From 9f007a0af25b4033834aa776166dead2aea2df5c Mon Sep 17 00:00:00 2001 From: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> Date: Sat, 17 Oct 2026 03:36:49 +0800 Subject: [PATCH] feat(site): add a playable research time machine Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> --- apps/presentation/site/README.md | 39 + .../site/public/lab/time-machine/data/ad.json | 2717 +++++++++++++ .../public/lab/time-machine/data/graph.json | 3486 +++++++++++++++++ .../lab/time-machine/data/portfolio.json | 1952 +++++++++ apps/presentation/site/scripts/seo-smoke.mjs | 18 +- .../site/scripts/time-machine.test.mjs | 64 + apps/presentation/site/src/App.tsx | 6 +- apps/presentation/site/src/main.tsx | 8 +- apps/presentation/site/src/page-metadata.ts | 11 + apps/presentation/site/src/render-static.tsx | 3 +- .../site/src/time-machine/TimeMachine.tsx | 177 + .../site/src/time-machine/records.ts | 44 + .../site/src/time-machine/scenes.json | 170 + .../site/src/time-machine/time-machine.css | 89 + 14 files changed, 8775 insertions(+), 9 deletions(-) create mode 100644 apps/presentation/site/public/lab/time-machine/data/ad.json create mode 100644 apps/presentation/site/public/lab/time-machine/data/graph.json create mode 100644 apps/presentation/site/public/lab/time-machine/data/portfolio.json create mode 100644 apps/presentation/site/scripts/time-machine.test.mjs create mode 100644 apps/presentation/site/src/time-machine/TimeMachine.tsx create mode 100644 apps/presentation/site/src/time-machine/records.ts create mode 100644 apps/presentation/site/src/time-machine/scenes.json create mode 100644 apps/presentation/site/src/time-machine/time-machine.css diff --git a/apps/presentation/site/README.md b/apps/presentation/site/README.md index 224d5fb678..b4e46f3166 100644 --- a/apps/presentation/site/README.md +++ b/apps/presentation/site/README.md @@ -109,6 +109,45 @@ not raw session logs. The full-screen viewer bundles only the two explicit live LoopX state, local status feeds, private registries, raw logs, or write APIs. +## Time Machine + +`/lab/time-machine/` is a read-only, decision-first public experience. Choose +for an Agent, see the actual recorded outcome, then optionally inspect the +original finding or browse each capture. The existing homepage research +collection discovers it. English is the default; `?lang=zh` selects Chinese, +and `?case=graph` or `?case=ad` opens another episode. Back/forward navigation +restores the episode. Choices live only in page memory; reloading resets them. +Both choices reveal the same historical outcome. They trigger no model, +installation, restore, live Goal operation or simulated counterfactual. + +The three scenes distinguish a Portfolio local/native reversal, a scoped Graph +finding with unproven native benefit, and an Ad screening/fresh-sample reversal. +These are solver-authored diagnostic observations, not independent replications, +countable benchmark results or evidence of a LoopX performance gain. The original +status and text remain visible, including uncertainty. The record slider reads +one original event in append order; it is not a historical-state projection. +There are no inferred dependencies or causal links. + +The 76/134/105 events are byte-identical public fixtures from +[the fixed research-history source](https://github.com/loopx-project/loopx/tree/44c7af3db271ab1991428d3ff1991ca922434eff/apps/presentation/dashboard/src/data/fixtures), +selected against the +[immutable public archive](https://huggingface.co/datasets/huangrt01/loopx-edgebench-research-checkpoints-20261015/tree/c858f51021dc1c1cf08263f9c04b692a5e29b40b). +The browser verifies the capture SHA-256 before enabling a choice. A missing, +changed or malformed capture clears the displayed records and offers retry; +old content is never relabeled as another case. No raw native session is bundled. +The existing Explore v0 owner defines record semantics; the local TS reader +only checks the pinned artifact shape and renders its text. No shared vocabulary, +capability or provider is introduced. The route loads separately from the home +page and the static build includes the English question and archive link. + +Run `node --test scripts/time-machine.test.mjs` for capture integrity, negative +cases and verbatim order checks, then `npm run build` and the ordinary SEO smoke. +The packaged browser journey must also cover choices/results/original text, +locale/case/history/reload, keyboard and narrow screens, plus a missing or +changed capture and successful retry. Build at both `/` and `/loopx/` to qualify +asset and navigation bases. Removing the route and its research entry reverts +the experience without touching Explore state or live Agents. + ## Search publication `npm run build` compiles the browser app and prerenders the homepage and the diff --git a/apps/presentation/site/public/lab/time-machine/data/ad.json b/apps/presentation/site/public/lab/time-machine/data/ad.json new file mode 100644 index 0000000000..fdda6585d4 --- /dev/null +++ b/apps/presentation/site/public/lab/time-machine/data/ad.json @@ -0,0 +1,2717 @@ +{ + "description": "Unchanged Explore history from a real, sealed research checkpoint later restored by follow-up experiments. Original text, timestamps, identities and references; no invented relationships.", + "source_sha256": "de04b679e9f382b650e891e5c7dd7146bc690500435c511f107c9d24da4f4efa", + "source_bytes": 145296, + "goal_id": "ad-warm-off-20261015-01", + "display_name": "广告优化 · 近期研究 checkpoint", + "provenance": { + "case_id": "ad_placement_optimization", + "source_run_id": "ad-warm-stride-off-20261015-01", + "checkpoint_id": "8656d5442a664ada9d8bdb4877349f43", + "checkpoint_sha256": "1d5c9691440d941424ab531b5632f4b5e2bb526fd6039e00421409f809bc41e3", + "captured_at": "2026-10-15T14:32:27.665890+00:00", + "captured_native_turns": 87, + "selection": "Fixed previously receiver-qualified second-generation natural Ad8656d5 point, selected before this launch from existing14:31Z qualification; same exact digest, no selection by score or later source feedback.", + "restored_by": [ + "ad-warm-stride-on-20261015-05", + "ad-warm-stride-off-20261015-05" + ], + "redactions": [], + "claim_scope": "diagnostic_only", + "score_countable": false, + "trajectory_publication": { + "status": "uploaded", + "repository": "https://huggingface.co/datasets/huangrt01/loopx-edgebench-research-checkpoints-20261015", + "revision": "c858f51021dc1c1cf08263f9c04b692a5e29b40b", + "archive": "ad_placement_optimization.zip", + "archive_sha256": "4a00c2e217ce3aac10fbbb3ec73b365342cfbdb11ee154311f2ff3df9b068fef" + }, + "boundary": "Captured Explore knowledge only; original native execution state, credentials, evaluator and models are not imported." + }, + "source_counts": { + "event_count": 105, + "event_kinds": { + "node": 48, + "finding": 57 + }, + "node_kinds": { + "question": 48 + }, + "statuses": { + "open": 48, + "tentative": 57 + }, + "first_recorded_at": "2026-10-14T18:55:41+00:00", + "last_recorded_at": "2026-10-15T14:23:55+00:00", + "edge_count": 0, + "parent_records": 0 + }, + "events": [ + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "71c462d361749b53", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T18:55:41+00:00", + "result_id": "question_ab81670234b2e784a2279332a8e487b8ec15ebead3b7dd8132552044d26c2f61", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:fecfe6289ac8d8f8", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Current regrowth solver on generated seeds 0,1,3,6,8,9 and 200–223; paired uniform and mild-bias variants under a 4.60s search budget.", + "title": "Does mildly biasing move selection toward less-satisfied ads improve score within the 4.60-second search budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "60fbbe8a76f28829", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-bias-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_ab81670234b2e784a2279332a8e487b8ec15ebead3b7dd8132552044d26c2f61", + "recorded_at": "2026-10-14T18:55:41+00:00", + "result_id": "finding_eaa58db7ef557a0b86d2823478d2a988223f47599d11e8743897da84c8dfcab7", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:d9aa789883c60741", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: f46b58e53ddfb6f035d5f5656ac1da43c867488487b47f5f467f8961c34f46d6 Applicability: Current regrowth solver on generated seeds 0,1,3,6,8,9 and 200–223; paired uniform and mild-bias variants under a 4.60s search budget. Observation: Native checkpoint b526bec8 passes 50/50 with score 49332642308. Mild bias gains 0.61M mean points on 24 fresh seeds, with mixed individual results; candidate f46b58e5 remains native-untested. Interpretation: Outcome: continue retained: - Source-bound native evaluation and best-state retention - Useful local work during the 3600-second cooldown changed: - Prefer the less-satisfied of two random ads on half the moves", + "tags": [ + "writeback-result" + ], + "title": "Does mildly biasing move selection toward less-satisfied ads improve score within the 4.60-second search budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4eccd4d8807300ac", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T19:17:59+00:00", + "result_id": "question_d745d82439ff49e61e889ffc15393a1bc86d19534eede5ecf30c5b4fc7c65332", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:92b2c7500cf92da8", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Current mild-bias solver; six tuning seeds, paired fresh seeds 300–315 and cut-repair replication 316–331; tested slide and one/four-blocker cut settings.", + "title": "Do blocker slides or alternative cut-and-regrow repairs improve satisfaction within the 4.60-second search budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "dede0c014e825ded", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-repair-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_d745d82439ff49e61e889ffc15393a1bc86d19534eede5ecf30c5b4fc7c65332", + "recorded_at": "2026-10-14T19:17:59+00:00", + "result_id": "finding_725bb4ef39d028c8e5fc20796b7da3cf261f1372fe49c313da60754f61e5c88d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:70370fe8823fd62d", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: f62ce9936d6b96123865247556f1160de8ef02e7d8dce4208742419e18004d26 Applicability: Current mild-bias solver; six tuning seeds, paired fresh seeds 300–315 and cut-repair replication 316–331; tested slide and one/four-blocker cut settings. Observation: Slides lost 474365.5 mean points on 16 fresh seeds. Alternative cuts gained 242797.1 then lost 282917.1, pooled -20060 on 32. All 104 outputs passed within 5 seconds; current f46b58e5 remains native-untested. Interpretation: Outcome: continue retained: - Current mild-bias candidate and separate native winner - Source-bound validation and submission cooldown changed: - Added paired evidence on fresh seeds 300–331 stopped: - Adoption of unconditional blocker slides - Adoption of alternative cuts at current settings", + "tags": [ + "writeback-result" + ], + "title": "Do blocker slides or alternative cut-and-regrow repairs improve satisfaction within the 4.60-second search budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9be581670fddeaee", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T19:42:53+00:00", + "result_id": "question_8a277db0ac93987eb58000636bf4ed6fbbe4d67266a51346d89d480841b8f044", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:a06c4de6c52233b0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Mild-bias solver; endpoint tuning on six seeds, paired fresh seeds 400–431, and native combined-source comparison with b526bec8.", + "title": "Does gentler cooling improve the regrowth solver within the 4.60-second search budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "27ddad958fdb878a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-floor-1", + "evidence:ad-native-2" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_8a277db0ac93987eb58000636bf4ed6fbbe4d67266a51346d89d480841b8f044", + "recorded_at": "2026-10-14T19:42:53+00:00", + "result_id": "finding_1fb166d0b50e9f530c80428b5a2bc1c1f07f117962204bf955281e3f4b962a48", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:fa53ab971ceb6f12", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: c77bbd69d02ef96cf9b05ba5cf4e2127a803608307e7cce82fc7440e0863bc65 Applicability: Mild-bias solver; endpoint tuning on six seeds, paired fresh seeds 400–431, and native combined-source comparison with b526bec8. Observation: Ending at 1e-4 gains 1.02M mean points on 32 fresh pairs, winning 26. Combined candidate c77bbd69 improves native total by 64880888 and passes 50/50. Free growth regresses; final polish adds only 74 points on saved layouts. Interpretation: Outcome: continue retained: - Feasible best-state retention and the 4.60-second search budget - Source-bound native evaluation and acceptance-based cooldown changed: - Final annealing temperature from 1e-7 to 1e-4 - Best native checkpoint to submission 3a5c8079e85b stopped: - Adoption of tested free-growth and hotter-endpoint settings", + "tags": [ + "writeback-result" + ], + "title": "Does gentler cooling improve the regrowth solver within the 4.60-second search budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "92032c72afb9b718", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T19:59:21+00:00", + "result_id": "question_9ab9d162ee69f3aa3ac48db7aaa2f8971b1463d86dc6b0eacc7eec64dd0eadfc", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:110a8496071a4f49", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Initial-temperature tuning on six seeds; 0.008 versus 0.032 paired on fresh seeds 500–531, with 1e-4 ending and native-tested c77bbd69 comparator.", + "title": "Does a warmer starting temperature improve the gently cooled regrowth solver within its 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2d7ec2378a491914", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-initial-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_9ab9d162ee69f3aa3ac48db7aaa2f8971b1463d86dc6b0eacc7eec64dd0eadfc", + "recorded_at": "2026-10-14T19:59:21+00:00", + "result_id": "finding_354ebc597b7824d9a5a6821590ad3cc5bac611e94b35d9efe8c79cd79e418f90", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:dd2511b3128163df", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 07d9f3cbf095f92c0ee3be2d47029653836753d85731af34cf92d56a24bee8ce Applicability: Initial-temperature tuning on six seeds; 0.008 versus 0.032 paired on fresh seeds 500–531, with 1e-4 ending and native-tested c77bbd69 comparator. Observation: Starting at 0.032 gains 1.96M and 2.98M mean points on two fresh batches, pooled 2.47M with 25/32 wins. Source 07d9f3cb is native-untested; c77bbd69 remains the separately preserved native winner. Interpretation: Outcome: continue retained: - The 1e-4 endpoint, feasible best-state retention and 4.60-second budget - Source-bound native evaluation and submission cooldown changed: - Starting temperature from 0.008 to 0.032 - Promising candidate to 07d9f3cb; native-best checkpoint remains c77bbd69", + "tags": [ + "writeback-result" + ], + "title": "Does a warmer starting temperature improve the gently cooled regrowth solver within its 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e8dec7c419fc6049", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T20:21:00+00:00", + "result_id": "question_9013d40fc8889b9328b94bc6a9f77c5274738b8702c35986bbee830179078905", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:f1fbcdaf2000bea1", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 07d9f3cb; initial 0.032/0.064/0.128 screened on six seeds, 0.064 versus 0.032 on fresh 600–631; endpoint 1e-4 and 4.60-second budget.", + "title": "Do adjacent hotter starts improve the current 0.032 annealing candidate?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "a570411e49a0fcea", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-hotter-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_9013d40fc8889b9328b94bc6a9f77c5274738b8702c35986bbee830179078905", + "recorded_at": "2026-10-14T20:21:00+00:00", + "result_id": "finding_2c33e96ed05a5b923134d61e6b593aed4f632816798e59b48b7ad795327ab69e", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:bbe6bad6763e2dd3", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 07d9f3cbf095f92c0ee3be2d47029653836753d85731af34cf92d56a24bee8ce Applicability: Source 07d9f3cb; initial 0.032/0.064/0.128 screened on six seeds, 0.064 versus 0.032 on fresh 600–631; endpoint 1e-4 and 4.60-second budget. Observation: 0.064 gained 1.12M on six screening seeds, then lost 0.29M and 1.17M in fresh batches, pooled -0.73M with 13/32 wins. 0.128 lost 2.12M in screening. Tests remained valid under 5s. Interpretation: Outcome: continue retained: - Candidate 07d9f3cb; native winner c77bbd69 separately - Endpoint 1e-4, 4.60-second budget and source-bound evaluation changed: - Added paired evidence on fresh seeds 600–631 stopped: - Adoption of tested 0.064/0.128 starts under the unchanged schedule", + "tags": [ + "writeback-result" + ], + "title": "Do adjacent hotter starts improve the current 0.032 annealing candidate?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8c001de31d6d67f8", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T20:43:03+00:00", + "result_id": "question_d3c06997e7913b35fd8381385a6a9bb58c751dc7263fd357c94baa51655eedd3", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:c2a37d7d8d640b33", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 07d9f3cb comparator, exponents 0.75/1/1.25; six screening seeds and 32 fresh pairs per changed exponent, adopted 87f3e340 and combined native comparison with c77bbd69.", + "title": "Does changing cooling progression improve the warm-start regrowth solver within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "eeff5b0a1364db35", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-curve-1", + "evidence:ad-native-3" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_d3c06997e7913b35fd8381385a6a9bb58c751dc7263fd357c94baa51655eedd3", + "recorded_at": "2026-10-14T20:43:03+00:00", + "result_id": "finding_df77392388ce4a0749133104be6d07ff908b128b9a75aeeff089ca9510c08185", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:121020900627cb71", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 87f3e340fceb1b1cd0e3ae8ff2756c67061ce0971de9099c6894856f55ddae05 Applicability: Source 07d9f3cb comparator, exponents 0.75/1/1.25; six screening seeds and 32 fresh pairs per changed exponent, adopted 87f3e340 and combined native comparison with c77bbd69. Observation: Earlier cooling gained 0.140M and 0.168M in fresh batches, pooled 0.154M with 13/32 wins; delayed cooling lost in both. The combined warmer-start/earlier-cooling source gains 100747013 native total points, 50/50 passed; individual native effects remain unknown. Interpretation: Outcome: continue retained: - Feasible best-state retention, endpoint 1e-4 and 4.60-second budget - Source-bound checkpoints and acceptance-based submission cooldown changed: - Cooling progression to p^0.75 - Native-best source to 87f3e340/submission 54b9cd2bb608 stopped: - Adoption of the tested delayed-cooling schedule", + "tags": [ + "writeback-result" + ], + "title": "Does changing cooling progression improve the warm-start regrowth solver within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7df922e95c8e0e72", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T21:06:40+00:00", + "result_id": "question_cd5526f82a411b9ed41b2ba4238107e32c941441b4b8ef443d617a8013a5ae88", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:7af9bc99a6197fc6", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 87f3e340; global modes screened on 800–805, mode 2 paired on 816–847; conditional mode 3 screened on 850–855 and paired on 856–887, same cooling and 4.60s budget.", + "title": "Do global or satisfaction-conditional late-small-step policies improve the native-tested annealing solver?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8b1d4682ac9fdfcc", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-steps-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_cd5526f82a411b9ed41b2ba4238107e32c941441b4b8ef443d617a8013a5ae88", + "recorded_at": "2026-10-14T21:06:40+00:00", + "result_id": "finding_a6220604a8e2af15490412da21ccf57539df8d0a8b9c77347f5e18b1be2798ef", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:1ab28a78c1a9b5bf", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 94268ddce13ed493435de99ec427f97841f27c1f8f8f63d7f76bb492a95c05a0 Applicability: Source 87f3e340; global modes screened on 800–805, mode 2 paired on 816–847; conditional mode 3 screened on 850–855 and paired on 856–887, same cooling and 4.60s budget. Observation: Global reweighting lost 0.024M and 0.063M in fresh batches, pooled -0.043M. Conditional reweighting lost 0.358M then gained 1.389M, pooled +0.516M with 20/32 wins. Its batch signs conflict; neither policy is adopted. Interpretation: Outcome: continue retained: - Native winner 87f3e340 and its cooling/budget - Conditional alternative 5dc41f3e without native-score attribution changed: - Added source-bound step-policy evidence on fresh seeds 800–887 - Next structural probe: regrowth following an area-neutral primary move stopped: - Adoption of the tested global late-small-step policies", + "tags": [ + "writeback-result" + ], + "title": "Do global or satisfaction-conditional late-small-step policies improve the native-tested annealing solver?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "10e9b3ca1c412d37", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T21:23:41+00:00", + "result_id": "question_9b0148646d3d45f542978ef0e09dead30ea296f69a2c4ca8e384da76b76c029d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:70d581ed7dc8407a", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native source 87f3e340; always or quarter-sampled neutral recovery for <=2 blockers, six screening seeds 900–905 and sampled recovery paired on 916–947, same cooling/proposals.", + "title": "Does neighbor regrowth after an area-neutral primary move improve packing under the 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3e7f889806ac49b1", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-neutral-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_9b0148646d3d45f542978ef0e09dead30ea296f69a2c4ca8e384da76b76c029d", + "recorded_at": "2026-10-14T21:23:41+00:00", + "result_id": "finding_1444e864dd58af87f21a263551730b981e8a0076243606b70e34adac4c6f2bcb", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:036031897a1df3ba", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 162d0dc44e286f2ed0be32237e0e11c0d87e78360e1b5e2ef7061b9f75827880 Applicability: Native source 87f3e340; always or quarter-sampled neutral recovery for <=2 blockers, six screening seeds 900–905 and sampled recovery paired on 916–947, same cooling/proposals. Observation: Quarter-sampled recovery gained 74223 points in six-case screening, then lost 608598 and 210423 in fresh batches, pooled -409510 with 16/32 wins. All 82 outputs passed; benefit under the tested conditions is unsupported. Interpretation: Outcome: replan retained: - Native winner 87f3e340, cooling progression and 4.60-second budget - Unadopted conditional-step checkpoint 5dc41f3e changed: - Next probe targets initial topology through weighted point partitions stopped: - Adoption of tested neutral-primary recovery settings", + "tags": [ + "writeback-result" + ], + "title": "Does neighbor regrowth after an area-neutral primary move improve packing under the 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e1e19d0ff9b45de2", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-4", + "evidence:ad-steps-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_cd5526f82a411b9ed41b2ba4238107e32c941441b4b8ef443d617a8013a5ae88", + "recorded_at": "2026-10-14T21:45:26Z", + "result_id": "finding_a6220604a8e2af15490412da21ccf57539df8d0a8b9c77347f5e18b1be2798ef", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:870d5589c0722884", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Source 5dc41f3e, AD_STEP_MODE=3, gains 27505940 native total points over 87f3e340 in accepted submission a83a3f1f24f2, all 50 passed. Adopted exact checkpoint. Local evidence remains mixed: prior fresh batches -357534 and +1388894.5; new 1100-1107 batch -555327.5; pooled 40 fresh pairs +301478.7, 23 wins/17 losses. Native details expose PASS statuses only. Global policies and other step settings remain separately scoped; generalization and per-case native gains are unknown.", + "title": "Conditional late-small steps gain 27505940 native total points, 50/50 passed" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "886dbae59eb18cd3", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T21:46:53+00:00", + "result_id": "question_db5ee516fa3119d2cca299ef34405ad741e28b0efc5ef7db74e7e9edbc781dfd", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:ad53317c82f62b54", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 4c245312, AD_INIT_MODE 0/1/2; screen 1000–1005, mode 1 fresh 1016–1047, cooler mode-1 starts screen 1200–1205. All search budgets 4.60s.", + "title": "Does weighted partition initialization improve final ad-placement score within the same time budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9a0cabca18fc1835", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-partition-1", + "evidence:ad-partition-cooling-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_db5ee516fa3119d2cca299ef34405ad741e28b0efc5ef7db74e7e9edbc781dfd", + "recorded_at": "2026-10-14T21:46:53+00:00", + "result_id": "finding_ccc76aed72d3ac3455ac241adcf690d969c6ecdeff300af6c44d59d86d519bc9", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:b157a7edac2db375", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 4c2453125a9b990e6872f475a26dec26bee3a2cc27b2d9c9694ff28d3d8423c4 Applicability: Source 4c245312, AD_INIT_MODE 0/1/2; screen 1000–1005, mode 1 fresh 1016–1047, cooler mode-1 starts screen 1200–1205. All search budgets 4.60s. Observation: Partition-only initial mean rose from 0.9200 to 0.9523 in screening, but fresh final batches gained 666255 and lost 860646, pooled -97195, 13/32 wins. Starts 0.016/0.008 lost 1.360M/2.090M versus warm partition on six new cases. All 124 partition-stage outputs passed. Interpretation: Do not adopt these variants. Hybrid gained 459115 in six-case screening but has no fresh replication; test it on current native winner 5dc41f3e. Better initialization does not establish final benefit, and the small cooler screen does not prove general inferiority.", + "tags": [ + "writeback-result" + ], + "title": "Does weighted partition initialization improve final ad-placement score within the same time budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "03abd6526fef423a", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:02:54+00:00", + "result_id": "question_7d9f65be0c83cc9fb71a1bfe72a81a31dc620ec0de49375409ea3ff8ac307c19", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:b06235f166a7216f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 5dc41f3e comparator and 72b03f01 hybrid; screening 1300–1305, fresh 1316–1347, same cooling/steps and 4.60s budget.", + "title": "Does hybrid greedy/weighted-partition initialization improve the native-winning conditional-step solver?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "16531e6b4fc0bd1a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-hybrid-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_7d9f65be0c83cc9fb71a1bfe72a81a31dc620ec0de49375409ea3ff8ac307c19", + "recorded_at": "2026-10-14T22:02:54+00:00", + "result_id": "finding_05dab6dd051fe64fb8e9c83316b11ac49d2e274950cfa83e39301b922799193d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:f83fda4352561dbd", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 72b03f016d6c378ef03a5a419b50f681cf8eae70da1dda1e06602779937be8e7 Applicability: Source 5dc41f3e comparator and 72b03f01 hybrid; screening 1300–1305, fresh 1316–1347, same cooling/steps and 4.60s budget. Observation: Hybrid gained 749595 and 332492 mean points in fresh batches, pooled +541044 with 22/32 wins. All 92 outputs passed. The 4.747s case reran at 4.614/4.616s with varying scores. Source 72b03f01 is the promising local candidate; its native effect is unknown. Interpretation: Outcome: continue retained: - Native-tested source 5dc41f3e and acceptance-based submission cooldown - Conditional steps, cooling schedule and 4.60-second total budget changed: - Hybrid initialization in solution.cpp and best-supported checkpoint - Next local parameter probe: conditional step satisfaction threshold stopped: - Attributing existing native feedback to the unsubmitted hybrid candidate", + "tags": [ + "writeback-result" + ], + "title": "Does hybrid greedy/weighted-partition initialization improve the native-winning conditional-step solver?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "a8062039e3095642", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:23:13+00:00", + "result_id": "question_6b12c07a469887a2d96291bf760bb754328154b8c155a7596cf7838dd35a01e2", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:f4541b0b76f1628e", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Hybrid 72b03f01, thresholds 0.95/0.98/0.995 screened on 1400–1405; 0.95 fresh paired on 1416–1447. Diagnostics verify changed paths; native transfer untested.", + "title": "Does changing the late-small-step satisfaction threshold improve hybrid annealing within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "aeef678a7978aaf5", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-threshold-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_6b12c07a469887a2d96291bf760bb754328154b8c155a7596cf7838dd35a01e2", + "recorded_at": "2026-10-14T22:23:13+00:00", + "result_id": "finding_ce5294c7d7f64e514f63767ba465a68f6ba7b66159ed843786ce1661478b29c3", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:e6640f3cd528e986", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 6fa273a2ac11bb5ca81178ec4dcc7e78b9cb2eab2b697bad808fc10632fe530f Applicability: Hybrid 72b03f01, thresholds 0.95/0.98/0.995 screened on 1400–1405; 0.95 fresh paired on 1416–1447. Diagnostics verify changed paths; native transfer untested. Observation: Threshold 0.95 lost 165687 in six-case screening, then gained 547434 and 770272 in fresh batches, pooled +658853 with 20/32 wins. All 84 outputs passed; both changed predicates executed. Its native effect remains unknown. Interpretation: Outcome: continue retained: - Hybrid initialization, cooling and 4.60-second budget - Separate native winner 5dc41f3e and source-bound checkpoints changed: - Late-small-step satisfaction threshold from 0.98 to 0.95 - Next probe targets rectangle-relative amplitude, rather than another threshold sweep stopped: - Adoption of the screened-only 0.995 variant", + "tags": [ + "writeback-result" + ], + "title": "Does changing the late-small-step satisfaction threshold improve hybrid annealing within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "a522326de2208481", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:37:53+00:00", + "result_id": "question_811011eb5010826448b388be29318ff0c6dedc8954c6c1fc24903c7c7c9aab81", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:69f3538c9024ab74", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 6fa273a2 vs modes 0/1/2 in f4c6f5e6; side/4 and side/8 screen 1500–1505, side/8 fresh 1516–1547. Native transfer untested.", + "title": "Do rectangle-relative late-step caps improve the threshold-0.95 hybrid within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fdff2b28eb40b773", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-relative-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_811011eb5010826448b388be29318ff0c6dedc8954c6c1fc24903c7c7c9aab81", + "recorded_at": "2026-10-14T22:37:53+00:00", + "result_id": "finding_b998ffa281ceb5c3ed0e10fd6eb1045c5892f4f0c76e6db94d053a8657797625", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:38643b13c0bb650f", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: f4c6f5e604cb554a904f1c20cb6daab0a6007b7771b0538a6d37ac91a00fa218 Applicability: Source 6fa273a2 vs modes 0/1/2 in f4c6f5e6; side/4 and side/8 screen 1500–1505, side/8 fresh 1516–1547. Native transfer untested. Observation: Side/8 cap screened +35546, then lost 449966 and 514496 in fresh batches, pooled -482231 with 13/32 wins. Side/4 screened -838579. All 84 outputs passed; diagnostics verified both changed paths. No cap is adopted. Interpretation: Outcome: continue retained: - Uncapped threshold-0.95 hybrid source 6fa273a2 and 4.60-second budget - Separate native winner 5dc41f3e and acceptance-based cooldown changed: - Next priority is native transfer of the locally supported hybrid/threshold candidate stopped: - Adoption of the tested side/4 and side/8 late-step caps", + "tags": [ + "writeback-result" + ], + "title": "Do rectangle-relative late-step caps improve the threshold-0.95 hybrid within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "14b1f32b227fd6a4", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:47:10+00:00", + "result_id": "question_b1c8a9d038aa09fc89f89d7ea9bb1f87af7097b6e528e68e7300af8b28c47e6d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:99044d79f7aa46e7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native source 6fa273a2/9ae3a0c9e323 versus 5dc41f3e/a83a3f1f24f2; combined initialization and threshold changes under 4.60 seconds.", + "title": "Do combined hybrid initialization and threshold 0.95 improve native score over the prior conditional-step winner?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8f76b34d8d0cf24a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-5" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_b1c8a9d038aa09fc89f89d7ea9bb1f87af7097b6e528e68e7300af8b28c47e6d", + "recorded_at": "2026-10-14T22:47:10+00:00", + "result_id": "finding_a5d457f710199a53121517ac5b0a30f80328beef27e9968a4a3d431bf9001237", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:6769dd1a3f1af277", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 6fa273a2ac11bb5ca81178ec4dcc7e78b9cb2eab2b697bad808fc10632fe530f Applicability: Native source 6fa273a2/9ae3a0c9e323 versus 5dc41f3e/a83a3f1f24f2; combined initialization and threshold changes under 4.60 seconds. Observation: Native score 49527208218, 50/50 passed, +1432069 over prior best. Source was compared with its accepted checkpoint before promotion. Public per-test details expose passed statuses only. Interpretation: Keep accepted source 6fa273a2 as native best. This small single-evaluation gain supports score_first selection but does not isolate the two changes or quantify runtime-dependent variability. Preserve former winner and continue fresh local experiments.", + "tags": [ + "writeback-result" + ], + "title": "Do combined hybrid initialization and threshold 0.95 improve native score over the prior conditional-step winner?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5ce24859c0a39159", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:56:40+00:00", + "result_id": "question_9517e66c3cac79b48d22611214c56847b7d16d57a579fbbd7d8b59b2aef96fcf", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:70dcc9b6a52ee7a7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 6fa273a2 vs be900c53 mode 2, floor 0.001 after 60% progress for satisfaction below 0.95; screen 1600–1605, fresh 1616–1647. Native transfer untested.", + "title": "Does a late temperature floor limited to poorly satisfied ads improve the native-winning hybrid under 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1f77ad39d40b356a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-poor-temperature-1", + "evidence:ad-poor-temperature-screen-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_9517e66c3cac79b48d22611214c56847b7d16d57a579fbbd7d8b59b2aef96fcf", + "recorded_at": "2026-10-14T22:56:40+00:00", + "result_id": "finding_782c7afbf8df95740b4bd7c669b7400ddfbbf8bd8bcf7b70130d44c2b6113b2e", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:15eb407db56be600", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 84c7e5b75c14f665f6009d226eb0c267d6205863de889b0baf051c633f5bf4d8 Applicability: Source 6fa273a2 vs be900c53 mode 2, floor 0.001 after 60% progress for satisfaction below 0.95; screen 1600–1605, fresh 1616–1647. Native transfer untested. Observation: Conditional floor 0.001 gained 82379 and 649154 in fresh batches, pooled +365767, 17/32 wins; paired SD 1.90M. All 64 new outputs passed. Both positive means support provisional local adoption; native transfer remains unknown. Interpretation: Outcome: continue retained: - Hybrid initialization, geometric invariants and 4.60-second budget - Native winner 6fa273a2 and acceptance-based cooldown changed: - Local candidate to source 84c7e5b7 with conditional late floor 0.001 - Next probe: floor qualification 0.90/0.98 versus 0.95 stopped: - Adoption of the screened-only conditional floor 0.0003", + "tags": [ + "writeback-result" + ], + "title": "Does a late temperature floor limited to poorly satisfied ads improve the native-winning hybrid under 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "a1aee13065a0390e", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:10:03+00:00", + "result_id": "question_3c76440eacd4ae2f338e427e06a2e6ea84a2ed5b60531fb2242fd5387687bacb", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:8eaa80cfccc17a0b", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 84c7e5b7 vs 793cf97a thresholds 0.90/0.95/0.98; screen 1700–1705, 0.90 fresh 1716–1747; floor 0.001 after 60%, 4.60-second budget. Native effect untested.", + "title": "Does changing the qualification threshold for the conditional late temperature floor improve ad-placement score?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b59f19a6c3638c96", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-floor-threshold-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_3c76440eacd4ae2f338e427e06a2e6ea84a2ed5b60531fb2242fd5387687bacb", + "recorded_at": "2026-10-14T23:10:03+00:00", + "result_id": "finding_488d172738506532c3bee8458b04f0a72b51837f0d3a26233329c980861446a5", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:70d447366999601f", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 69fa3f047c7495f89adf1f482fb1132e2047c467ee69a54cbeed37a4460c3fd5 Applicability: Source 84c7e5b7 vs 793cf97a thresholds 0.90/0.95/0.98; screen 1700–1705, 0.90 fresh 1716–1747; floor 0.001 after 60%, 4.60-second budget. Native effect untested. Observation: Threshold 0.90 gained 363686 and 224516 in fresh batches, pooled +294101, 17/32 wins, SD 1.69M. All 86 outputs passed. A 0.98 run took 4.853s, repeated at 4.619s; cause unresolved. Fresh max 4.627s. Native effect unknown. Interpretation: Outcome: continue retained: - Conditional floor 0.001 after 60%, step qualification 0.95 and 4.60-second budget - Source-bound native winner 6fa273a2 changed: - Local candidate floor qualification from 0.95 to 0.90, source 69fa3f04 - Next probe: proposal mix emphasizing shape changes or translations stopped: - Adopting the screened-only 0.98 heat qualification", + "tags": [ + "writeback-result" + ], + "title": "Does changing the qualification threshold for the conditional late temperature floor improve ad-placement score?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "60ca4fccaa57486b", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:22:04+00:00", + "result_id": "question_9ee7d67ab9d2a39f8e7fdb7f47e5d3f6ea706a346301011b353bc02417a7bb94", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:3423f906da44e6aa", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 69fa3f04 vs 06b784bd modes 0/1/2; screening 1800–1805, more-shape fresh 1816–1847. Same heat, initialization and 4.60-second budget. Native transfer untested.", + "title": "Does shifting proposal probability from expansion to shape changes or translations improve ad-placement score?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ca517941e2db701a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-move-mix-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_9ee7d67ab9d2a39f8e7fdb7f47e5d3f6ea706a346301011b353bc02417a7bb94", + "recorded_at": "2026-10-14T23:22:04+00:00", + "result_id": "finding_741b895b6b01d120b4029b078872801be05282f05e5dde0c3bf23a432926037a", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:f4a7133ff606130a", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 1fe4e7f2418f44a66eb83bbaa9c78cb9973c0219fd517e04f639b9c3370cc8b6 Applicability: Source 69fa3f04 vs 06b784bd modes 0/1/2; screening 1800–1805, more-shape fresh 1816–1847. Same heat, initialization and 4.60-second budget. Native transfer untested. Observation: Shape mix gained 274281 and 870507 in fresh batches, pooled +572394, 14/32 wins; paired SD 2.60M. Extra translations screened -624303. All 84 outputs passed. Mean gains meet the local rule despite more losses than wins; native transfer unknown. Interpretation: Outcome: continue retained: - Hybrid, conditional heat and 4.60-second budget - Separate native winner 6fa273a2 and accepted-result cooldown changed: - Local proposal mix to 26/11/19/30/11/3, source 1fe4e7f2 - Next step: evidence/cost route review before further tuning stopped: - Adoption of the screened-only extra-translation mix", + "tags": [ + "writeback-result" + ], + "title": "Does shifting proposal probability from expansion to shape changes or translations improve ad-placement score?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ee0ccd9cc2461508", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:46:49+00:00", + "result_id": "question_7d465c70037e8d26766168f64370dc3942339fc32f2484174936df298eba185d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:319871c40d1275d1", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 1fe4e7f2 versus 78f28710 modes 0/1/2; screen 1900–1905, high-loss-first fresh 1916–1947. Native transfer untested.", + "title": "Does prioritizing contracted blockers by lost satisfaction improve compound recovery within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9480394a66b19290", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-regrow-order-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_7d465c70037e8d26766168f64370dc3942339fc32f2484174936df298eba185d", + "recorded_at": "2026-10-14T23:46:49+00:00", + "result_id": "finding_c652328314cf9ac2757f85dfcbdb9907bb719cf1a1e36185926640694663a9f2", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:4631e4b73c1ab3c0", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 78f287102965442f2b18f9cb6dcbb678a2a480345ab79b674c075d25a1379225 Applicability: Source 1fe4e7f2 versus 78f28710 modes 0/1/2; screen 1900–1905, high-loss-first fresh 1916–1947. Native transfer untested. Observation: High-loss-first gained in screening but lost pooled fresh score. Heat/shape candidate gained 954331 native points; its effects are combined. Interpretation: Outcome: continue retained: - Source 1fe4e7f2 as current/native best; geometry and 4.60-second budget - Input-order compound recovery and source-bound accepted feedback changed: - Next probe: target-area shape goals for poor ads, rather than recovery ordering stopped: - Adoption of tested high-loss-first ordering; low-loss-first remains screened only", + "tags": [ + "writeback-result" + ], + "title": "Does prioritizing contracted blockers by lost satisfaction improve compound recovery within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2941f61b7ab77172", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:57:13+00:00", + "result_id": "question_c1493e636c25a00dac91855c3bd63efc4c99e73ef715d75a7e7131fd4da50b1c", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:341dd80674551954", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source 1fe4e7f2 versus cd62c408 modes 0/1/2: 25/50/75% below 0.95; screen 2000–2005, 50% fresh 2016–2047. Native transfer untested.", + "title": "Does increasing target-area shape probability only for poorly satisfied ads improve score within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "492416d19bdcb422", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-shape-goal-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_c1493e636c25a00dac91855c3bd63efc4c99e73ef715d75a7e7131fd4da50b1c", + "recorded_at": "2026-10-14T23:57:13+00:00", + "result_id": "finding_94c2c7d8ca560bdfdef01e0295157a0445dab5ba5f1a2bae47fc5e16da44d552", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:383a21b45a034221", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: cd62c408ed2642e6c4b0b70f54b9e51165205280f9243d31bb2596343841fa67 Applicability: Source 1fe4e7f2 versus cd62c408 modes 0/1/2: 25/50/75% below 0.95; screen 2000–2005, 50% fresh 2016–2047. Native transfer untested. Observation: 50% lost both fresh means, pooled -512583. One short diagnostic found 67–72% of poor target proposals contain another required point. Interpretation: Outcome: continue retained: - Native/current 1fe4e7f2, original 25% target probability, compound scoring and 4.60-second budget changed: - Next probe: point-aware orthogonal alignment, original-edge clamp versus point centering stopped: - Adoption of tested 50% target frequency; 75% is screened only", + "tags": [ + "writeback-result" + ], + "title": "Does increasing target-area shape probability only for poorly satisfied ads improve score within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3eca4a69d9fe1c7f", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T00:10:24+00:00", + "result_id": "question_bfaf907dad21fb0a2ef6b9037fb4ac48a276a63bf42d197ef7916a170069567f", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:4c4b9557019e92f7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native 1fe4e7f2 versus 1b5e6a66 modes 0/1/2; screen 2100–2105, centering fresh 2116–2147; promoted a28d2fe0. Native transfer untested.", + "title": "Does point-aware alignment of poor-ad target shapes improve score within the same 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d90b51db1e9e270b", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-point-align-1", + "evidence:ad-point-bound-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_bfaf907dad21fb0a2ef6b9037fb4ac48a276a63bf42d197ef7916a170069567f", + "recorded_at": "2026-10-15T00:10:24+00:00", + "result_id": "finding_acbecd211715cacb9843a732900ef4feeb9d6ddfca695d4bfb92f73805c6f731", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:aadf9da9d3014474", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: a28d2fe0e0ccc1995e690b8b37694d1fb51796444306b63a6a6940b68e8e7060 Applicability: Native 1fe4e7f2 versus 1b5e6a66 modes 0/1/2; screen 2100–2105, centering fresh 2116–2147; promoted a28d2fe0. Native transfer untested. Observation: Centering gained in both fresh batches. Oracle-checked point bounds force 1.28M local loss/case, about 17% of residual loss; joint packing can force more. Interpretation: Outcome: continue retained: - Geometry, original 25% target frequency, heat and 4.60-second budget - Separate native winner 1fe4e7f2 and accepted-result cooldown changed: - Local candidate to centered alignment, source a28d2fe0 - Next probe: bias selection by gap or relative satisfaction to point-feasible ceiling stopped: - Adoption of screened-only old-edge clamp", + "tags": [ + "writeback-result" + ], + "title": "Does point-aware alignment of poor-ad target shapes improve score within the same 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1f3a852297c6d286", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T00:24:52+00:00", + "result_id": "question_fcf244e904ed1101979079ae1047fed93aed941db3e01b74b534fd8c7659d29a", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:3aed994b70db859c", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Local a28d2fe0 versus e602d140 modes 0/1/2; screen 2200–2205, relative policy fresh 2216–2247. Native transfer untested.", + "title": "Does move selection using point-feasible satisfaction ceilings improve score within 4.60 seconds including preprocessing?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1765f37e22248f73", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-ceiling-bias-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_fcf244e904ed1101979079ae1047fed93aed941db3e01b74b534fd8c7659d29a", + "recorded_at": "2026-10-15T00:24:52+00:00", + "result_id": "finding_bc15ab1fd37bd50fa817e395f433961bed52215616d420b74aad573e8562984b", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:33c189cb96dcbffe", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: e602d1406e425d4c233b85c9b054cb676b1c65c01a5d9908d51d60c26ff50e35 Applicability: Local a28d2fe0 versus e602d140 modes 0/1/2; screen 2200–2205, relative policy fresh 2216–2247. Native transfer untested. Observation: Relative bias gained in screening but lost pooled fresh score. Bounded cases lost 564377 on average; post hoc groups do not isolate causality. Interpretation: Outcome: continue retained: - Local a28d2fe0, native 1fe4e7f2, raw selection and 4.60-second budget changed: - Next probe: mandatory-point clipping of translation proposals stopped: - Adoption of tested relative-ceiling selection; gap policy screened only", + "tags": [ + "writeback-result" + ], + "title": "Does move selection using point-feasible satisfaction ceilings improve score within 4.60 seconds including preprocessing?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "65fc3ba69962e72b", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T00:36:45+00:00", + "result_id": "question_ab6e1ffda47c3f35ac605ad01dcad90b93606ab46251f6619c0b40a90acffcd5", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:a031abef30980e7b", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Local a28d2fe0 versus 36bf6a8e modes 0/1/2; screen 2300–2305, poor-ad policy fresh 2316–2347. Native transfer untested.", + "title": "Does mandatory-point clipping of size-preserving translations improve score within the same 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ec3978aaaa873678", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-point-translate-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_ab6e1ffda47c3f35ac605ad01dcad90b93606ab46251f6619c0b40a90acffcd5", + "recorded_at": "2026-10-15T00:36:45+00:00", + "result_id": "finding_af9fca9733b50f24e9f9bae38eed3a8243c59c31a6193efc726defbbc4917017", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:985846b0f728db4f", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 36bf6a8e8fa35c419ea2b0ddca5c616b749d4d6ed459f2b5914a99e9996d18b1 Applicability: Local a28d2fe0 versus 36bf6a8e modes 0/1/2; screen 2300–2305, poor-ad policy fresh 2316–2347. Native transfer untested. Observation: Clipping removed diagnosed point contacts but failed fresh score adoption. Two short cases show frequent point contacts in poor-ad expansions; their benefit remains unproven. Interpretation: Outcome: continue retained: - Local a28d2fe0, native 1fe4e7f2, original translations and 4.60-second budget changed: - Next probe: clip ordinary expansion proposals at mandatory points stopped: - Adoption of poor-ad translation clipping; all-ad clipping screened only", + "tags": [ + "writeback-result" + ], + "title": "Does mandatory-point clipping of size-preserving translations improve score within the same 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "78d35ef4bc35d6a3", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-7", + "evidence:ad-point-align-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_bfaf907dad21fb0a2ef6b9037fb4ac48a276a63bf42d197ef7916a170069567f", + "recorded_at": "2026-10-15T00:51:46+00:00", + "result_id": "finding_72dfad365e222f610ec1f0342629e8de866c6b518d94d4c2cf490701b0221b5c", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:3eabffbaac4db33e", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: a28d2fe0e0ccc1995e690b8b37694d1fb51796444306b63a6a6940b68e8e7060 Applicability: Native 1fe4e7f2 versus 1b5e6a66 modes 0/1/2; screen 2100–2105, centering fresh 2116–2147; promoted a28d2fe0. Native transfer untested. Observation: Native a28d2fe0 lost 4152903, so restored 1fe4e7f2. Expansion clipping on a28 lost both fresh means; public native details do not diagnose geometry. Interpretation: Outcome: continue retained: - Native winner/source 1fe4e7f2, raw scoring, geometry checks and 4.60-second budget - Source-bound accepted feedback and separate checkpoints changed: - Root restored after byte comparison; next review tests repair-only shape fallback stopped: - Native adoption of a28d2fe0 and local adoption of tested all-ad expansion clipping", + "tags": [ + "writeback-result" + ], + "title": "Does point-aware alignment of poor-ad target shapes improve score within the same 4.60-second budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1490121bf3cb2a54", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T01:12:24+00:00", + "result_id": "question_0817dc8c12fb21bf9de1d35c07ad4b5f6e1f1b9a5f1936d744b90be914888b41", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:6bdf35fdd9bdb1dc", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native 1fe4e7f2 vs c14ed515 modes 0/1/2; screen 2500–2505, center fresh 2516–2547, a28d2fe0 head pairs 2560–2567; promoted b53ff507. Native untested.", + "title": "Does repairing only point-blocked poor-ad target shapes improve score within 4.60 seconds while preserving point-free originals?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3d502b1dda512d07", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-repair-only-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_0817dc8c12fb21bf9de1d35c07ad4b5f6e1f1b9a5f1936d744b90be914888b41", + "recorded_at": "2026-10-15T01:12:24+00:00", + "result_id": "finding_ccba9edec9ad3041b7cd5eefdaf8a744707e47d6ba9554f2d74dee679ffe8b28", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:a1823502b04e981a", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: b53ff507c22d9aa16782267a3f24c039d23d8ae07b51f14117abdc759171b80f Applicability: Native 1fe4e7f2 vs c14ed515 modes 0/1/2; screen 2500–2505, center fresh 2516–2547, a28d2fe0 head pairs 2560–2567; promoted b53ff507. Native untested. Observation: Both fresh means gained versus 1fe4e7f2. Conditional centering lost to always-centering on 8 matched cases; native discrepancy remains unexplained. Interpretation: Outcome: continue retained: - Native winner 1fe4e7f2, raw scoring and 4.60-second total budget - Exact source checkpoints and unfavorable matched comparisons changed: - Local candidate to blocked-only centered repair, source b53ff507 - Next local probe: neighbor-contraction-aware repaired placement stopped: - Adoption of screened-only blocked-edge clamp", + "tags": [ + "writeback-result" + ], + "title": "Does repairing only point-blocked poor-ad target shapes improve score within 4.60 seconds while preserving point-free originals?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0a45ec2dc5415589", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T01:34:24+00:00", + "result_id": "question_f6098061334c931cd036136e7982c57e452b1f999ef7cf4ac13d255dbe27b7ac", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:bfe60cf4214fab3a", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Local b53ff507 vs eb7fc1ed modes 0/1/2: screen 2600–2605, selected 1 fresh 2616–2647. Optimized 9c4b5f58 mode 1 skips fixed positions; fresh 2716–2747. Native untested.", + "title": "Does choosing repaired target-shape positions by immediate neighbor contraction score improve satisfaction within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f81475fe0b3b6b44", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-repair-position-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_f6098061334c931cd036136e7982c57e452b1f999ef7cf4ac13d255dbe27b7ac", + "recorded_at": "2026-10-15T01:34:24+00:00", + "result_id": "finding_8b5af65306992ff6d4030d0feca76b0f568a5a287ec72f95d84d7f159b576ff0", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:dca46228f5221c7d", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 9c4b5f58e470fb2b8bbc49fd6244d0a4ae45ea751d91f192baf7ecf115a1a370 Applicability: Local b53ff507 vs eb7fc1ed modes 0/1/2: screen 2600–2605, selected 1 fresh 2616–2647. Optimized 9c4b5f58 mode 1 skips fixed positions; fresh 2716–2747. Native untested. Observation: v1 fresh means -122470/+120533; v2 skips fixed positions and gives -414183/+516443. Both mixed; raw preview gains did not establish a reliable final score benefit. Interpretation: Outcome: continue retained: - Local b53ff507, separate native 1fe4e7f2, raw scoring and 4.60-second budget - Point-free original proposals, exact source checkpoints and unfavorable cases changed: - Next local probe: narrower repair eligibility after checking concentration stopped: - Adoption of tested neighbor-contraction-ranked repaired placements", + "tags": [ + "writeback-result" + ], + "title": "Does choosing repaired target-shape positions by immediate neighbor contraction score improve satisfaction within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fca1de3633750447", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-8", + "evidence:ad-repair-threshold-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_0817dc8c12fb21bf9de1d35c07ad4b5f6e1f1b9a5f1936d744b90be914888b41", + "recorded_at": "2026-10-15T01:51:36+00:00", + "result_id": "finding_a1c2c96e36357bb9e4ae18b454064810baec740e44b15f6073ce91588ff91ba1", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:a752135633adb710", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: b53ff507c22d9aa16782267a3f24c039d23d8ae07b51f14117abdc759171b80f Applicability: Native 1fe4e7f2 vs c14ed515 modes 0/1/2; screen 2500–2505, center fresh 2516–2547, a28d2fe0 head pairs 2560–2567; promoted b53ff507. Native untested. Observation: Native8 lost852499 despite50/50 valid. Narrow.85 lost both fresh means. Restored1fe4e7f2; retained submitted sources. On native-base short runs,5–8-blocker negative improving compounds occurred1278/832 times. Interpretation: Outcome: continue retained: - Native winner1fe4e7f2, raw objective, exact checkpoints and4.60-second budget - Native/local losses and source-bound accepted identities changed: - Root restored after comparing b53 submitted bytes; next probe expands regrowth blocker limit stopped: - Native adoption of b53ff507 and local adoption of tested.85 repair qualification", + "tags": [ + "writeback-result" + ], + "title": "Does repairing only point-blocked poor-ad target shapes improve score within 4.60 seconds while preserving point-free originals?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "40984bc700bfc559", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T02:04:52+00:00", + "result_id": "question_d83ebcca721d665ed191d2b0b26ee6dd823c3edc691bd344d2895caae7a9e680", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:d479869026c4c13d", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Local native1fe4e7f2 vs fc2d5a63 cap4/8/12. Screen2900–2905, selected cap12 fresh2916–2947; short diagnostics2900/2901 and full boundary tests. Native untested.", + "title": "Does increasing compound-regrowth blocker ceilings from4 to8 or12 improve score within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4efd765b03f60b4f", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-regrow-limit-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_d83ebcca721d665ed191d2b0b26ee6dd823c3edc691bd344d2895caae7a9e680", + "recorded_at": "2026-10-15T02:04:52+00:00", + "result_id": "finding_3572bcb769c1689c4b08543b5d49cf73b14b9830477d54bf081231cf5b41a263", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:16e9e088591fa15b", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: fc2d5a6343027d9e768005b3cde0552aa378632778b4491a5a350143d50cdeaa Applicability: Local native1fe4e7f2 vs fc2d5a63 cap4/8/12. Screen2900–2905, selected cap12 fresh2916–2947; short diagnostics2900/2901 and full boundary tests. Native untested. Observation: Cap12 fresh means+214336/-317364; no adoption. All124 poor baseline ads touched neighbors;87.17% of561 directed contact edges led to ads with satisfaction>=.95. Interpretation: Outcome: continue retained: - Native1fe4e7f2, exact checkpoints, raw objective and4.60-second budget - Mixed batch signs and previous native/local counterevidence changed: - Next probe: touching-neighbor move selection from poor ads stopped: - Adoption of tested cap12; cap8 remains screened only", + "tags": [ + "writeback-result" + ], + "title": "Does increasing compound-regrowth blocker ceilings from4 to8 or12 improve score within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "11d929ae5df7355b", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T02:21:47+00:00", + "result_id": "question_deb85156cbffd56ae7249a29ae3275de36d31a5da9b1b0de32f179f6a765b505", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:602335e5225f5fc3", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native1fe4e7f2 vs e1610cc9 modes0/25%/50%; screen3000-3005,selected50% fresh3016-3047. Short diagnostics3000/3001 and boundary checks. Native untested.", + "title": "Does redirecting moves from poor ads to edge-touching neighbors improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "113b03a73a5df152", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-neighbor-selection-2" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_deb85156cbffd56ae7249a29ae3275de36d31a5da9b1b0de32f179f6a765b505", + "recorded_at": "2026-10-15T02:21:47+00:00", + "result_id": "finding_f773119d927cc6ef79b5768e32bddde1508b2f7c5ddd26dfa7231c2d624b6005", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:7974f941b9436862", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: e1610cc99143e7b7273029e6ed7ebbf35c7273287d2f266f19e7e7f92e7f922f Applicability: Native1fe4e7f2 vs e1610cc9 modes0/25%/50%; screen3000-3005,selected50% fresh3016-3047. Short diagnostics3000/3001 and boundary checks. Native untested. Observation: Redirect50% failed adoption: fresh-853647/+184142, pooled-334753/32. Short runs found80/125 positive feasible shrink-plus-origin-regrowth previews; their repeated trajectory gains do not predict final score. Interpretation: Outcome: continue retained: - Native1fe4e7f2, exact checkpoints, raw acceptance and4.60-second budget - Unfavorable local/native evidence and all completed outputs changed: - Next probe: explicitly coordinate neighbor shrink with original poor-ad regrowth stopped: - Adoption of tested50% redirect;25% screened only", + "tags": [ + "writeback-result" + ], + "title": "Does redirecting moves from poor ads to edge-touching neighbors improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "576c85c68743fa7a", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T03:03:55+00:00", + "result_id": "question_4a4b9d0f1f3a7491c288a88d63a095b2782e82ce1468aab89620efcc240bb1d3", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:5f78652b11b40e98", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native1fe vs f97/f279/ef844 pair variants,25%/50%; fresh3116-3147/3216-3247,triples3300-3331; oracles/traces,cost cull; native9ba52.", + "title": "Do coordinated neighbor-shrink and poor-ad-regrowth compounds improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "65b7e661d1744b6c", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-neighbor-pair-1", + "evidence:ad-neighbor-pair-fast-1", + "evidence:ad-native-9" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_4a4b9d0f1f3a7491c288a88d63a095b2782e82ce1468aab89620efcc240bb1d3", + "recorded_at": "2026-10-15T03:03:55+00:00", + "result_id": "finding_aebef46f1edf55168b6bdc91acd8cb6d4b954df11fdbd4a242d20d5f7052faf5", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:ebb2381305c868f1", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: ba52c69d869a8ab02a8a38c5f1c79479c5460dea23a19f0945c2a180145e92b5 Applicability: Native1fe vs f97/f279/ef844 pair variants,25%/50%; fresh3116-3147/3216-3247,triples3300-3331; oracles/traces,cost cull; native9ba52. Observation: Native9/ba52 lost4046621 despite50/50;root restored1fe. Pair pooled means-150905/-337717/-91909. Contact cull preservedfixedtraces and lowereddriverCPU,but no finalscoregain was established. Interpretation: Outcome: replan retained: - Native1fe4e7f2,raw objective,4.60-second budget and exact checkpoints - All local/native losses and validated constraint/cost evidence changed: - Next probe: audit exact point-containing-proposal rejection on the native winner stopped: - Adoption of tested coordinated shrink/regrowth variants,including nativeba52", + "tags": [ + "writeback-result" + ], + "title": "Do coordinated neighbor-shrink and poor-ad-regrowth compounds improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "aadba742ae840efd", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T03:31:25+00:00", + "result_id": "question_b9b83e3f9907332facc046ebc67466f9b9f7f7b71b6582be42e795e3c961ff6c", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:4edf31e5c7993581", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native1fe vs f2d27d54 group0/2/4neighbors; screen3400-3405,selected2 fresh3416-3447; state/RNG oracles,diagnostics,boundaries; promoted e42bffdd, native untested.", + "title": "Do rare small-group ruin and greedy reconstruction improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3f83bfc21c1a229f", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-group-rebuild-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_b9b83e3f9907332facc046ebc67466f9b9f7f7b71b6582be42e795e3c961ff6c", + "recorded_at": "2026-10-15T03:31:25+00:00", + "result_id": "finding_e00299eff8baf49c6b693af9fe7938388f3772aa3c15b55610bb6f9255879f42", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:f5a77aab137e63c8", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: e42bffdd3b9ddca826c029ca7978bef91a849a03124cac6c96111108c66d4724 Applicability: Native1fe vs f2d27d54 group0/2/4neighbors; screen3400-3405,selected2 fresh3416-3447; state/RNG oracles,diagnostics,boundaries; promoted e42bffdd, native untested. Observation: Two-neighbor rebuild gained261971/630746 fresh,+446359 pooled,17/32; e42 promoted locally. Large variation and prior native/local disagreement remain; native transfer untested. Interpretation: Outcome: replan retained: - Native1fe best and exact checkpoints; raw objective and4.60-second total budget - Unfavorable local/native evidence and complete tester/state checks changed: - Local source to e42bffdd,rare group3 ruin/reconstruction - Next: source-bound native transfer; during cooldown test partial group resets stopped: - Adoption of screened-only group5 treatment", + "tags": [ + "writeback-result" + ], + "title": "Do rare small-group ruin and greedy reconstruction improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "25c14032b59d5416", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T03:47:51+00:00", + "result_id": "question_80417e49afcb5a884e68746a73d1a265d54190591eb76db5ee583ba54df9cfec", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:9f18a9232146b44e", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Local e42 vs0413 reset0/half/quarter; screen3500-3505,selectedquarter fresh3516-3547; geometry/state/RNG oracles,diagnostics,boundaries; native untested.", + "title": "Do contained half- or quarter-side seeds improve small-group reconstruction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ce8046b1acc5529e", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-group-reset-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_80417e49afcb5a884e68746a73d1a265d54190591eb76db5ee583ba54df9cfec", + "recorded_at": "2026-10-15T03:47:51+00:00", + "result_id": "finding_a6fd0f2fd65a637a8b97451b546efcabd429c46c764dda379dea4fe856319afd", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:cd78ca2df7226c2f", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 0413f0362ffd7be08eb328740ca82e2f0d0c661262564373c2cb2e930f41d798 Applicability: Local e42 vs0413 reset0/half/quarter; screen3500-3505,selectedquarter fresh3516-3547; geometry/state/RNG oracles,diagnostics,boundaries; native untested. Observation: Quarter seeds gave+632347/-249875 fresh,+191236/32,17wins. Gain did not repeat;root e42 retained. Valid containment/state/RNG checks and89outputs do not establish native transfer. Interpretation: Outcome: continue retained: - Local e42 and native1fe separately; exact checkpoints and unfavorable results - Group2/1x1 method,raw objective,4.60-second total budget changed: - Next: native transfer; during cooldown test group opportunity period stopped: - Adoption of tested quarter-side resets; half-side remains screened only", + "tags": [ + "writeback-result" + ], + "title": "Do contained half- or quarter-side seeds improve small-group reconstruction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "64a8c75309e2a47b", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-10" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_b9b83e3f9907332facc046ebc67466f9b9f7f7b71b6582be42e795e3c961ff6c", + "recorded_at": "2026-10-15T04:11:12Z", + "result_id": "finding_012c9d08d147", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:7e9930d533bfddc7", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Submission b4b066102b19 (agent-10), source e42bffdd3b9ddca826c029ca7978bef91a849a03124cac6c96111108c66d4724; score 49524696922 versus 49528162549. Earlier local +446358.53125/32 did not transfer to retained native best. Public details expose only PASS statuses. Period8192 is a different native-untested challenger. Root restored to native1fe after submitted-byte comparison.", + "title": "Native e42 at period 4096 passed 50/50 and scored 3,465,627 below the incumbent" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4349b30e93505d30", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T04:14:53+00:00", + "result_id": "question_b287a269366a3a265143881f92cd0ed85d64b9c2ab8f993aaefdae34b0618e6b", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:3d46983decac34b1", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Local e42/4096 vs5df masks2047/8191;screen3600-3605,selected8192 fresh3616-3647;head3660-3667 vs1fe/e42. Normal f64e native untested; native10 separately tests parent e42.", + "title": "Do group-reconstruction periods2048 or8192 improve satisfaction over4096 within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7e0a238006e3afa8", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-10", + "evidence:ad-group-period-1", + "evidence:ad-residual-3" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_b287a269366a3a265143881f92cd0ed85d64b9c2ab8f993aaefdae34b0618e6b", + "recorded_at": "2026-10-15T04:14:53+00:00", + "result_id": "finding_c22b94ebfcd4cc73ee677aba26199c44fd84ccdb20e797f7a10f61eda6333e18", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:d0a34b9fa6bb38e1", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: f64e28af6d059a1ca608819ca8fe98c2fd1d20e9b95ff448c8733a053ad90b43 Applicability: Local e42/4096 vs5df masks2047/8191;screen3600-3605,selected8192 fresh3616-3647;head3660-3667 vs1fe/e42. Normal f64e native untested; native10 separately tests parent e42. Observation: Native e42/4096 lost3465627 despite50/50. Period8192 gained288562/32 vs e42,but head lost986699/8 vs1fe. Restored1fe after submitted-byte comparison; f64e preserved. Local residuals remain blocked/undersized. Interpretation: Outcome: replan retained: - Native1fe best; source-bound native feedback and all unfavorable results - Period8192 challenger f64e separately; raw objective and4.60-second budget changed: - Root restored to1fe; next audit exact rejection cost on the incumbent stopped: - Native adoption of e42 and promotion of f64e over the incumbent on current evidence", + "tags": [ + "writeback-result" + ], + "title": "Do group-reconstruction periods2048 or8192 improve satisfaction over4096 within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "cfbe99ba93131af6", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T04:31:53+00:00", + "result_id": "question_03925b2e839f3062f80a70d47ff3e303ab3b5b2f557003b9b7ba67a6e1c5afb2", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:e7ca77da026b6a2d", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Local native1fe vs c42 AD_POINT_REJECT0/1; fixed traces on3416-3427; screen3700-3705,fresh3716-3747; checkpoint1b2c,boundaries. Native untested.", + "title": "Does exact early rejection of point-containing proposals save CPU and improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "bafc033f7ba6ce43", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-point-reject-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_03925b2e839f3062f80a70d47ff3e303ab3b5b2f557003b9b7ba67a6e1c5afb2", + "recorded_at": "2026-10-15T04:31:53+00:00", + "result_id": "finding_d90a6fd6e470055fc55eb2d89f3983542157e02251fb0e458781fba0b7c247af", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:f6969db1e6474615", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 1b2ca3e77025ea4613ecdac045662adfec146388c23565231ed42c69ffee77c5 Applicability: Local native1fe vs c42 AD_POINT_REJECT0/1; fixed traces on3416-3427; screen3700-3705,fresh3716-3747; checkpoint1b2c,boundaries. Native untested. Observation: 300k exact calls matched. Driver CPU medians fell3.44%/2.14%;fresh score means -400242/+549105,+74432 pooled/32. Native1fe stays root;1b2c candidate preserved. Short diagnostics found point-containing proposals in15.29% of calls. Interpretation: Outcome: continue retained: - Native1fe best and all prior local/native counterevidence; raw objective,4.60-second budget - Exact cost candidate1b2c separately,native untested; mixed final-score evidence changed: - Next test static mandatory-point rejection before rectangle scans; validate exactness and useful cost first stopped: - Score-based promotion from the mixed fresh result; this is not permanent route rejection", + "tags": [ + "writeback-result" + ], + "title": "Does exact early rejection of point-containing proposals save CPU and improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7d14b6259ae40cbe", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T04:51:02+00:00", + "result_id": "question_d7bf21e35a5020da892b55125b5ce6c4559091f8c3ecd3ef31dd8b9a5c598637", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:2bf4d703bf8075f0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native1fe vs867prefix and7ef gated2x;reused3416-3427,query/state oracles,3800-3805diagnostics,boundaries;no score sweep/native. GNU11.4/gprof2.38.", + "title": "Do static mandatory-point prefix queries reduce move evaluation cost within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "743dd11de8e43063", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-point-prefilter-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_d7bf21e35a5020da892b55125b5ce6c4559091f8c3ecd3ef31dd8b9a5c598637", + "recorded_at": "2026-10-15T04:51:02+00:00", + "result_id": "finding_baee54b256301eb9fb4881d1bbcde26aca8714002fbdcbb17b601007a38193e3", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:6c3b39b7e7cba545", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 18257363eda48b9ead849e18a94d33d9ea607288913c769921a678915099ca71 Applicability: Native1fe vs867prefix and7ef gated2x;reused3416-3427,query/state oracles,3800-3805diagnostics,boundaries;no score sweep/native. GNU11.4/gprof2.38. Observation: Prefix passed exactness but CPU rose10.66%;2x gating recovered2.99% vsunconditional,still6.99% overnative. Init~.03ms;profile query inlined,root cause unresolved. Root1fe preserved;no score sweep. Legacy cost varied across drivers. Interpretation: Outcome: replan retained: - Native1fe best,raw satisfaction,4.60-second budget and all unfavorable local/native evidence - Early1b2c separately,native untested;scope cost claims to their exact driver changed: - Switch from static queries to removing redundant cut containment checks using unchanged neighbor anchors stopped: - Score sweeps and promotion of both tested prefix variants after failed cost criteria;no universal route ban", + "tags": [ + "writeback-result" + ], + "title": "Do static mandatory-point prefix queries reduce move evaluation cost within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "33a18ae62f455d7b", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T05:08:44+00:00", + "result_id": "question_a9bab617045455016a90803b37c08476283f3cbac321270ea26e884a1a78731d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:4aca457b53707b81", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native1fe vsd7a modes0/1/2 on3416-3427 traces;3900-3905screen,3916-3947fresh,boundaries;normal2f6c native11/50cases. GNU11.4/O2.", + "title": "Do anchor-only legal-cut predicates reduce move cost and improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f7e1c05a9c7d59ed", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-cut-legality-1", + "evidence:ad-native-11" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_a9bab617045455016a90803b37c08476283f3cbac321270ea26e884a1a78731d", + "recorded_at": "2026-10-15T05:08:44+00:00", + "result_id": "finding_cc8ee7c65a29f6cf9049dfe1ba736a6fbde808aaed2952ce9db8cf13dee1ed44", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:81f043886b873767", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 2f6cdae0da953ee11c3b4ac8797d24829f5c0230fe2ea18c1d2983783851051d Applicability: Native1fe vsd7a modes0/1/2 on3416-3427 traces;3900-3905screen,3916-3947fresh,boundaries;normal2f6c native11/50cases. GNU11.4/O2. Observation: Exact predicates reduced fixed-driver CPU6.55%;fresh local scores stayed mixed. Native11 scored49540117810,+11955261,50/50. Promoted submitted2f6c after byte comparison;prior1fe retained. One timed native result does not certify transfer. Interpretation: Outcome: replan retained: - Raw satisfaction,4.60-second budget,and all negative/mixed local and native evidence - Old native1fe and unsubmitted1b2c separately;driver-specific cost claims remain scoped changed: - Root/best-native/best-supported now exact native11/2f6c;next test unrolled cuts on this incumbent stopped: - No local two-positive score qualification claimed;no prefix adoption or native retry during cooldown", + "tags": [ + "writeback-result" + ], + "title": "Do anchor-only legal-cut predicates reduce move cost and improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1b157f116ade9c12", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T05:22:33+00:00", + "result_id": "question_3691eda592e455d5b45c77456d1d5cc8f1a67040abdc68b11f7b78772e691185", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:cf58d2ecd97db8c9", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native2f6c vsc0ea modes0/1;3916-3927 saved layouts/exact traces,4000-4005screen,4016-4047fresh,boundaries. GNU11.4/O2;normal ea288 native untested.", + "title": "Does unrolling legal-cut cases reduce move cost and improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ddb9fe90f54d8956", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-cut-unroll-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_3691eda592e455d5b45c77456d1d5cc8f1a67040abdc68b11f7b78772e691185", + "recorded_at": "2026-10-15T05:22:33+00:00", + "result_id": "finding_cf1963ec2655b42ebd231810170b53df6944adea79372c0dc836a47ca65469b4", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:2e36f5ceced51635", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: ea2880053fef8a156e6111ab7692768a2c43d986b399519d257aeec0666a1bcf Applicability: Native2f6c vsc0ea modes0/1;3916-3927 saved layouts/exact traces,4000-4005screen,4016-4047fresh,boundaries. GNU11.4/O2;normal ea288 native untested. Observation: Unroll cost fell3.55%,but fresh score means+288984/-289572,pooled-294/32. Kept native2f6c/root;ea288 retained separately. Static regrowth audit shows only two distinct extents in four direction iterations;cache benefit unmeasured. Interpretation: Outcome: continue retained: - Native2f6c best,raw satisfaction,4.60-second budget and all unfavorable local/native evidence - Exact-cost route;valid unroll ea288 is a separate,native-untested contender changed: - Next measure cached regrowth extents on2f6c with exact state/RNG and cost controls stopped: - Local score promotion of unroll after mixed batches;no claim of equivalence or universal route rejection", + "tags": [ + "writeback-result" + ], + "title": "Does unrolling legal-cut cases reduce move cost and improve satisfaction within4.60seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "cf66d79a4c62eb12", + "event_kind": "node", + "evidence_refs": [ + "evidence:ad-grow-cache-1", + "evidence:ad-grow-integer-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T06:03:27Z", + "result_id": "node_18d6354b9412", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:096bf8c2b19a12e4", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Compare original2f6c with float-cache1cd9 and integer-cache4de on saved3916-3927 layouts under GNU11.4/O2;600k exact move calls,10.59M rounding checks,384million-call cost traces. No full score sweeps after failed cost criteria.", + "title": "Do cached target extents improve move cost on native2f6c?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "69b5596f571f47ca", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-grow-cache-1", + "evidence:ad-grow-integer-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "node_18d6354b9412", + "recorded_at": "2026-10-15T06:03:49Z", + "result_id": "finding_b06df8730244", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:b78ebf024c14a13a", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "GNU11.4/O2;2f6c vs1cd9 float cache:300k exact moves,192million-call traces,CPU reduction0.3224%/1.0852%,pooled0.2881%,8/12positive.2f6c vs4de integer cache:300k exact moves,10.59M numeric checks,192traces,reductions-2.3693%/-0.9356%,pooled-0.9407%,3/12positive. Each failed >=1%in both halves and>=8/12pooledpositive. All24short diagnostic outputs valid. Skipped full score sweeps and stopped further extent variants in this segment;no universal arithmetic or score-transfer claim. Sources and all unfavorable rows preserved separately.", + "title": "Both tested extent caches missed the controlled CPU adoption rule" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "85a068192c22e027", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-12", + "evidence:ad-cut-unroll-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_3691eda592e455d5b45c77456d1d5cc8f1a67040abdc68b11f7b78772e691185", + "recorded_at": "2026-10-15T06:05:47Z", + "result_id": "finding_b596d4e94b69", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:b58a2dc5c3d9bc22", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Exact frozen ea2880053fef8a156e6111ab7692768a2c43d986b399519d257aeec0666a1bcf,submission45798148902f/agent-12 accepted2026-10-15T06:02:12Z observed. Score49504247688/50valid versus retained5600e8405946/2f6c/49540117810,delta-35870122. Prior fixed-driver CPU reduction3.55%and mixed fresh-294/32 did not transfer to native gain. Current/frozen byte comparison preceded restoration of2f6c;candidate retained separately. Public details expose50PASS but no per-case scores or layouts. No permanent unrolling ban or attribution to rejected bias/extent variants;next native no earlier07:02:12Z.", + "title": "Native12 unrolled cuts passed50cases but lost35870122 against native2f6c" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0c45f5c2bc7d5a5a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-bias-quarter-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_ab81670234b2e784a2279332a8e487b8ec15ebead3b7dd8132552044d26c2f61", + "recorded_at": "2026-10-15T06:06:16Z", + "result_id": "finding_59223d3ac909", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:10f9486b559a9212", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "On native2f6c compare unchanged f8ac52cb921dc99b5a9506799f9557e9b5a5ab1a62c951afd20031b8ed4da600/default25% vs50%,same4.60second budget. Screen4200-4205 delta-779000/6;validation4216-4231 +23560/16;replication4232-4247 -1262560/16. Pooled fresh-619500/32,9wins,pairedSD1742163. All76comparisons and2policyboundaries valid;max4.64809s/RSS5708KiB,plus exact selected-root smoke valid. One candidate,alternating/reverse order,no retune or sign-based stopping. No local/native promotion. Earlier50%vsuniform result used older solver;this scoped loss does not prove globally optimal bias or predict75%.", + "title": "Reducing worst-of-two selection from50%to25% failed fresh score replication" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1fda4fa31f22047f", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T06:08:38+00:00", + "result_id": "question_f384d7aba1c65ef76c68ced0c365e07e98ebc7bba050c2873e8a59d0f89c0a80", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:487e8db659630fc0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "GNU11.4/O2;2f6c vs1cd9 float-cache and4de integer-cache;saved3916-3927 layouts,600kexactmoves/384million-calltraces;no fullscores/native for caches.", + "title": "Do float and integer extent caches meet the paired CPU cost rule on native2f6c?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ae90915475eef42d", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-grow-cache-1", + "evidence:ad-grow-integer-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_f384d7aba1c65ef76c68ced0c365e07e98ebc7bba050c2873e8a59d0f89c0a80", + "recorded_at": "2026-10-15T06:08:38+00:00", + "result_id": "finding_e37843479c3370ba28c23735d3c2c81692dcdcbfd8f4b0515892f1fad05d4365", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:c2a89417d4e0bb07", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 4de01f4e9190b3c5a625df08b96a880789cb119ebe05290b7f3918a551bcb49d Applicability: GNU11.4/O2;2f6c vs1cd9 float-cache and4de integer-cache;saved3916-3927 layouts,600kexactmoves/384million-calltraces;no fullscores/native for caches. Observation: Both passed300kexactmoves. Float-cache CPU reduction0.322%/1.085%,pooled0.288%;integer-cache-2.369%/-0.936%,pooled-0.941%. Both failed useful-cost rule;no score sweeps. Numeric10.59Mchecks passed. Interpretation: Outcome: continue retained: - Native2f6c best,raw satisfaction,4.60-second budget;all scoped local/native counterevidence - Unroll native12 regression and bias25 loss recorded separately;no universal route bans changed: - Next test75%frequency with samples2 on unchanged2f6c;native cooldown gates only submission stopped: - Further extent variants in this segment and score sweeps after failed cost criteria - Promotion of bias25 and native12/unroll after measured losses", + "tags": [ + "writeback-result" + ], + "title": "Do float and integer extent caches meet the paired CPU cost rule on native2f6c?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8bef96ee99621a59", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-bias-threequarter-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_ab81670234b2e784a2279332a8e487b8ec15ebead3b7dd8132552044d26c2f61", + "recorded_at": "2026-10-15T06:23:39+00:00", + "result_id": "finding_0b94f1178d858a5fd50fd3b07b10f3212b0b79ff702b1305de4d2240eec1f51f", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:42a8905fa918921c", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 5d2d5179d7dd074b7850cedb5277670af5ba6f67f628ca3cd93341bf4d61504e Applicability: Current regrowth solver on generated seeds 0,1,3,6,8,9 and 200–223; paired uniform and mild-bias variants under a 4.60s search budget. Observation: 75%fresh means+913080/+692922;pooled+803001/32,20wins. All79valid,max4.69018s/RSS5784KiB. Current75%layouts:99.972%loss undersized;153/158poor blocked,98thin. Local gain is not native proof. Interpretation: Outcome: continue retained: - Raw satisfaction,geometry,4.60s budget;native2f6c49540117810 kept separately - All source-bound unfavorable local/native evidence,including25%,unroll12 and extent variants changed: - Root/best-supported now exact75%5d2d after two positive fresh batches;native untested - Next test fading compactness acceptance energy;raw SAT remains cache,best and output objective stopped: - Further frequency variants in this segment;freeze75% before native/auxiliary experiments", + "tags": [ + "writeback-result" + ], + "title": "Does mildly biasing move selection toward less-satisfied ads improve score within the 4.60-second search budget?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "25421722b3a52414", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-13", + "evidence:ad-bias-threequarter-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_ab81670234b2e784a2279332a8e487b8ec15ebead3b7dd8132552044d26c2f61", + "recorded_at": "2026-10-15T07:06:41Z", + "result_id": "finding_d7edf8f45570", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:a4864dff20b600aa", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Exact5d2d/75% had fresh local+913080/+692922,pooled+803001/32. Native13/0fb595013e1e scored49530923907,50/50,delta-9193903vsnative2f6c49540117810. Comparedfrozenbytes thenrestored2f6c root/best-supported;75%checkpoint andallnegativeevidence retained. Compactness/rotation were distinctunsubmittedchildren. Scoped native/localdisagreement,no universalbiasban.", + "title": "75%selection local gain regressed in native13;restored50%incumbent" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "104fabbb02870777", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T07:07:47+00:00", + "result_id": "question_80c0b60c597f6c9e372266eae6816d1cf19c070f1ed0218ab61baf8fc915fdfc", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:3415cd924342006a", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Parent 75% 5d2d; compactness 1aad and rotation 6f7; generated 4400-4405/4416-4447 and 4500-4505/4516-4547. Native13 tested parent 5d2d only.", + "title": "Do fading compactness or rare 90-degree rotations improve packing within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9e4ded7645a68b90", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-compactness-1", + "evidence:ad-rotation-1", + "evidence:ad-native-13" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_80c0b60c597f6c9e372266eae6816d1cf19c070f1ed0218ab61baf8fc915fdfc", + "recorded_at": "2026-10-15T07:07:47+00:00", + "result_id": "finding_dc9ed2ebba4878413d92ec4141ee59143a0ff53c51c3b025ad80936815c65d7a", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:c17675122115c6f3", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 6f7af3b50eb38018c8da14e12825bb2fefd444b631288c577c5ef1e405193e6c Applicability: Parent 75% 5d2d; compactness 1aad and rotation 6f7; generated 4400-4405/4416-4447 and 4500-4505/4516-4547. Native13 tested parent 5d2d only. Observation: Compactness fresh means -105523/-305211, pooled -205367; rotation +39619/-1266868, pooled -613625. Neither qualified. Native13/5d2d scored 49530923907, 50/50, -9193903 vs 2f6c; restored incumbent after exact source comparison. 171 local outputs valid. Interpretation: Outcome: continue retained: - Raw satisfaction, geometry and 4.60s budget; native 2f6c best remains 49540117810 - All checkpoints and scoped unfavorable evidence, including positive local 75% results changed: - Restored root/best-supported to 2f6c after native13 regression - Next measure late stagnation before selecting a reset/reheat policy stopped: - Promotion of these compactness/rotation variants and native13 candidate", + "tags": [ + "writeback-result" + ], + "title": "Do fading compactness or rare 90-degree rotations improve packing within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ab265d80f6d57f28", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T07:36:49+00:00", + "result_id": "question_266f993ce456486507f4ac323dd1380dfbd0cd473d05b20cc0650a46015a138d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:f091629f62bf7c74", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native 2f6c/50%; diagnostic 20c174 seeds4600-4615; combined81e seeds4700-4705/4716-4747; reset-onlyb2 seeds4800-4805/4816-4847. Normal4.60s.", + "title": "Does late stagnation justify resets, with or without a short heat pulse, within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e88ca4c378570ad5", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-stagnation-1", + "evidence:ad-restart-1", + "evidence:ad-reset-only-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_266f993ce456486507f4ac323dd1380dfbd0cd473d05b20cc0650a46015a138d", + "recorded_at": "2026-10-15T07:36:49+00:00", + "result_id": "finding_a4720f6091a9e0b8c1adc7ff2cce7baa1be55aa870f38ee33cecd5950ac70397", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:2e9477b46c178f44", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: b2b6e75be63070c6b455cfb4de536025a41cf0c036ca107138a0bd2086ea52aa Applicability: Native 2f6c/50%; diagnostic 20c174 seeds4600-4615; combined81e seeds4700-4705/4716-4747; reset-onlyb2 seeds4800-4805/4816-4847. Normal4.60s. Observation: 16/16 cases had 71%-97% late meaningful-stale samples. Combined fresh means -31666/+178284, pooled +73309/32; reset-only -165592/-1283814, pooled -724703/32. Neither qualified. All 186 outputs valid; retained native 2f6c. No new native submission. Interpretation: Outcome: continue retained: - Native 2f6c winner, raw satisfaction and 4.60s budget; all scoped local/native counterevidence - All tested sources and complete pairs, including positive combined screen and pooled mean changed: - Validated observer, controller and resets; 900k fixed moves plus independent accounting checks - Next compare heat-only policy on native parent, preserving state at the pulse trigger stopped: - Local promotion of these fixed combined and reset-only policies; no universal route ban", + "tags": [ + "writeback-result" + ], + "title": "Does late stagnation justify resets, with or without a short heat pulse, within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "a195e6cd191a63ba", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-native-14", + "evidence:ad-restart-1", + "evidence:ad-heat-only-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_266f993ce456486507f4ac323dd1380dfbd0cd473d05b20cc0650a46015a138d", + "recorded_at": "2026-10-15T08:16:24Z", + "result_id": "finding_c6a276e98b1d", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:73774c2ac63c47e8", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Exact combined restart81e native14/acd2f362485c scored49555673017,50/50,+15555207 versus native11/2f6c. Compared frozen submitted/current sources before adopting81e to root,best-supported and best-native. Combined local fresh-31666/+178284 remain mixed; heat-only2452 fresh+163138/-758362,pooled-297612/32,not promoted and not submitted. Native result supports this fixed combination; it does not isolate component effects. Cap prototypes remain separate. Next native eligible>=09:03:21Z from acceptance observation.", + "title": "Combined restart became native best; heat-only stayed mixed" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1eef9a1d81ebbd85", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T08:23:36+00:00", + "result_id": "question_412941fac497498beb43bf2e1a25366012b381209f3cbb76c5375906616cc055", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:b47298da35f12b39", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Parent native14/81e; cap121, normal4.60s, generated5100-5105/5116-5147. Rejection diagnostic used prior2f6c. Native14 evaluated81e only.", + "title": "Does retrying point-blocked expansion with an anchor-capped edge improve native14 within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "6ce1b3f52f5e73d2", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-rejection-1", + "evidence:ad-anchor-cap-native14-1", + "evidence:ad-native-14" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_412941fac497498beb43bf2e1a25366012b381209f3cbb76c5375906616cc055", + "recorded_at": "2026-10-15T08:23:36+00:00", + "result_id": "finding_b300007c0e88e08564d1a949b6cdc870bba08903dd964d84b7628b55b298f9f3", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:c74d3e07122e9b33", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 121686c909bb70b0f0ccf3839258467f1ce49e144becfcc9b5ee869f03e09c29 Applicability: Parent native14/81e; cap121, normal4.60s, generated5100-5105/5116-5147. Rejection diagnostic used prior2f6c. Native14 evaluated81e only. Observation: Native14/81e gained 15555207 with 50/50. Heat-only fresh +163138/-758362; cap121 -972111/+292679, pooled -339716/32. Neither qualified locally. Cap geometry and 300k rebased state/RNG steps passed; all 193 local outputs valid. Interpretation: Outcome: replan retained: - Raw satisfaction, geometry and 4.60s budget; all source-bound local/native counterevidence - Native14/81e remains separate from unsubmitted heat-only2452 and cap121 changed: - Adopted exact native14/81e to root, best-supported and best-native after source comparison - Next measure capped retry outcomes by phase/SAT before choosing a focused gate stopped: - Promotion of heat-only2452 and cap121 under the fixed local rule; no universal route ban", + "tags": [ + "writeback-result" + ], + "title": "Does retrying point-blocked expansion with an anchor-capped edge improve native14 within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2f94034b9d4b3946", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T08:44:39+00:00", + "result_id": "question_74e30f8a9e6dd5ad2e00e7a4a828255aa9159733531c1dd18bd690f61017cf78", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:1293a95167a76222", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native14/81e parent; focus4d2, normal4.60s, generated5300-5305/5316-5347. Diagnostic on unconditional cap121 used5200-5215; no new native evaluation.", + "title": "Does restricting anchor-capped retries to ads below 95% satisfaction improve native14 within 4.60 seconds?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "801d24b1662940b0", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-cap-outcomes-1", + "evidence:ad-cap-focus-1", + "evidence:ad-anchor-cap-native14-1" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_74e30f8a9e6dd5ad2e00e7a4a828255aa9159733531c1dd18bd690f61017cf78", + "recorded_at": "2026-10-15T08:44:39+00:00", + "result_id": "finding_eead1a0cfd05cdb722bd580dc07327fd374f0053cec90211988b80b45ab379c9", + "run_id": "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd:e34b45111db62f56", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 4d2dcc7a630fd4eb7cd891e8ef972f172e2d43c40bf73fd9afe7591fdc245605 Applicability: Native14/81e parent; focus4d2, normal4.60s, generated5300-5305/5316-5347. Diagnostic on unconditional cap121 used5200-5215; no new native evaluation. Observation: SAT>=.95 caused 86.6% of retries but also gains. Focus4d2 fresh means -646966/-1053295, pooled -850130/32,11 wins. Both failed. New 600k state/RNG/accounting steps passed;101 local outputs valid. Retained native81e; no new native result. Interpretation: Outcome: continue retained: - Raw satisfaction, geometry and 4.60s budget; native14/81e best 49555673017 - Complete unconditional/focused cap pairs and valid4d2 as a separate native-untested checkpoint changed: - Measured cap outcomes by phase/SAT and validated the focused gate independently - Next audit maximum anchored empty rectangles to separate fixed-neighbor room from proposal limitations stopped: - Local promotion of4d2 and further frequency variants in this segment; no universal cap or native-evaluation ban", + "tags": [ + "writeback-result" + ], + "title": "Does restricting anchor-capped retries to ads below 95% satisfaction improve native14 within 4.60 seconds?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d1dc57d3d1d330f5", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T12:00:33+00:00", + "result_id": "question_bbd4c0bac14dad72981efa45a500c9428495bd7b0b3e373296ef1ad9534f6031", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source3809 versus native81e;normal4.60s,local6700-6731,shared-prefix diagnostic6700-6715,native agent-17 a70f6773ad73.", + "title": "Does a4.58s triad interruption with original cooling and returned unused time improve native81e?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f4c4d8c3f0b5c237", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-triad-continuation-native-1", + "evidence:ad-triad-continuation-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_bbd4c0bac14dad72981efa45a500c9428495bd7b0b3e373296ef1ad9534f6031", + "recorded_at": "2026-10-15T12:00:33+00:00", + "result_id": "finding_3d3bce36c72bcd19496580c330f30c13fc7efc148d3c87733670c40cc637ada8", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 38096ac7daac48b06c0f5e275661fda6e15cbac1e52ae4f1782e9214b12362c8 Applicability: Source3809 versus native81e;normal4.60s,local6700-6731,shared-prefix diagnostic6700-6715,native agent-17 a70f6773ad73. Observation: Native3809 passed50/50 but lost9925530; conditional shared-prefix gains did not establish whole-solver improvement. Current root matches81e after source comparison and valid smoke. Interpretation: Outcome: replan retained: - Native best81e,raw satisfaction,geometry and5s/1GB - 3809 checkpoint,114prior valid outputs and all scoped local/native losses changed: - Completed source-bound native continuation and restored verified81e - Phase-room successor now uses native loss and preserves restored root stopped: - Promotion or repeat submission of unchanged3809 - Inferring universal joint-move failure or causal explanation from aggregate score", + "tags": [ + "writeback-result" + ], + "title": "Does a4.58s triad interruption with original cooling and returned unused time improve native81e?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "04f567fbb5380a37", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T12:16:44+00:00", + "result_id": "question_5935a0503ad58db6a52e9ac8d5be2810b03bf38d7442ff66efd98c8765d9572f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source81e,observer0e5c,normal4.60s,fresh6800-6815,60%/80%/4.58s;unchanged kernel and sequential offline witnesses.", + "title": "Do earlier81e best states offer materially more realizable triad room than4.58s states?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5eb6b14b150c9014", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-phase-triad-room-1", + "evidence:ad-triad-continuation-native-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_5935a0503ad58db6a52e9ac8d5be2810b03bf38d7442ff66efd98c8765d9572f", + "recorded_at": "2026-10-15T12:16:44+00:00", + "result_id": "finding_db53b29e5b638e48d00658fd7d9ca5eef3ef6f3f55713fef3c53677a31390c13", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 0e5ca741b5c79a26059373b65fe5d74f2dd3392587003f44852c75021383faf3 Applicability: Source81e,observer0e5c,normal4.60s,fresh6800-6815,60%/80%/4.58s;unchanged kernel and sequential offline witnesses. Observation: Earlier wins12/16and10/16cases but means are0.628/0.546of late;late gain is concentrated. Offline feasible gains recover1.038/0.918/1.694%normalized residual at3.10/3.36/3.40ms mean sweep cost. Neither earlier phase meets frozen routing criterion. Interpretation: Outcome: replan retained: - Native best81e49555673017,raw satisfaction,geometry and5s/1GB - 3809 loss9925530 and all complete scoped local/native counterevidence changed: - Validated read-only phase observer and complete source-bound offline query/replay - Next test expands feasible three-ad region before another timing/frequency variant stopped: - Claiming native benefit from offline gains or majority signs alone", + "tags": [ + "writeback-result" + ], + "title": "Do earlier81e best states offer materially more realizable triad room than4.58s states?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b07d11d812dd144b", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T12:31:33+00:00", + "result_id": "question_a2790fa9aa8dfa49c86a816874e3f1cd1ea701df9ee297e215d42c7e648fc25e", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact81e,extended helper8587,box/board/union offline sweeps,fresh6900-6915,normal4.60s final layouts;fixed foreign ads.", + "title": "Does expanding the feasible region materially improve three-ad allocation on81e final layouts?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0e15713e0a9bfd34", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-expanded-triad-room-1", + "evidence:ad-phase-triad-room-1", + "evidence:ad-triad-continuation-native-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_a2790fa9aa8dfa49c86a816874e3f1cd1ea701df9ee297e215d42c7e648fc25e", + "recorded_at": "2026-10-15T12:31:33+00:00", + "result_id": "finding_c3e28eab07e387f2cb160bd806d5f6808249b1b77b5b6796b56f5dd6a0fd8c5f", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 8587f24faf3d86dd1b4fb31ea4718389c9a6735a64f0c8bb29795622149618d1 Applicability: Exact81e,extended helper8587,box/board/union offline sweeps,fresh6900-6915,normal4.60s final layouts;fixed foreign ads. Observation: Union mean increment314,median0.010points;11raw positives include6subpoint changes,only5official gains. Extra mean8.08ms,max full sweep51.12ms. Frozen rule favors runtime; complete gain/cost/resolution evidence favors a fourth-blocker probe next. Interpretation: Outcome: replan retained: - Native best81e49555673017,raw objective,geometry and5s/1GB - 3809 and prior native/local losses;full data and original frozen recommendation changed: - Validated explicit-region extension and box/board/union sequential witnesses - Replanned from nominal sign-count recommendation to four-ad allocation research stopped: - Treating subpoint raw gains as material native-scale progress - Promoting a region candidate from conditional witness gains alone", + "tags": [ + "writeback-result" + ], + "title": "Does expanding the feasible region materially improve three-ad allocation on81e final layouts?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "85e36e604fb8d24d", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T12:45:23+00:00", + "result_id": "question_98bf1547fa6e6e620cd272423557ad6a20d68ef29e41aa91d5d6cbe90bc7fddf", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact81e,four helperd3dd,initial deficient-center contact groups,fresh7000-7015,normal4.60s finals;foreign ads fixed.", + "title": "Does releasing a fourth touching blocker materially improve joint allocation over matching triads on81e final layouts?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4277619acf216052", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-four-ad-room-1", + "evidence:ad-expanded-triad-room-1", + "evidence:ad-triad-continuation-native-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_98bf1547fa6e6e620cd272423557ad6a20d68ef29e41aa91d5d6cbe90bc7fddf", + "recorded_at": "2026-10-15T12:45:23+00:00", + "result_id": "finding_46844c13377dfd5d0ec1c8c9f03213230d4a4e1dc5771f0c31e8a0f10e57fe4b", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: d3dd213ada5df32616372abdb914fcc933fa76b62e918570cb8339523a09fc23 Applicability: Exact81e,four helperd3dd,initial deficient-center contact groups,fresh7000-7015,normal4.60s finals;foreign ads fixed. Observation: Union adds1289mean/0median points over matching triads;3official gains,96%from7000. Extra50.55ms,full sweep57.20mean/150.53maxms. Eligible centers hold78.65%residual but incremental recovery is0.01425%. Interpretation: Outcome: replan retained: - Native best81e49555673017;raw objective,integer geometry,5s/1GB - 3809and prior pair/group/shrink-regrow native/local losses changed: - Validated memoized four-way guillotine search and matching triad controls - Select joint blocker-trim choices before regrow,without common slicing requirement stopped: - Copying expensive full four-ad sweep into runtime from conditional witness gain", + "tags": [ + "writeback-result" + ], + "title": "Does releasing a fourth touching blocker materially improve joint allocation over matching triads on81e final layouts?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "6e71a650f84ad98d", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T13:00:59+00:00", + "result_id": "question_fdea237f1b9cb741e1748f28b342afa63c2d2d9f70d24f0ae66d7bff570bb601", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact81e,trim helper596a,fresh7100-7115finals,coarse frozen proposals plus separately frozen adaptive boundary reuse;<=3blockers,foreign ads fixed.", + "title": "Do joint blocker-trim choices improve attainable end-state gain beyond81e independent-best trim and regrowth?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e0d5aaaa45aecc0a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-blocker-trim-room-1", + "evidence:ad-four-ad-room-1", + "evidence:ad-triad-continuation-native-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_fdea237f1b9cb741e1748f28b342afa63c2d2d9f70d24f0ae66d7bff570bb601", + "recorded_at": "2026-10-15T13:00:59+00:00", + "result_id": "finding_455ec6c0dbd136e5127a0bb6488e45e8b3b7bfc586071aa47b66fe20a9070511", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 596a0a58d14ae88d1db09b175f22d97948e88ec99a0c2edfd3b1ee108ae026d0 Applicability: Exact81e,trim helper596a,fresh7100-7115finals,coarse frozen proposals plus separately frozen adaptive boundary reuse;<=3blockers,foreign ads fixed. Observation: Both coarse and adaptive fine schedules showzero joint-choice increment. Finer base proposals instead gain6376mean points on16/16reused layouts at2.31meanms. Adaptive diagnostic changes proposal scale;native benefit unknown. Interpretation: Outcome: replan retained: - Native best81e49555673017;raw objective,integer geometry,5s/1GB - 3809and earlier triad,pair,random group/shrink-regrow losses changed: - Validated joint trim search,exact base control,rollback and real transactions - Select bounded deterministic base boundary polish;preserve parent annealing timeline stopped: - Adding orientation/order enumeration without demonstrated end-state gain", + "tags": [ + "writeback-result" + ], + "title": "Do joint blocker-trim choices improve attainable end-state gain beyond81e independent-best trim and regrowth?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "db4104b1fb367288", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T13:30:43+00:00", + "result_id": "question_834801b15c96f5c64b78a85613e64820251184825956baaf98ea1fab546e6fea", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exactd71vs81e,one10ms-capped final base pass,32fresh normal pairs7200-7231andnative48eb5473e82b/agent18;5s/1GB.", + "title": "Does the original-clock post-greedy boundary-polish policy improve native raw score over81e?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "57b37be2ff401e81", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-boundary-polish-native-1", + "evidence:ad-boundary-polish-1", + "evidence:ad-boundary-review-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_834801b15c96f5c64b78a85613e64820251184825956baaf98ea1fab546e6fea", + "recorded_at": "2026-10-15T13:30:43+00:00", + "result_id": "finding_6cfeb36900ecbbe2bcffa048107ba4ca0210baf8f95b5a2d5f2244f30ed15bed", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: d71c786874c423547e9b597a29dab000503acb97fd66c862c24ee0e6b4346841 Applicability: Exactd71vs81e,one10ms-capped final base pass,32fresh normal pairs7200-7231andnative48eb5473e82b/agent18;5s/1GB. Observation: Semantic prefix and actual post-greedy commits are verified,all outputs valid.32normal pairs are mixed;one native result loses23189487. Root81e restored after exact source comparison. Compiled-prefix throughput remains unmeasured;no cause inferred. Interpretation: Outcome: replan retained: - Native best81e49555673017;raw objective,5s/1GB,integer geometry - d71/3809and all earlier native/local losses,complete source-bound evidence changed: - Completed candidate/native continuation and restored supported incumbent - Next distinguish compiled-prefix throughput from conditional pass gain and timed variance stopped: - Promoting d71 from offline gain or successful output validity - Another late frequency/reservation variant without new discriminating evidence", + "tags": [ + "writeback-result" + ], + "title": "Does the original-clock post-greedy boundary-polish policy improve native raw score over81e?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "211a9b848b8a7fe9", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T13:48:59+00:00", + "result_id": "question_f5005110a0c1b31702898f5779a83009a7e3aad3d8de4141e14b6f2d955d141c", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact81e/d71/GCC11.4-O2,four retained7200/7203/7214/7230 inputs,4M fixed moves,CPU0,balanced orders;not native causality.", + "title": "Does adding d71 boundary-polish code consistently slow an otherwise equivalent parent search, and does outlining correct it?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e0dcb06789439ed6", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-prefix-throughput-1", + "evidence:ad-boundary-polish-native-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_f5005110a0c1b31702898f5779a83009a7e3aad3d8de4141e14b6f2d955d141c", + "recorded_at": "2026-10-15T13:48:59+00:00", + "result_id": "finding_75100e751792d35924b48cd9e96f142b9362975e04a5cfef0986e5103e7ba4be", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 81e58e2378ef6bf7379a638dc3f66a5129b879e8a6213bf4f86911797ceb08bc Applicability: Exact81e/d71/GCC11.4-O2,four retained7200/7203/7214/7230 inputs,4M fixed moves,CPU0,balanced orders;not native causality. Observation: Layout-sensitive local CPU effects exist,but disabled is often faster and outlining has no consistent benefit. Profiles place92-94%in unchanged attempt/regrow. No causal native explanation or score gain is established. Interpretation: Outcome: replan retained: - Native best81e49555673017,integer geometry,4.60s/5s/1GB - d71/3809and prior unrolling/pair/group native/local losses changed: - Completed controlled throughput,hot-code and C++ profile readback - Next test exact SIMD overlap scan on incumbent with unchanged search semantics stopped: - Promoting noinline correction or inferring native loss from code size - Further arbitrary layout tuning on four reused inputs", + "tags": [ + "writeback-result" + ], + "title": "Does adding d71 boundary-polish code consistently slow an otherwise equivalent parent search, and does outlining correct it?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "21146d952b4301ad", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T14:11:05+00:00", + "result_id": "question_339452d5d092b904c5ce404b369367cf6a26845fdf21e44936f847825679cc3f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact9dc/SSE2 versus81e GCC11.4-O2;4retained fixed-work fixtures and32fresh7300-7331pairs;native pending.", + "title": "Does exact SIMD overlap filtering on81e improve whole-solver native raw score through lower geometry cost?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3fc2bc5dc66383c6", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-simd-overlap-1", + "evidence:ad-prefix-throughput-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_339452d5d092b904c5ce404b369367cf6a26845fdf21e44936f847825679cc3f", + "recorded_at": "2026-10-15T14:11:05+00:00", + "result_id": "finding_ae361e99732e2ec1510f8f81387e661a5e38ecd1ef7e1d241c23085ef8d2680b", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 9dcadc6a4646c25e05cbfc109bfaccbad9edfb04009e4714db3feea70fbe24f0 Applicability: Exact9dc/SSE2 versus81e GCC11.4-O2;4retained fixed-work fixtures and32fresh7300-7331pairs;native pending. Observation: Predicates and complete trajectories match;controlled cost is lower. Fresh score comparison is uncertain. Root9dc is prepared,not native-adopted;cooldown pending. Best81e and all failed checkpoints remain intact. Interpretation: Outcome: continue retained: - Raw native score objective,integer geometry,4.60s/5s/1GB and best81e49555673017 - Prior d71/3809/unrolling and neighborhood losses changed: - Implemented and froze9dc;completed local proof,cost and fresh comparisons - Bound native readiness Monitor and independent read-only-root P2 continuation - Recover original bound writeback after pre-writeback wait projection failed stopped: - Claiming native improvement or completion from CPU/local checks", + "tags": [ + "writeback-result" + ], + "title": "Does exact SIMD overlap filtering on81e improve whole-solver native raw score through lower geometry cost?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "028fcedccb3d3193", + "event_kind": "node", + "goal_id": "ad-warm-off-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T14:23:55+00:00", + "result_id": "question_e28c5409393bb4c2918ca9093763b356a3dbc5796ba3807556925ccc46b0625c", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "7102vs9dc,GCC11.4-O2/CPU0,4reused55-192ad inputs;semantics and local throughput only.", + "title": "Does exact SSE2 obstacle-bound reduction in growAvailable improve total fixed-work cost on9dc?" + }, + { + "agent_id": "warm-off-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "40de83b7cae96371", + "event_kind": "finding", + "evidence_refs": [ + "evidence:ad-simd-grow-1", + "evidence:ad-simd-overlap-1", + "checkpoint:112c48732389b1d06d32e206058e37823611657e3173e146affdda8a7d2e9bbd" + ], + "goal_id": "ad-warm-off-20261015-01", + "node_id": "question_e28c5409393bb4c2918ca9093763b356a3dbc5796ba3807556925ccc46b0625c", + "recorded_at": "2026-10-15T14:23:55+00:00", + "result_id": "finding_61f797e745ccf3f328e36152a98098cdd4a549f11396823c571b48d3b6c4e274", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 7102cfee317acb20ef0b101da01956bfb1b3e1525b48eab5a56f582022ae9942 Applicability: 7102vs9dc,GCC11.4-O2/CPU0,4reused55-192ad inputs;semantics and local throughput only. Observation: Geometry and complete trajectories match;balanced total timings show no supported cost benefit or size threshold.7102 stays local;9dc native request remains pending. Interpretation: Outcome: replan retained: - Raw native-score objective;integer geometry,4.60s/5s/1GB;best81e49555673017 - Prepared9dc and prior d71/3809/unroll/neighborhood counterevidence changed: - Completed independent growAvailable SIMD proof and total-cost comparison - Return to original9dc source-bound native continuation stopped: - Promoting7102 or tuning a size threshold on4reused fixtures", + "tags": [ + "writeback-result" + ], + "title": "Does exact SSE2 obstacle-bound reduction in growAvailable improve total fixed-work cost on9dc?" + } + ] +} diff --git a/apps/presentation/site/public/lab/time-machine/data/graph.json b/apps/presentation/site/public/lab/time-machine/data/graph.json new file mode 100644 index 0000000000..582a136137 --- /dev/null +++ b/apps/presentation/site/public/lab/time-machine/data/graph.json @@ -0,0 +1,3486 @@ +{ + "description": "Unchanged Explore history from a real, sealed research checkpoint later restored by follow-up experiments. Original text, timestamps, identities and references; no invented relationships.", + "source_sha256": "b105de360ed7d015ae5d7bf96b545879e12507b1fc97eaeb5fb6fa2c0bbc00ab", + "source_bytes": 200333, + "goal_id": "graph-warm-on-20261015-02", + "display_name": "Graph · 近期研究 checkpoint", + "provenance": { + "case_id": "graph_node_classification", + "source_run_id": "graph-warm-stride-on-20261015-02", + "checkpoint_id": "4af468bfd8b84edb9cea23c8665ecaca", + "checkpoint_sha256": "9e65de32844ceb15931bc8aa9809eab6e626fcb85258845fd0adf628d2c4dee1", + "captured_at": "2026-10-15T13:47:07.731851+00:00", + "captured_native_turns": 100, + "selection": "Reuse the on-arm point independently selected earlier by longest captured native history,100 Turns, unrelated to score. Known hard-case plateau means purposive canary, not representative/general effect.", + "restored_by": [ + "graph-replan-fresh-20261016-05", + "graph-replan-retained-20261016-05" + ], + "redactions": [], + "claim_scope": "diagnostic_only", + "score_countable": false, + "trajectory_publication": { + "status": "uploaded", + "repository": "https://huggingface.co/datasets/huangrt01/loopx-edgebench-research-checkpoints-20261015", + "revision": "c858f51021dc1c1cf08263f9c04b692a5e29b40b", + "archive": "graph_node_classification.zip", + "archive_sha256": "46192895c4c3ed2a88036f5b2c59c66670d69e3943559ad28786de19e247b97c" + }, + "boundary": "Captured Explore knowledge only; original native execution state, credentials, evaluator and models are not imported." + }, + "source_counts": { + "event_count": 134, + "event_kinds": { + "node": 66, + "finding": 68 + }, + "node_kinds": { + "question": 66 + }, + "statuses": { + "open": 66, + "confirmed": 18, + "tentative": 16, + "refuted": 34 + }, + "first_recorded_at": "2026-10-14T10:10:59+00:00", + "last_recorded_at": "2026-10-15T13:45:49+00:00", + "edge_count": 0, + "parent_records": 0 + }, + "events": [ + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5058ae93d3c152a5", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T10:10:59+00:00", + "result_id": "question_d1c06e650ab08c9f17d46e766d4014990a667e618962bf14e30eb0dbd855ef0c", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:8f1538d1beb465fe", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Synthetic five-class graphs with five degree regimes, visible anchor labels and held-out target nodes.", + "title": "Can degree-conditioned neighbor-label features improve classification when class compatibility changes with node degree?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0e29ef7aa5ca4c70", + "event_kind": "finding", + "evidence_refs": [ + "experiments/conditional_degree_results.json", + "experiments/interaction_path_validation.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_d1c06e650ab08c9f17d46e766d4014990a667e618962bf14e30eb0dbd855ef0c", + "recorded_at": "2026-10-14T10:10:59+00:00", + "result_id": "finding_6bf2223ebaf51c1b5f36d4df053de90eaf84fb2e84a2a878ed974ee022d107bb", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:666326daedda2947", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: f0c468856ff3308ef875bc5a7303db357b5e6c45285738041b55e79547ab4189 Applicability: Synthetic five-class graphs with five degree regimes, visible anchor labels and held-out target nodes. Observation: Across 3 seeds, v1 test accuracy was 39.5-45%; v2 reached 100%. With strong word features both reached 100% and v2 retained the original feature model. Interpretation: Degree interactions cover an additive-feature limitation on these generated graphs. Squirrel improvement is unproven; native evaluation remains pending.", + "tags": [ + "writeback-result" + ], + "title": "Can degree-conditioned neighbor-label features improve classification when class compatibility changes with node degree?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "32fc61e89f94a9f6", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T10:25:20+00:00", + "result_id": "question_f0e9d0057d3db3b6d5e2fe471c6281f53be9bc391afc5ecca5c5fe3c8bee9a82", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:8bce2e75fbf51380", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated five-class target graphs with masked anchors; 54 cases across three seeds, signal strength, hub noise and train coverage.", + "title": "How do graph-column normalization strengths compare on rare-anchor and frequent-nuisance-hub classification probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "05cb45cb6dd309a7", + "event_kind": "finding", + "evidence_refs": [ + "experiments/normalization_sweep.json", + "experiments/normalization_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_f0e9d0057d3db3b6d5e2fe471c6281f53be9bc391afc5ecca5c5fe3c8bee9a82", + "recorded_at": "2026-10-14T10:25:20+00:00", + "result_id": "finding_c91ec76fe1e48044a79d59ad12fbc435683dd82f08bd09d1539b2276604fdab1", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:8dc99ccec79d8b0b", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 7dd7fd69d5299524de232b8515d1ff74d6203ee4ad55d5bf187cf788cbbece82 Applicability: Generated five-class target graphs with masked anchors; 54 cases across three seeds, signal strength, hub noise and train coverage. Observation: Mean target accuracy: quarter power 0.973156, half power 0.973202, raw 0.972392. Half-minus-quarter effects changed sign by seed. Interpretation: No consistent benefit supports adding normalization search to v2 from these probes. Keep this scoped negative evidence; it does not prove a Squirrel or universal preference.", + "tags": [ + "writeback-result" + ], + "title": "How do graph-column normalization strengths compare on rare-anchor and frequent-nuisance-hub classification probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1191ebc32ecca4d5", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T11:21:17+00:00", + "result_id": "question_837813bd922eafba85e8199e8d9ecc144502cf411d882a81eec9fea1897f6ce0", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:7a62316fb09afa32", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Owned generated correlated-kernel and full-size noisy joint-neighborhood fixtures; two CPU threads.", + "title": "Does unconditional direct solution of the quadratic ridge system improve whole-operation CPU time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d40ab0aaa592b7c0", + "event_kind": "finding", + "evidence_refs": [ + "experiments/kernel_solver_diagnosis.json", + "experiments/full_solver_diagnosis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_837813bd922eafba85e8199e8d9ecc144502cf411d882a81eec9fea1897f6ce0", + "recorded_at": "2026-10-14T11:21:17+00:00", + "result_id": "finding_8153c1c20750584818a8abf7038571330b8afa9633b502cb1febc36b4ec039ec", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:df760c7507418855", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: c9b28aa0afcfe22b240eef98a9f6932d8456679b14ac4672f2eae563c545339d Applicability: Owned generated correlated-kernel and full-size noisy joint-neighborhood fixtures; two CPU threads. Observation: Three warmed alternating pairs: direct/CG median time 0.467 for the kernel solve, 1.006 for the full operation. Predictions agreed; max full confidence difference 0.000021. Interpretation: Component benefit did not transfer to the full fixture. Retain native v3; test conditioning-specific solver choice. This does not establish a native-worker bottleneck.", + "tags": [ + "writeback-result" + ], + "title": "Does unconditional direct solution of the quadratic ridge system improve whole-operation CPU time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "17fe29d3c78722c1", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T11:38:59+00:00", + "result_id": "question_1d51acb49056043e9f17409f391ef9ee122b9d2f0d73fc73d12af0f670ab585d", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:c399fb557f354882", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated correlated kernels and full-size graphs; fixed inputs, two CPU threads.", + "title": "Can selective direct solving preserve predictions while reducing CPU cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "6ba5f512dc1b5917", + "event_kind": "finding", + "evidence_refs": [ + "experiments/adaptive_analysis.json", + "experiments/adaptive_kernel_refined.json", + "experiments/adaptive_full_grid.json", + "experiments/runtime_adaptive_skip.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_1d51acb49056043e9f17409f391ef9ee122b9d2f0d73fc73d12af0f670ab585d", + "recorded_at": "2026-10-14T11:38:59+00:00", + "result_id": "finding_37a707b5dc22a329050bd643c24cb06a7d26a847d47105550174a06a12359fd9", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:54a2d6a8839255a4", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: b3f47a70024a30978116c95e3c68a47b00a0207f7c82ff61c55c523d9d175e5c Applicability: Generated correlated kernels and full-size graphs; fixed inputs, two CPU threads. Observation: Refined fallback took 0.773x and 0.841x on slow kernels; fast case 0.996x. Four full graph outputs matched exactly. Feature-heavy median time ratio 1.027, interval 0.995-1.181. Interpretation: Scoped component gains support native testing of v4. Mixed whole-operation timings do not prove a native gain. Best-supported v3 remains separate; the broader cutoff was dropped.", + "tags": [ + "writeback-result" + ], + "title": "Can selective direct solving preserve predictions while reducing CPU cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ce9a453d568f452d", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T12:10:19+00:00", + "result_id": "question_f35d7d8d581df5272c349683357afb47da23378eaedc773da660ac558f60b095", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:380b38518be78cce", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated repeated-row kernels and full-size graphs with 0-85% duplicate neighborhoods; two CPU threads.", + "title": "Can exact duplicate-neighborhood compression reduce quadratic ridge cost while preserving label multiplicities?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "045dd02b7ccca667", + "event_kind": "finding", + "evidence_refs": [ + "experiments/compression_analysis.json", + "experiments/compression_bypass_grid.json", + "experiments/validation_compression.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_f35d7d8d581df5272c349683357afb47da23378eaedc773da660ac558f60b095", + "recorded_at": "2026-10-14T12:10:19+00:00", + "result_id": "finding_9e2ccbbf7bfba71615c6c1aeafefa885a4a6fdb1acb46f29d5a31751959c82a8", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:a4896283d03b0d93", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 7c6bd0e45a6d6b75612e3bac74dc4715b26296198b94c93c40eacaa3c0784307 Applicability: Generated repeated-row kernels and full-size graphs with 0-85% duplicate neighborhoods; two CPU threads. Observation: Nine double-solve oracle cases had max score error 0.000123. Full-size high-duplication runtime ratio 0.649; predictions matched, max confidence difference 0.000043. No-duplicate ratio 1.021. Interpretation: Label weights and the ridge objective are retained. Native utility is unknown; 30%/60% duplicate probes skipped quadratic screening and do not measure compression cost.", + "tags": [ + "writeback-result" + ], + "title": "Can exact duplicate-neighborhood compression reduce quadratic ridge cost while preserving label multiplicities?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4f3bc80311fb6e5f", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T12:43:47+00:00", + "result_id": "question_37f97bda50ba6caf8d7f58fd029c6570aa6811a46fc53b42f1cad8c4a36d2922", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:89b836be049d01a1", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated full-size joint-neighborhood graphs with partial duplication and feature controls; two CPU threads.", + "title": "Can linear-family preselection hide useful quadratic neighborhood predictions?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1416dd99233fb18c", + "event_kind": "finding", + "evidence_refs": [ + "experiments/quality_gate_grid.json", + "experiments/quality_gate_negative.json", + "experiments/validation_quality_gate_path.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_37f97bda50ba6caf8d7f58fd029c6570aa6811a46fc53b42f1cad8c4a36d2922", + "recorded_at": "2026-10-14T12:43:47+00:00", + "result_id": "finding_b5860e437b2905fbc0b9db2bbdc3ba567783817eee67e0ecc433bf4e65c00b1a", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:fb9f856daf362787", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 8473cd99c1070625c7bc22988a3a6a8729ea620962b5f39fc1af74df9c7163ea Applicability: Generated full-size joint-neighborhood graphs with partial duplication and feature controls; two CPU threads. Observation: At 30%/60% duplication, accuracy rose 41.2%->85.0% and 61.9%->88.2%; old gate made no kernel calls. Strong-word accuracy stayed 100%. Independent-edge control kept accuracy but cost 32% more time. Interpretation: Linear suitability does not bound nonlinear capacity. The quality-headroom gate covers missed models, with measured extra-cost risk. Native score improvement remains unproven.", + "tags": [ + "writeback-result" + ], + "title": "Can linear-family preselection hide useful quadratic neighborhood predictions?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "de3978f8c58fadba", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T13:00:10+00:00", + "result_id": "question_047f0de9c40a43e1fccf763d4e2bf7f25090bc700917cd4cdb691b0a3a283396", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:8849922124beec6d", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated full-size joint-neighborhood graphs spanning duplicate levels and blend choices; two CPU threads.", + "title": "Does skipping a zero-weight final linear refit preserve predictions and reduce CPU time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7b93d3f7c3ca4a36", + "event_kind": "finding", + "evidence_refs": [ + "experiments/skip_linear_grid.json", + "experiments/validation_skip_linear_path.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_047f0de9c40a43e1fccf763d4e2bf7f25090bc700917cd4cdb691b0a3a283396", + "recorded_at": "2026-10-14T13:00:10+00:00", + "result_id": "finding_48d6ee486c9808fc28cfff6d1794f960b5d8646124729410fb65f634e498f412", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:2c2e7577a7915771", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 21d5d67193f38c2654f3eba27af6607c86607c3f967ff6cd5722b031028b06c2 Applicability: Generated full-size joint-neighborhood graphs spanning duplicate levels and blend choices; two CPU threads. Observation: Five probes had identical predictions/confidences. Pure nonlinear choices removed one refit; partial and word-only retained it. High-duplication median ratio was 0.969; controls varied. Interpretation: The omission preserves predictions here. No broad timing benefit is established; native v7 utility remains unknown.", + "tags": [ + "writeback-result" + ], + "title": "Does skipping a zero-weight final linear refit preserve predictions and reduce CPU time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "92d717faa3dde719", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T13:27:25+00:00", + "result_id": "question_bf6f6e5cce831d26742df437fa494083010bde972bea8b280aadd8c383f7155d", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:56f3ba0a9ed5d03f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated margin controls, 21 graph probes and 20 flat-reliability counterexamples; two CPU threads.", + "title": "How does validation-conditioned margin backoff affect confidence calibration?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "37eb9c268c806b35", + "event_kind": "finding", + "evidence_refs": [ + "experiments/calibration_analysis.json", + "experiments/calibration_counterexamples.json", + "experiments/calibration_order_diagnosis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_bf6f6e5cce831d26742df437fa494083010bde972bea8b280aadd8c383f7155d", + "recorded_at": "2026-10-14T13:27:25+00:00", + "result_id": "finding_74dad17b9740fded3aa30641a5ccd0f6fff69d0b53e98560f894af10c4eadfc2", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:300905fd71ea6f92", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: c5bc1944ea6df76cf20d2660a18948deb2aa7100f9938d7d3e3781b037a1b04a Applicability: Generated margin controls, 21 graph probes and 20 flat-reliability counterexamples; two CPU threads. Observation: Conditional backoff improved 4/21 graphs, preserved 17, and worsened 1/20 flat-reliability controls. Mean graph ECE15 fell 0.0822 to 0.0614. Unconditional anchoring hurt all 20 controls. Interpretation: Retain conditional backoff for native testing. Preserve counterevidence and numerical limits; native utility is unknown.", + "tags": [ + "writeback-result" + ], + "title": "How does validation-conditioned margin backoff affect confidence calibration?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4121eec4072bf156", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T14:04:24+00:00", + "result_id": "question_8764bb1684508c4ec01265a66b7de14ea0e100e2d29a2644809f6a9c65c8a7a2", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:3e07312f648d9175", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native ten-split aggregate comparison of v6 and v9, with combined calibration and refit-skipping changes.", + "title": "Did v9 improve native score over v6?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "bc666bb5de4c98d0", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v9_result.json", + "experiments/class_ambiguity_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_8764bb1684508c4ec01265a66b7de14ea0e100e2d29a2644809f6a9c65c8a7a2", + "recorded_at": "2026-10-14T14:04:24+00:00", + "result_id": "finding_3aef561aa2c0294ff5314e042f97ad1f4eda98524e2f9bdd8ae20520a0e27386", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:c5cdcb9b3a96cc77", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 0e0b57c920c8955d174490ab224cab3783c7f67f83f3f646bcc4243514c15edc Applicability: Native ten-split aggregate comparison of v6 and v9, with combined calibration and refit-skipping changes. Observation: V9 passed 4/4 but scored 0.733380 versus 0.733547. Classification metrics matched; ECE rose 0.03587 to 0.03753. Time penalties were 0.020219 and 0.020217. Interpretation: Restore v6. Local ambiguity correction is scoped evidence; this native comparison did not improve. Branch coverage and timing variance remain unresolved.", + "tags": [ + "writeback-result" + ], + "title": "Did v9 improve native score over v6?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "026f22f103e6cc95", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T14:27:20+00:00", + "result_id": "question_42118afac2527d32d14a9f48de592ac22641721f623f58c318cfa368be4a615f", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:6a3be6ecb49f62ee", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated full-size graphs at two seeds, controlled duplicates and two CPU threads.", + "title": "Can preconditioning or unique-row construction reduce quadratic ridge cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0004758a8eaf90e5", + "event_kind": "finding", + "evidence_refs": [ + "experiments/quadratic_cost_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_42118afac2527d32d14a9f48de592ac22641721f623f58c318cfa368be4a615f", + "recorded_at": "2026-10-14T14:27:20+00:00", + "result_id": "finding_3b85ebfe0b6628149200b34687aadda3f7cef7ef572be1a65def73bb46366383", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:319330c85b437c51", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 9fbcad36a9454f86d8f2a485e69aead8d7b06c52a363901b4c1546ea8147873e Applicability: Generated full-size graphs at two seeds, controlled duplicates and two CPU threads. Observation: PCG cut a refit from 57 to 28 iterations without consistent total gains. Unique construction gave 0.83-0.90 timing ratios on six duplicated probes. Ten graphs preserved predictions; all constructed kernels matched. Interpretation: Test isolated v10 natively. Preserve control timing variability and v6 best; native utility remains unknown.", + "tags": [ + "writeback-result" + ], + "title": "Can preconditioning or unique-row construction reduce quadratic ridge cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "da374c78b2629a98", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T15:03:54+00:00", + "result_id": "question_18859158cb8a55a0217e62a2ce9583e0d274852bd0f1a9cb4a1fcdcb1cfff9b7", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:901488fa80041e27", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native ten-split v6/v10 aggregate comparison of unique-row quadratic-kernel representation.", + "title": "Did v10 improve native score over v6?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "cd038db592eb0815", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v10_result.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_18859158cb8a55a0217e62a2ce9583e0d274852bd0f1a9cb4a1fcdcb1cfff9b7", + "recorded_at": "2026-10-14T15:03:54+00:00", + "result_id": "finding_1a43b5b34a7f81c177563934cbd7f2b906cd6fd1c180693caf9c0875245813ec", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:4e80bc04fefca766", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 9fbcad36a9454f86d8f2a485e69aead8d7b06c52a363901b4c1546ea8147873e Applicability: Native ten-split v6/v10 aggregate comparison of unique-row quadratic-kernel representation. Observation: V10 passed 4/4 over ten splits: score 0.735142 versus 0.733547. Quality aggregates including ECE matched. Mean time was 6.681 versus 7.824 seconds; time penalty 0.018622 versus 0.020217. Interpretation: Promote exact v10. One aggregate per source does not prove nodewise equality or branch causality. Evaluate v11 separately.", + "tags": [ + "writeback-result" + ], + "title": "Did v10 improve native score over v6?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3045ddd25aadf2e0", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T15:39:33+00:00", + "result_id": "question_b0cb6cc00894155c7a03d4d18088b36e86c6257b446f52818cd72ca9d5371df7", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:534578a5d349f86e", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated repeated-neighborhood graphs, independent Gram/final-refit oracles and twelve full-size probes at two seeds.", + "title": "Does duplicate-column compression reduce undirected graph-product cost while preserving the ridge objective?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "03771643fbd63720", + "event_kind": "finding", + "evidence_refs": [ + "experiments/sparse_columns_analysis.json", + "experiments/sparse_columns_cg_diagnosis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_b0cb6cc00894155c7a03d4d18088b36e86c6257b446f52818cd72ca9d5371df7", + "recorded_at": "2026-10-14T15:39:33+00:00", + "result_id": "finding_9bad97ef7134bfb81be3e44d61465cff1aa8783c188bba40fd7dcffa3d762e1e", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:a60d0768f0d12b2b", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 927082086a80e64223b91cced22ec5546dbf107ebd0fefff5ee146cc24358835 Applicability: Generated repeated-neighborhood graphs, independent Gram/final-refit oracles and twelve full-size probes at two seeds. Observation: Gram error below 1.3e-7; final-refit score error below 0.000188. Twelve graph outputs matched. Active timing ratios 0.756-0.964. Two 110-step screening cases exceeded 0.0005; one also failed row-only. Interpretation: Keep v12 for native testing and preserve capped-screening counterevidence. Exact mathematical column equivalence and local timing do not certify native utility.", + "tags": [ + "writeback-result" + ], + "title": "Does duplicate-column compression reduce undirected graph-product cost while preserving the ridge objective?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5ad34ec20cb8e6db", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T16:03:13+00:00", + "result_id": "question_8bdf4e20186cd996d12de3c7ab42e3c695ec2084ac513a69c60fd56cda0d9f5d", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:ec468d9d337d0ffe", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native ten-split aggregate comparison of v10 and combined sparse graph row/column v12.", + "title": "Did combined v12 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "c88aeb04b6ae3a7b", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v12_result.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_8bdf4e20186cd996d12de3c7ab42e3c695ec2084ac513a69c60fd56cda0d9f5d", + "recorded_at": "2026-10-14T16:03:13+00:00", + "result_id": "finding_c9bd0db6d0d227fda5038015c612d8e4bdf1fded08ea4858ef2df20f60c8167c", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:4b095c4df58465f7", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 927082086a80e64223b91cced22ec5546dbf107ebd0fefff5ee146cc24358835 Applicability: Native ten-split aggregate comparison of v10 and combined sparse graph row/column v12. Observation: V12 passed 4/4 but scored 0.732153 versus 0.735142. Mean time fell from 6.681 to 5.654 seconds; quality fell 0.004659. ECE improved, but all classification aggregates fell. Interpretation: Restore exact v10. Local cost and mathematical equivalence did not establish native score utility. One aggregate comparison cannot isolate row from column changes or identify the responsible split.", + "tags": [ + "writeback-result" + ], + "title": "Did combined v12 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "77430c993b125589", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T16:34:39+00:00", + "result_id": "question_ba834979e34884ab5176d91b821e55bc2d00cd0518ba235b34ab59f7cc0503f4", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:20aa4364a851fa30", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Generated symmetric/directed repeated rows, independent oracles, 90 fits and 22 whole graphs.", + "title": "Can forward-only graph row compression preserve v10 arithmetic while reducing CPU cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4c34bf4bed40e5a2", + "event_kind": "finding", + "evidence_refs": [ + "experiments/sparse_forward_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_ba834979e34884ab5176d91b821e55bc2d00cd0518ba235b34ab59f7cc0503f4", + "recorded_at": "2026-10-14T16:34:39+00:00", + "result_id": "finding_58ef563d9b71bcf1af7186449f30dc2f84a0a6da2b19052d5adc11c1a0eec849", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:0c54aefb035a22de", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 74325bc08d4ba8c43cdcca33d8db256284308ca401d473b73c0fa2f60a8ee5b3 Applicability: Generated symmetric/directed repeated rows, independent oracles, 90 fits and 22 whole graphs. Observation: V14 matched v10 in 90 fits and 22 graph outputs. Active component ratios were 0.838-0.921; six of eight whole medians improved. Five original screening oracle failures and two small timing regressions remain. Interpretation: Test exact v14 against native-best v10. Generated parity and cost do not establish native utility or equality across backends.", + "tags": [ + "writeback-result" + ], + "title": "Can forward-only graph row compression preserve v10 arithmetic while reducing CPU cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2f06731cb5f0cb85", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T17:56:07+00:00", + "result_id": "question_f4c6c14e058614350b77fb7e6fb0bc160dc980cd76816af340cceb709e7ba743", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:59c4ab0368f6a0c2", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native ten-split aggregate comparison of v10 and forward-only rows plus Python feature buffers in v15.", + "title": "Did combined v15 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5e74134cf99cdd87", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v15_result.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_f4c6c14e058614350b77fb7e6fb0bc160dc980cd76816af340cceb709e7ba743", + "recorded_at": "2026-10-14T17:56:07+00:00", + "result_id": "finding_b9eeb645d9a533831b48227ee1f00944f9fb39f2d31e860627e581de74b412d3", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:df351c6521c492c1", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: a70ae49d5c4c71c91588c80a90d8e47a92a22e7fe7ecaf1074eca23641963f5a Applicability: Native ten-split aggregate comparison of v10 and forward-only rows plus Python feature buffers in v15. Observation: V15 passed 4/4 but scored 0.734671 versus 0.735142. All quality aggregates matched. Mean time rose from 6.681 to 7.024 seconds; memory rose from 27 to 84 MB with zero resource penalty. Interpretation: Restore exact v10. Local bitwise fidelity did not establish native runtime utility. The combined aggregate does not identify the responsible branch or collector.", + "tags": [ + "writeback-result" + ], + "title": "Did combined v15 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1bb2c64a49dccf80", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T18:13:30+00:00", + "result_id": "question_9b136bdfadf59058258b5aa4e13a7eecf366b1e29ce78c8d2bd815c427c45845", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:3e435541bf39aabc", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Owned generated graphs comparing v10, forward-only v14 and buffered v16, with separate ordinary and allocation-tracked modes.", + "title": "Can row expansion buffer reuse preserve v10 arithmetic and improve complete-solve time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "88617e908f2ce9a4", + "event_kind": "finding", + "evidence_refs": [ + "experiments/row_buffer_analysis.json", + "experiments/row_buffer_allocation_solve.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_9b136bdfadf59058258b5aa4e13a7eecf366b1e29ce78c8d2bd815c427c45845", + "recorded_at": "2026-10-14T18:13:30+00:00", + "result_id": "finding_88604bfd87506b2c3c2e80ff267c3be094700f186c3e96edb16439c388d1e897", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:e3fae3fb095ed8b5", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 0e5fe2cd7dda4685891b231aab5f590dc3494a6fa473247843f78b9053a8042c Applicability: Owned generated graphs comparing v10, forward-only v14 and buffered v16, with separate ordinary and allocation-tracked modes. Observation: V16 matched 90 capped fits and complete outputs. Six of eight active medians improved versus v10; two active and three bypass cases slowed. Four mode comparisons matched complete diagnostics; v16/v14 timing was mixed. Interpretation: Native utility and the collector remain unknown. Keep native-best v10; retain v16 for an isolated native test. Buffer reuse is not a confirmed advantage over v14.", + "tags": [ + "writeback-result" + ], + "title": "Can row expansion buffer reuse preserve v10 arithmetic and improve complete-solve time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8931ad762213dd4d", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T18:43:54+00:00", + "result_id": "question_e2ae08e3027c6c6b5b81304e56b79730972b625390ccc46aa30dd7e7743586ba", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:b369841682f920bc", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native ten-split aggregate comparison of v10 and forward-only rows with expansion buffers in v16.", + "title": "Did v16 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d1f0fcaf89da34f8", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v16_result.json", + "experiments/native_v16_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_e2ae08e3027c6c6b5b81304e56b79730972b625390ccc46aa30dd7e7743586ba", + "recorded_at": "2026-10-14T18:43:54+00:00", + "result_id": "finding_984f3a703da09c6c3195a9d43ebe9e00a11a4833ab10e3270286e0d0cd5dbbec", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:b8da80adf1dc3afb", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 0e5fe2cd7dda4685891b231aab5f590dc3494a6fa473247843f78b9053a8042c Applicability: Native ten-split aggregate comparison of v10 and forward-only rows with expansion buffers in v16. Observation: V16 passed 4/4 but scored 0.733389 versus 0.735142. Quality aggregates matched; mean runtime rose from 6.681 to 7.881 seconds. Memory remained near 27 MB with no resource penalty. Interpretation: Restored exact v10 before selecting the next candidate. Generated parity and local savings did not demonstrate native score utility. Sequential aggregates do not isolate cost causes or prove nodewise equality.", + "tags": [ + "writeback-result" + ], + "title": "Did v16 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "cb606071d76203ac", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T18:59:25+00:00", + "result_id": "question_7e658ef3f52429c001696aae7847bb251aa4993f3e2a3e35ed124b612e7c3325", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:8a6ce038bd9e63d0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Four owned 5201-node controls, seed6137, five alternating complete-solve pairs each, comparing fixed v17 and adaptive v18.", + "title": "Did full-train kernel-order selection improve local time-adjusted quality over fixed cubic?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e064577a550176fe", + "event_kind": "finding", + "evidence_refs": [ + "experiments/adaptive_graph_order_analysis.json", + "experiments/adaptive_graph_order_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_7e658ef3f52429c001696aae7847bb251aa4993f3e2a3e35ed124b612e7c3325", + "recorded_at": "2026-10-14T18:59:25+00:00", + "result_id": "finding_ea3084550f5a33d5708071b37238bae1cb78ce5ac613c251d388805b5d504c8b", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:0a0673bf960c5851", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: c251b40b3febc5302f9a78ac8a38194a38b17a37595b9872b5426b7aa2de0239 Applicability: Four owned 5201-node controls, seed6137, five alternating complete-solve pairs each, comparing fixed v17 and adaptive v18. Observation: V18 median local score deltas ranged from -0.001588 to -0.000133. Two quality gains, one loss and one bypass tie were retained. Numerical and interface checks passed. Interpretation: Retain v18 without selection. Its measured local tradeoff did not justify more fits. Bypass timing variation is not attributed to an unexecuted selector; native utility remains unknown.", + "tags": [ + "writeback-result" + ], + "title": "Did full-train kernel-order selection improve local time-adjusted quality over fixed cubic?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b290717dc1612a01", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T19:33:50+00:00", + "result_id": "question_e4d6b38b32084b01f31ad3471ab20bdd8643abe51a82d3749eddc005f8d7fa1e", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:a3db67ac6a1d5183", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native ten-split aggregate comparison of v10 with cubic graph kernels and canonical node sorting in v17.", + "title": "Did fixed cubic v17 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "358ba7107b7bb7d4", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v17_result.json", + "experiments/native_v17_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_e4d6b38b32084b01f31ad3471ab20bdd8643abe51a82d3749eddc005f8d7fa1e", + "recorded_at": "2026-10-14T19:33:50+00:00", + "result_id": "finding_55c4c90a03982f3fd02be96b9e5b3061876595875b8a918ff107a2740a87b8aa", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:fcf5a8a06a75a7b5", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 4f6e5d918b82fd81b03192aa75853ea5e52c4d585cc44ebab9085999a26d9f99 Applicability: Native ten-split aggregate comparison of v10 with cubic graph kernels and canonical node sorting in v17. Observation: V17 passed 4/4 but scored 0.727049 versus 0.735142. Quality fell 0.006225; mean time rose from 6.681 to 7.890 seconds. Memory stayed near 27 MB with zero resource penalty. Interpretation: Restore v10. Generated interaction gains did not establish native utility. Aggregate feedback cannot isolate cubic order from row sorting or attribute individual splits.", + "tags": [ + "writeback-result" + ], + "title": "Did fixed cubic v17 improve native score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "46d9c295209cb135", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T19:46:59+00:00", + "result_id": "question_03cedd45d80752e5c83d88001fad58d5fcbdd3898a0017653346cad8a59c1257", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:79db0c3c7e9ab5a3", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Six owned 5201-node graph controls, seeds6137/8119, plus generated small and larger signal probes; compare v20 with full-train v18.", + "title": "Does matching the regularization and blend grid repair 512-node screen fidelity?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f434c975c0d05537", + "event_kind": "finding", + "evidence_refs": [ + "experiments/grid_graph_screen_analysis.json", + "experiments/screen_twin_fidelity.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_03cedd45d80752e5c83d88001fad58d5fcbdd3898a0017653346cad8a59c1257", + "recorded_at": "2026-10-14T19:46:59+00:00", + "result_id": "finding_1305e8cf3315b619994aed87d3d0edd2bf31113598df9631c38d29ba4bb85bfc", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:a74eebf838d38cae", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: dcbe9415cff71d840a5f38f25834355bdcd80486e2d474587c7aa2945ce4f22b Applicability: Six owned 5201-node graph controls, seeds6137/8119, plus generated small and larger signal probes; compare v20 with full-train v18. Observation: V20 passed sampling/interface checks and recovered one small-case loss, but chose3 versus full18 degree2 on all six full-size controls. Three active local score deltas versus v10 were positive and three negative. Interpretation: Retain20. Isolated v21 all-train twin counts also repaired none. The tested parameter-grid and twin changes did not fix target-subsampling fidelity; native utility remains unknown.", + "tags": [ + "writeback-result" + ], + "title": "Does matching the regularization and blend grid repair 512-node screen fidelity?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d82342f73cde6dd1", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T20:10:19+00:00", + "result_id": "question_ef65fa1830ed5bef6e3a886e24869e20c81e57a3e4a419c87892bdc9fa8dccfb", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:dd8ffc9425b92194", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Six active owned5201-node graph controls across seeds6137/8119 and one word bypass; five alternating uninstrumented pairs per control.", + "title": "Did covariance reformulation make all-target basis screening locally worthwhile?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0b255e938483bad8", + "event_kind": "finding", + "evidence_refs": [ + "experiments/covariance_basis_screen_analysis.json", + "experiments/covariance_basis_screen_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_ef65fa1830ed5bef6e3a886e24869e20c81e57a3e4a419c87892bdc9fa8dccfb", + "recorded_at": "2026-10-14T20:10:19+00:00", + "result_id": "finding_e3ba65bcfbea0f15ee8c59bfbd773294dc499098872f2d3bb2649d01a4c81311", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:a84e1b808c69a084", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 4d0b42ff9278d6591a9c7bf165b32c27657bf18970ad747bbfa2c3de3be46f72 Applicability: Six active owned5201-node graph controls across seeds6137/8119 and one word bypass; five alternating uninstrumented pairs per control. Observation: Covariance23 passed72 references and seven output equalities with22. Six active medians were0.95–1.00x22 but1.08–1.20xv10; all six local score deltas versus10 were negative. Five of six full-train order choices matched. Interpretation: Keep v10. Tested fidelity gains did not repay screening cost. Bypass quality tied; its timing is not credited to an unexecuted screen. Native23 utility remains unknown. Try a single cubic challenger without basis decomposition.", + "tags": [ + "writeback-result" + ], + "title": "Did covariance reformulation make all-target basis screening locally worthwhile?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "29f02f6b6a86f483", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T20:33:33+00:00", + "result_id": "question_106de3aab5a108eec0643015165a57710459c6e6e80fcd4b36a7eeaef48046f3", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:301a5d439dcb0101", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native ten-split public aggregate comparison; v26 uses original squared precision and one full-target fixed-.02 cubic challenger with residual-checked large cubic fits.", + "title": "Did the full-target one-cubic challenger v26 improve native final_score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "51bab38649353ff8", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v26_result.json", + "experiments/native_v26_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_106de3aab5a108eec0643015165a57710459c6e6e80fcd4b36a7eeaef48046f3", + "recorded_at": "2026-10-14T20:33:33+00:00", + "result_id": "finding_dc4b868a4e98b3d08a0027a52c65476945b9633f2de9caa0ee1bd06163b86b7f", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:69c13b1c480a9078", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: ef6e8d6889e1d9cb19eb357d50df0bb2bcd7b2415a9ec998ea6e43349e700c00 Applicability: Native ten-split public aggregate comparison; v26 uses original squared precision and one full-target fixed-.02 cubic challenger with residual-checked large cubic fits. Observation: V26 passed4/4. Quality rose0.001524 and class metrics rose about0.0018, but time rose6.68to9.59s; final_score fell0.002247 to0.732895. Memory stayed near27MB with no resource penalty. Interpretation: Restore10. Adaptive selection improved native classification quality but its observed runtime did not repay that gain. Retain26 as a quality candidate for cost investigation. Sequential aggregates cannot identify split choices, isolate precision/canonicalization/selection causes or establish a native hotspot. All local timing and negative evidence remain separate.", + "tags": [ + "writeback-result" + ], + "title": "Did the full-target one-cubic challenger v26 improve native final_score over v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2704bab1d0fabc6b", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T21:09:15+00:00", + "result_id": "question_23a52d7e116f624a5ce34a03defcf469f3615bab4562dff09f23a6366cf65733", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:dd59de1d19042a84", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Two owned5010-node cubic-selected graphs, six owned5201-node squared controls and one word bypass; three alternating uninstrumented pairs per case, separate counters/profile.", + "title": "Did a64-step cubic CG cap preserve tested fast convergences while shortening two owned large cubic-selected solves?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fbe8b4accde57400", + "event_kind": "finding", + "evidence_refs": [ + "experiments/cubic_fit_diagnosis.json", + "experiments/short_cubic_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_23a52d7e116f624a5ce34a03defcf469f3615bab4562dff09f23a6366cf65733", + "recorded_at": "2026-10-14T21:09:15+00:00", + "result_id": "finding_0e14026f09034a945a1289502ca43e01d1c2d5fab34b0aa438234ae24fd1aa5a", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:badee61701c2d0fc", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: a4b46c7d97e5c292f0f93fc1f6c22fcb9410572aa3b6abba3d0eb7c39b5bdd53 Applicability: Two owned5010-node cubic-selected graphs, six owned5201-node squared controls and one word bypass; three alternating uninstrumented pairs per case, separate counters/profile. Observation: V28 passed10 interface,4 path and8 numeric checks. All9 outputs equal26; fast CG counts stay32–46 and slow fits fall160to64 before fallback. Two large solve ratios28/26=.8119/.7869. Interpretation: Conditional early fallback reduced owned slow-case cost while retaining tested fast convergence. Keep native-best10 separately;28 is staged for native verification. Six unchanged-path ratios.9511–1.0142 and bypass.9675 are timing variation, not credited gains. Direct27 saved more on slow cases but replaced convergent CG unnecessarily. Generated zero-word directed controls, Python/C-call profiling and three timing pairs cannot explain the native26 regression or certify native score, memory, p95 or all-input equivalence.", + "tags": [ + "writeback-result" + ], + "title": "Did a64-step cubic CG cap preserve tested fast convergences while shortening two owned large cubic-selected solves?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f63bbe22d9f6bc8c", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T21:26:15+00:00", + "result_id": "question_dd8bee50fcc6572f758074bcb06d00fcddfd417022d22248f04beaab8a971504", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:4d5b9a5eb917af4b", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Nine owned complete-graph controls with three alternating timing pairs; six balanced follow-up pairs on two5010-node cubic-selected graphs, eight counted paths and six independent760-node PSD kernels.", + "title": "Does reducing the large stable cubic CG cap from64to48 provide enough marginal benefit to replace the staged64-step candidate?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b37eab266d312e5d", + "event_kind": "finding", + "evidence_refs": [ + "experiments/cg48_cubic_analysis.json", + "experiments/cg48_marginal_cost.json", + "experiments/validation_cg48_crossover.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_dd8bee50fcc6572f758074bcb06d00fcddfd417022d22248f04beaab8a971504", + "recorded_at": "2026-10-14T21:26:15+00:00", + "result_id": "finding_e52449937c4fba88e4c11f5430b20fbe8b9d0077a5247064d782b9b65fa93517", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:3cc03aafc0e1f387", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 7f9c779df03e0f99728eb9cf32663e5ab1b1d2765183ea320f3e5a15f039d15e Applicability: Nine owned complete-graph controls with three alternating timing pairs; six balanced follow-up pairs on two5010-node cubic-selected graphs, eight counted paths and six independent760-node PSD kernels. Observation: Nine outputs equal28; fast counts stay32-46. Follow-up ratios29/28=.9823/.9331 with3/6 and6/6 wins. Unchanged bypass moved7%. A49-step reference convergence becomes a48-step direct fallback; predictions match. Interpretation: Retain29 unsubmitted and keep28 staged. Observed marginal gains favor48 on one owned slow case but are mixed on the other, while the boundary probe proves that earlier direct fallback can add work for inputs converging between caps. Independent errors remained at most1.03e-6, with direct fallback near8e-15 on the49-step case. Reused18 unchanged-path checks via byte/path proof;4 large invariances,8 counters and6 PSD cases executed. This does not establish native equivalence or a universal performance win;29 remains a viable alternative. Native-best10 is preserved; current history has11 completed accepted submissions. Next evaluate exact28 after eligibility.", + "tags": [ + "writeback-result" + ], + "title": "Does reducing the large stable cubic CG cap from64to48 provide enough marginal benefit to replace the staged64-step candidate?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e77917c809095676", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T21:43:06+00:00", + "result_id": "question_4eef36ff960defc759e63c20d5c7a93dc868ce618b31460e592ebb2a6055a51e", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:0ed36928272f229f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native public full-ten-split aggregate comparison: exact submitted28 versus best10, with26 as the quality/time reference. Local32 results are separate.", + "title": "Did the64-step cubic cap v28 improve native final_score over native-best v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9444ece35221f64f", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v28_result.json", + "experiments/native_v28_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_4eef36ff960defc759e63c20d5c7a93dc868ce618b31460e592ebb2a6055a51e", + "recorded_at": "2026-10-14T21:43:06+00:00", + "result_id": "finding_d98fc9c313b33c7f81d5ed137c57e1c6173b9a69d97524c28b81682654feb00f", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:5c8ed971a6835dc3", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: a4b46c7d97e5c292f0f93fc1f6c22fcb9410572aa3b6abba3d0eb7c39b5bdd53 Applicability: Native public full-ten-split aggregate comparison: exact submitted28 versus best10, with26 as the quality/time reference. Local32 results are separate. Observation: V28 passed4/4 at0.73466655 versus10 at0.73514207. Quality metrics exactly match26; average time fell9.59to8.20s and score rose0.00177158 over26. Gap to10 is0.00047553; resource penalty stayedzero. Interpretation: Restore exact10 after verifying current/submitted source identity; retain28. Native classification quality improvement was preserved with lower observed time, but final_score still missed the incumbent. This aggregate comparison does not identify split choices or a native hotspot, nor prove that all timing change is caused by the loop cap. Owned combined32 subsequently passed existing references and invariances, preserving9 generated outputs with mixed timing evidence; it is staged unsubmitted with native utility unknown. Next test checked fallback factorization and evaluate the strongest exact candidate after22:29:13UTC.", + "tags": [ + "writeback-result" + ], + "title": "Did the64-step cubic cap v28 improve native final_score over native-best v10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f05465220d77cf5c", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T21:57:20+00:00", + "result_id": "question_4233aadb12ab9073e43ae423e6dd3f0bbe3986eb1d133a4fc28072f85c58c24c", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:b307b29a62fef1a2", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Two owned5010-node cubic graphs, six5201-node fast controls and a word bypass; three timing pairs, six balanced seed89 pairs, independent full/compact and asymmetric references.", + "title": "Did residual-checked Cholesky reduce owned complete-solve cubic fallback cost while preserving existing numerical and invariance guards?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1bc71ca13435d32b", + "event_kind": "finding", + "evidence_refs": [ + "experiments/checked_cholesky_analysis.json", + "experiments/checked_cholesky_cost.json", + "experiments/checked_cholesky_marginal_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_4233aadb12ab9073e43ae423e6dd3f0bbe3986eb1d133a4fc28072f85c58c24c", + "recorded_at": "2026-10-14T21:57:20+00:00", + "result_id": "finding_130a6d9a9fae200b91c7196ce514b6014426281347ca869a75b6cc6ad6a49005", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:7330064c42df5c26", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 3a9feb23b8919f8ac6802854f0568424c55a1b285b369498e4172016666dcebf Applicability: Two owned5010-node cubic graphs, six5201-node fast controls and a word bypass; three timing pairs, six balanced seed89 pairs, independent full/compact and asymmetric references. Observation: V33 passed14 references and8 invariances. Nine predictions/quality values match32; max confidence delta5.63e-13. Slow ratios33/32=.9769/.8796; seed89 six-pair follow-up=.9588 with6/6 wins. Four large fits use Cholesky without LU. Interpretation: Stage33 unsubmitted; preserve native-best10 and32. Checked factorization, tight true-residual acceptance and two corrections retain LU when Cholesky fails or residuals remain high. Independent LU/analytic references and asymmetric probes exercise success/refinement/rejection. Initial fixture wrongly assumed1e-7 asymmetry would finish in two corrections; that case and failure log remain, and1e-9 was added for successful-refinement coverage without changing numerical guards. Six fast paths and bypass never execute the new factorization; their timing movement is variation, not credited gains. Local inputs and these small samples cannot certify native equivalence, native runtime cause, memory or p95. Next compare direct-first checked Cholesky before the next eligible native test.", + "tags": [ + "writeback-result" + ], + "title": "Did residual-checked Cholesky reduce owned complete-solve cubic fallback cost while preserving existing numerical and invariance guards?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5ade724357f15153", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T22:11:17+00:00", + "result_id": "question_dd341dcf280eca74312d1ae31982163c7af50a7054563c94cd99c7fd012e1c3d", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:4eaa709f57da64e7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Two owned5010-node cubic graphs, six5201-node fast controls and a word bypass; three timing pairs plus six balanced pairs on two fast controls, independent references and invariances.", + "title": "Does direct-first checked Cholesky improve owned cubic solve cost without a sustained fast-control penalty versus the staged CG-first solver?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7a539c4ea45a175d", + "event_kind": "finding", + "evidence_refs": [ + "experiments/direct_checked_cholesky_analysis.json", + "experiments/direct_checked_cholesky_cost.json", + "experiments/direct_checked_cholesky_fast_followup.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_dd341dcf280eca74312d1ae31982163c7af50a7054563c94cd99c7fd012e1c3d", + "recorded_at": "2026-10-14T22:11:17+00:00", + "result_id": "finding_4eefe20d55868881912e6ad2c4edaeb9fa24277f9d0e8ed3b3c86f38f2d8fa60", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:c90115f00fdb8c37", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 715b14fc38e2b381f38d3d92c6f077054bbb748db30065c7d56f88716c7a506a Applicability: Two owned5010-node cubic graphs, six5201-node fast controls and a word bypass; three timing pairs plus six balanced pairs on two fast controls, independent references and invariances. Observation: V34 passed14 references and8 invariances. Nine predictions/quality values match33/32; max confidence delta5.63e-13. Slow ratios34/33=.8772/.8210. Two fast follow-ups=.9632(6/6 wins) and.9901(4/6 wins); no LU on actual graph fits. Interpretation: Stage34 unsubmitted and preserve native-best10 plus33/32. Only high-precision cubic CG is skipped; checked Cholesky, tight residual acceptance, two corrections and LU recovery remain. Six fast controls now factor once and some compact cases refine, while ordinary precision and bypass work are unchanged. Initial fast ratios.9447-1.0272 and bypass1.0046 are retained counterevidence. Follow-ups use exactly matching input hashes, balanced order and excluded warmups; the second still has two losing pairs. Independent full-expanded LU and analytic/asymmetric probes preserve numerical bounds and exercise recovery. Local data and these small samples cannot certify native equivalence, global speedup, memory or p95. Next test validation-row-only projections while respecting the native cooldown.", + "tags": [ + "writeback-result" + ], + "title": "Does direct-first checked Cholesky improve owned cubic solve cost without a sustained fast-control penalty versus the staged CG-first solver?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "bf3ea0d77c764c8c", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T22:21:53+00:00", + "result_id": "question_bfecef8a9e8ec578b1e8560b32a8bcc75020a367bfc7108a4c44ff71d1092fc4", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:565407be697d17c4", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Two owned5010-node cubic graphs, six5201-node squared controls and a word bypass; three timing pairs, six balanced pairs on three controls, full/compact projection references and invariances.", + "title": "Does restricting kernel screening projections to validation rows preserve model outputs while improving owned whole-solve time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9b4cb3bbd0470d69", + "event_kind": "finding", + "evidence_refs": [ + "experiments/validation_projection_analysis.json", + "experiments/validation_projection_cost.json", + "experiments/validation_projection_followup.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_bfecef8a9e8ec578b1e8560b32a8bcc75020a367bfc7108a4c44ff71d1092fc4", + "recorded_at": "2026-10-14T22:21:53+00:00", + "result_id": "finding_3f63d9b2fffc7727fa33080994e11f322be0b709a66d35b1185a71bd5417d170", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:925945776cee2ca8", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 3ca70479b4e7485f47bb858906c20041913f0954d0b0827ebd7500800cde66cb Applicability: Two owned5010-node cubic graphs, six5201-node squared controls and a word bypass; three timing pairs, six balanced pairs on three controls, full/compact projection references and invariances. Observation: V35 passes10 interface,74 numeric and8 invariance checks; all9 complete outputs and quality values equal34. Balanced full-row ratio=.9581(6/6 wins), compact=.9985(4/6), cubic=.9878(3/6). Maximum partial/full projection error1.73e-6. Interpretation: Stage35 unsubmitted; retain34 and native-best10. All fitting targets and solver coefficients are unchanged; only validation predictions are projected in screening, with compact requested groups deduplicated. Final full projections remain original. Empty/repeated/reversed requests, full/compact cubic LU references and actual graph fitting boundaries pass unchanged limits. Full-row benefit repeats in six balanced pairs, but the cubic result is uncertain and compact essentially tied. Initial compact slowdowns and losing follow-up pairs remain counterevidence. Timed solves were uninstrumented; attribution ran separately. Generated inputs cannot certify native equivalence, broad speedup, memory or p95. Native evaluation remains required after the cooldown.", + "tags": [ + "writeback-result" + ], + "title": "Does restricting kernel screening projections to validation rows preserve model outputs while improving owned whole-solve time?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "901a31b34c95221a", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T22:42:17+00:00", + "result_id": "question_7f2781f61a8d9d5a408ea010bfaf565af76ce25a2fd8a3b6c5ed03e28b375a2b", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:91ef846f49b6d01f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Native public full-ten-split aggregate comparison of exact35 with10 and28. Local36 is a separate retained experiment; its results do not enter native35 metrics.", + "title": "Did exact35 improve native final_score over best10 after its locally validated cubic fitting and projection changes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "758d437a10983c6a", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v35_result.json", + "experiments/native_v35_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_7f2781f61a8d9d5a408ea010bfaf565af76ce25a2fd8a3b6c5ed03e28b375a2b", + "recorded_at": "2026-10-14T22:42:17+00:00", + "result_id": "finding_94ea0e0a35790029c6f2ac8871a9735c450e581563445b3f5ee0da05ff5b72a4", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:e48b4f877195f910", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 3ca70479b4e7485f47bb858906c20041913f0954d0b0827ebd7500800cde66cb Applicability: Native public full-ten-split aggregate comparison of exact35 with10 and28. Local36 is a separate retained experiment; its results do not enter native35 metrics. Observation: V35 passed4/4 at.73443058 versus10 at.73514207 and28 at.73466655. Quality equals28; mean time8.180vs8.196s fell slightly but mean log-time penalty.020858vs.020622 rose. Resource penalty stayedzero. Free readback matches. Interpretation: Restore exact10 only after matching current/submitted source, acceptedID b9de1f562180, round agent-13 and public full10 result. Keep35 source and scoped local cost evidence. Class metrics match28 exactly, while ECE differs by one rounding unit. The arithmetic mean runtime is not the selection statistic: final_score averages per-split log penalties, and its geometric time ratio rose1.02388 over28. Sequential aggregates cannot identify a native stage, affected split, nodewise equivalence or timing cause. Local fitting/projection gains therefore do not certify native utility. The subsequent36 operator trial is retained separately because it changed selector choices despite passing independent numeric checks; its partial timing summary lacks raw pairs and supports no speed claim.", + "tags": [ + "writeback-result" + ], + "title": "Did exact35 improve native final_score over best10 after its locally validated cubic fitting and projection changes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "6924e0e4de7cbc63", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T22:53:29+00:00", + "result_id": "question_ce7c3a77316f5486bf47a34c98b330a5b6cdc21461fdc6f99754e8312926688a", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:a77b797726310751", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Owned generated CPU workloads against10: full original transpose and dense products,110/160 CG caps,train-only selection and train/valid refit. Native utility untested.", + "title": "Can restricting only sparse forward training rows preserve capped fits and model selection while reducing owned whole-solve cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2136de6048c321e5", + "event_kind": "finding", + "evidence_refs": [ + "experiments/validation_sparse_forward_summary.json", + "experiments/sparse_forward_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_ce7c3a77316f5486bf47a34c98b330a5b6cdc21461fdc6f99754e8312926688a", + "recorded_at": "2026-10-14T22:53:29+00:00", + "result_id": "finding_a02665d463fd2df09fad89fbec100b2866296830b2bf05abf4bce72e14ca418f", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:e5a516402f405e88", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 0e9057705b3aa3e5c1f24476bb7d0967f1e66cb4c80f25d2d028c53b02430e05 Applicability: Owned generated CPU workloads against10: full original transpose and dense products,110/160 CG caps,train-only selection and train/valid refit. Native utility untested. Observation: Nine operators and54 capped fits equal10 exactly;54 independent LU references and20 interface/graph checks pass. Ten5201-node solves preserve full outputs/diagnostics. Forty balanced pairs:37 faster,geometric ratio0.9124; case medians0.838to0.970. Interpretation: Stage exact37 unsubmitted and retain native-best10. Compared with36, preserving both original transpose and dense arithmetic restores the formerly failing6137/.6 selection; this joint intervention does not identify a sole reduction cause. Four sparse forward gathers and three full dense passthrough preparations are observed per solve, with every original transpose retained and no test targets fitted. Numeric limits are unchanged. Ordinary timings include one excluded warmup per model and four balanced alternating pairs per case, persisted after each pair; constructor counters run separately. Local gains do not certify native benefit, memory,tail latency or a native hotspot. Earlier16/35 native regressions and36 selector failure remain relevant counterevidence. Reconcile history and exact source before native evaluation after23:29:45UTC.", + "tags": [ + "writeback-result" + ], + "title": "Can restricting only sparse forward training rows preserve capped fits and model selection while reducing owned whole-solve cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4304b21f1bd550b8", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T23:07:27+00:00", + "result_id": "question_cacf2fa19356a1682037d28d8c20f76ac8d1d7fced50fbef3cf7b0c5c20c02e1", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:dc67aea553c7f8b6", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Ten generated5201-node CPU workloads;38 versus37 and10 with full original transpose/dense products and unchanged CG budgets. Operator target order includes unique-sorted and fallback inputs.", + "title": "Does eliminating inverse sparse-row remapping improve owned whole-solve time beyond candidate37?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "db814e0bd346a08f", + "event_kind": "finding", + "evidence_refs": [ + "experiments/validation_original_order_summary.json", + "experiments/original_order_cost.json", + "experiments/original_order_fixture_correction.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_cacf2fa19356a1682037d28d8c20f76ac8d1d7fced50fbef3cf7b0c5c20c02e1", + "recorded_at": "2026-10-14T23:07:27+00:00", + "result_id": "finding_4eb0a26573b93c20e80e90e992f5c8afa0e0f8a9ee71b4231b54b8c61cfae386", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:c70a0ac8d7bd326d", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 382109a5ebb6a3f07d9171e429f4a9aac4bea8fe32dcaabfa47789319dd8ed55 Applicability: Ten generated5201-node CPU workloads;38 versus37 and10 with full original transpose/dense products and unchanged CG budgets. Operator target order includes unique-sorted and fallback inputs. Observation: 38 passes54 exact capped fits and independent LU bounds,20 interface/graph checks and ten exact whole solves.38/37 geometric ratio0.9969,28/60 wins,five faster case medians.37/10 ratio0.9144,54/60 wins. No native evaluation. Interpretation: Retain38 as uncertain incremental cost evidence and keep37 staged. Six three-model permutations per case balance every execution position, with one excluded warmup/model and each raw round saved immediately. Constructor observations run separately and verify original transpose/dense identities and fit boundaries. The initial counter wrongly expected raw train-list order; all three solvers apply unique() before normal screening. Correcting only that expectation left candidate source and every numeric/output gate unchanged. Preserve the failed harness/log and six initial rounds outside final aggregates. Arbitrary unsorted operator targets remain covered by independent capped-fit checks. Mixed local costs do not identify a timing cause or native utility. Explore shared shifted-Krylov screening for the three ridge values using independent full-target/residual references and explicit selector comparisons; retain37 for source-matched evaluation after23:29:45UTC.", + "tags": [ + "writeback-result" + ], + "title": "Does eliminating inverse sparse-row remapping improve owned whole-solve time beyond candidate37?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7a2982185e53fc55", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T23:19:44+00:00", + "result_id": "question_f199a7dbf22f189cf51c5348190a9467c5b2bf0ae750127dc22d1c82e1bd576e", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:0daeacb3e60944fe", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Owned CPU references and ten generated5201-node solves against37. Three screening ridge values share products; final single-ridge refit and nonlinear fits remain unchanged. Native utility unknown.", + "title": "Can a shared shifted-CG trajectory reduce owned ridge-screening cost while preserving independent fit accuracy and classifier behavior?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e1e179a2f0bfcede", + "event_kind": "finding", + "evidence_refs": [ + "experiments/validation_shifted_ridge_summary.json", + "experiments/validation_shifted_ridge_math.json", + "experiments/shifted_ridge_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_f199a7dbf22f189cf51c5348190a9467c5b2bf0ae750127dc22d1c82e1bd576e", + "recorded_at": "2026-10-14T23:19:44+00:00", + "result_id": "finding_5de2e39b4528373cf1cec7ee60efd9b35b8731f0f82dc6d3bdd5d0cce63c55db", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:120bb55e5bea5c87", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 88b0d95e8506cff43ffa6d0f6902f2bb4d2ae5cf02983fc63bc982f3f1662f31 Applicability: Owned CPU references and ten generated5201-node solves against37. Three screening ridge values share products; final single-ridge refit and nonlinear fits remain unchanged. Native utility unknown. Observation: 48 coefficient/residual and54 full-target LU references pass;max true residual8.94e-5 and fit error2.03e-4.20 interface/graph checks pass. Ten solves preserve choices,predictions and quality;max confidence delta1.2e-7.39/40 paired wins;geometric39/37 ratio0.8582. Interpretation: Stage exact39 and retain37 plus native-best10. Smallest-ridge capped trajectory remains bitwise equal; positive shifts differ numerically within unchanged independent bounds. Nine duplicate-target fallbacks remain exact. Separate observations verify four train-only shared screening paths and one original160-step train/valid final refit,excluding test targets. Full diagnostic tables match in seven of ten cases,while selected settings and predicted classes match in all ten. Four balanced alternating ordinary pairs per case exclude warmups and instrumentation; every raw pair persists. Whole-algorithm savings do not identify a native hotspot or certify native quality,memory or tail latency. Retain16/35 native regressions and36/38 counterevidence. The small full-target references have at most97 feature columns; higher-rank owned systems remain a useful coverage probe while native cooldown lasts. Reconcile current history and exact source before evaluation after23:29:45UTC.", + "tags": [ + "writeback-result" + ], + "title": "Can a shared shifted-CG trajectory reduce owned ridge-screening cost while preserving independent fit accuracy and classifier behavior?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fa260266f5bfa04e", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T23:28:53+00:00", + "result_id": "question_efa961907f212e1084214a3007db34359ba1ba92061a352b92882a717d26ad5d", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:612f564441203647", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Two seeds;960 nodes,720 targets,wide sparse and controlled spectral systems at110/160;local320/640 followups. Thresholds and candidate39 source/budgets remain unchanged.", + "title": "Do39 and37 retain independent fit accuracy on higher-rank owned ridge systems at existing iteration caps?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e88119dc51ece799", + "event_kind": "finding", + "evidence_refs": [ + "experiments/shifted_high_rank_summary.json", + "experiments/validation_shifted_high_rank.json", + "experiments/validation_shifted_high_rank_caps.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_efa961907f212e1084214a3007db34359ba1ba92061a352b92882a717d26ad5d", + "recorded_at": "2026-10-14T23:28:53+00:00", + "result_id": "finding_cdf20eaee330b345fd953ebe95ea0bac678b709f2a8b1bfec3e62c7285ce0247", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:8f47398d451b48b8", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 88b0d95e8506cff43ffa6d0f6902f2bb4d2ae5cf02983fc63bc982f3f1662f31 Applicability: Two seeds;960 nodes,720 targets,wide sparse and controlled spectral systems at110/160;local320/640 followups. Thresholds and candidate39 source/budgets remain unchanged. Observation: 36 comparisons:32 both-pass,four shared low-ridge failures,no39-only failure. Local640 allowances stop at320/318;score errors fall below5e-4,but seed89 residual2.0837e-4 still fails2e-4.39 remains source-identical and unsubmitted. Interpretation: Keep exact39 ready and retain every failure. Broad-spectrum low-ridge cases expose inherited capped accuracy limits:source37 and39 share exact base scores at all tested budgets. Larger local allowances improve the scores but do not make every residual pass. Matching320/640 stops distinguish a remaining stop/finite-precision limitation from simple cap exhaustion;the sole rounding cause and native relevance remain unproved. Original110 observations match the repeated inputs exactly. A metadata sanity check was corrected from steps<320 to steps<=320 with640 stopping early;no numerical criterion changed. Preserve the earlier small-reference and whole-solve evidence with this wider scope. Prepared immutable source and a one-shot wrapper record actual acceptance identity before awaiting feedback. Current history has13 completed submissions;no new evaluation occurred. Native utility remains unknown until source-matched full-ten-split feedback after23:29:45UTC.", + "tags": [ + "writeback-result" + ], + "title": "Do39 and37 retain independent fit accuracy on higher-rank owned ridge systems at existing iteration caps?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5298c48b793b2c40", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T23:36:26+00:00", + "result_id": "question_3f8eefa28372120d0803913e1fd4b67fa6f52d2fd514adcfa84c8834b6d0124e", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:bb144c81e158a717", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source-matched public full-ten-split comparison of39 and10. No native37 result;local residual diagnostics are separate evidence.", + "title": "Did exact39 improve native final_score over native-best10 after sparse training-row forward products and shared shifted screening?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9147419e99ce83bf", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v39_result.json", + "experiments/native_v39_analysis.json", + "experiments/shifted_residual_diagnosis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_3f8eefa28372120d0803913e1fd4b67fa6f52d2fd514adcfa84c8834b6d0124e", + "recorded_at": "2026-10-14T23:36:26+00:00", + "result_id": "finding_c223536af24c96f074aaaf3a32c17fccb5d5fe9f3fffd7b5afd1cde0b2a02b33", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:1c025e6de0371443", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 88b0d95e8506cff43ffa6d0f6902f2bb4d2ae5cf02983fc63bc982f3f1662f31 Applicability: Source-matched public full-ten-split comparison of39 and10. No native37 result;local residual diagnostics are separate evidence. Observation: 39 passed4/4 at0.733212 versus10 at0.735142. Quality fell0.004509;mean time fell6.68to5.09s. Resource penalty stayedzero. ActualID6faaad36be39/agent-14 and free readback match the immutable source. Interpretation: Restore source-verified10 and retain39. The tested combined implementation did not beat the incumbent. Native aggregates cannot distinguish the row-forward and shifted-screening changes, identify selected models or prove per-node equivalence;37 has not been evaluated natively. Local prototypes preserved selections but did not generalize to native quality. Higher-rank base-cap failures are shared and do not establish the native cause. New local640 diagnostics reproduce prior coefficients and stops:tracked residuals9.18e-5/8.75e-5,recomputed float2.136e-4/2.008e-4,double reference2.084e-4/1.907e-4. Both recursive and operator-precision discrepancies exist on those fixtures. Build higher-rank owned graph selection probes for10/37/39 before stacking restarts;keep production budgets and numeric limits unchanged.", + "tags": [ + "writeback-result" + ], + "title": "Did exact39 improve native final_score over native-best10 after sparse training-row forward products and shared shifted screening?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8f78dbed75c03935", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T23:47:13+00:00", + "result_id": "question_02f13c30e52f4ac90f969eb0b23e59c72f1aa53a75f55314f502dbc00262ff58", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:2861c80145956b7d", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Two seeds at2010/5201nodes,unique neighborhoods,2089features,512independent sampled adjacency rows;plus one noisy-label5201case. Three exact sources10/37/39.", + "title": "Do higher-rank owned graph selector probes distinguish the sparse-row and shifted-screening changes behind native39 quality loss?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ceb59de5e1b64e39", + "event_kind": "finding", + "evidence_refs": [ + "experiments/high_rank_selection_analysis.json", + "experiments/high_rank_selection.json", + "experiments/high_rank_noisy_selection.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_02f13c30e52f4ac90f969eb0b23e59c72f1aa53a75f55314f502dbc00262ff58", + "recorded_at": "2026-10-14T23:47:13+00:00", + "result_id": "finding_b16c45965ea2d928d89c3c40d8951537baef0d2c7c5ad0d26021ce73121e1c2d", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:f42e2e6cd9122df6", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 88b0d95e8506cff43ffa6d0f6902f2bb4d2ae5cf02983fc63bc982f3f1662f31 Applicability: Two seeds at2010/5201nodes,unique neighborhoods,2089features,512independent sampled adjacency rows;plus one noisy-label5201case. Three exact sources10/37/39. Observation: Five cases/15instrumented solves:37matches10 in65fits and all outputs.39positive-shift scores differ up to4.3753e-4;three noisy word-only screen qualities change. All selected settings and test predictions still match;max confidence delta1.55e-6. Interpretation: These probes extend coverage but do not identify the native cause. The first four cases saturated nearperfect classification;one separate30percent random-relabeling condition used the remaining three-solve allocation and exercised nonlinear screening. Rank512 is an independent sampled-row lower bound,not exact full rank. Smallest-ridge and final-refit scores remain exact;positive shifted trajectories differ. The noisy word-only ridge2quality changes do not win selection,so no changed-choice replay is triggered. Preserve every trace and masked-label proof;do not turn local parity into native equivalence. Native37 remains untested and39combines both mechanisms. Prioritize an isolated source-bound37evaluation after eligibility over unchanged parity repetitions or immediate residual restarts. Current solution stays verified native-best10.", + "tags": [ + "writeback-result" + ], + "title": "Do higher-rank owned graph selector probes distinguish the sparse-row and shifted-screening changes behind native39 quality loss?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "bdabca7ac6635fb2", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-14T23:55:38+00:00", + "result_id": "question_0349fc5778ed4052044fb43e81436a7c3b596d21969fcddde047e4a927f4eccd", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:77dd2e3090bb5d8e", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact40versus10;ten directed single/pair/triple structural cases,two sizes/two seeds. Independent full/compact/weighted-duplicate mixed-label oracle checks and selected-path invariances.", + "title": "Does training-centered quadratic ridge with an unpenalized intercept improve owned structural classification over10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "5eeaeeda4a252226", + "event_kind": "finding", + "evidence_refs": [ + "experiments/centered_kernel_analysis.json", + "experiments/validation_centered_kernel_math.json", + "experiments/centered_kernel_graph.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_0349fc5778ed4052044fb43e81436a7c3b596d21969fcddde047e4a927f4eccd", + "recorded_at": "2026-10-14T23:55:38+00:00", + "result_id": "finding_f0f03374656a1d22b277b9fa61fb799e03bb38ad49b0a7345276a99f170f27c1", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:c4133d596c592402", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 4ebe7c5aa31b19f8f15f0319d7f283999002a934f1835f408fafc88c0fbd08e5 Applicability: Exact40versus10;ten directed single/pair/triple structural cases,two sizes/two seeds. Independent full/compact/weighted-duplicate mixed-label oracle checks and selected-path invariances. Observation: 36dense augmented-system oracles and12translations pass5e-4;10interface and4selected-kernel invariances pass. Across10graph cases,all scored predictions match10;quality deltas-4.79e-5to2.00e-4,2positive/5negative/3zero. Interpretation: Retain40 unsubmitted,unstaged;keep isolated37 ready and native-best10 unchanged. Training-weighted centering preserves known-target multiplicities and fits an unpenalized intercept;the oracle validates this changed objective rather than original ridge parity. No robust local classification gain supports added cost investigation on these cases. Some unscored anchor predictions change;they are not a target-quality gain. The initial ownership assertion compared row indices with node IDs under shuffling;original harness/report/log retained,only assertion mapping corrected. Reused allten comparisons and finished four invariances without repeating the grid or changing source/numeric thresholds. These finite generated conditions do not rule out other centering benefits or prove native utility. Separate37preparation uses current14completed native history,an immutable source and a fail-closed one-shot wrapper;no evaluation was invoked.", + "tags": [ + "writeback-result" + ], + "title": "Does training-centered quadratic ridge with an unpenalized intercept improve owned structural classification over10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "affc21db1cb1fb36", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T00:03:40+00:00", + "result_id": "question_996b04517ca2917f97a8c4f22d019ee2829fe987e09e02f8bac5d580a82df991", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:60bc512c3f3ab687", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "27initial linear/twin candidates on six owned graph bundles;30synthetic reliability/shift controls. Train-only scores,no nonlinear pool or final refit.15-bin ECE proxy.", + "title": "Does cross-fitted calibration-aware screening selection improve independent quality over classification-only selection?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "dc1374976baced70", + "event_kind": "finding", + "evidence_refs": [ + "experiments/crossfit_selection_analysis.json", + "experiments/crossfit_selection_counterexample.json", + "experiments/crossfit_selection_probe.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_996b04517ca2917f97a8c4f22d019ee2829fe987e09e02f8bac5d580a82df991", + "recorded_at": "2026-10-15T00:03:40+00:00", + "result_id": "finding_6a39fb305ea233de69ff76322c79d497fc4250b1946e44dcdfa53688d5e5ecf0", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:ba6e737e8942c8bc", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 66aa427296fc20c8de94f7d870901abb0a7724f17ecea9a3631236e135f046c1 Applicability: 27initial linear/twin candidates on six owned graph bundles;30synthetic reliability/shift controls. Train-only scores,no nonlinear pool or final refit.15-bin ECE proxy. Observation: One6graph choice gains0.000617;one30control loses0.01429 including classification-0.01132. Leakage/order checks pass. Six alternating timing pairs per graph retain added selection medians16-46ms. Flat failure replays exactly. Interpretation: Do not promote a whole-classifier change. The flat-reliability control selects a lower validation classification score on an optimistic crossfit ECE estimate;its full-validation calibrator then has worse heldout ECE and classification. Excluding labels from their own calibration fit prevents that leakage but does not guarantee ranking accuracy. Candidate2 pooled ECE0.01821 versus mean fold0.04176 shows signed-bin cancellation across folds;this does not establish a unique cause. The one graph gain changes twin blend within a shared ridge fit and preserves measured class metrics. A distinct next probe should freeze primary predictions and choose only confidence geometry,calibrating correctness of the fixed primary;use fresh seeds for confirmation and retain confidence-loss controls. This can reuse one ridge fit but still needs cost/full-path evidence. Current37stays ready;native-best10 and prior calibration counterexamples remain unchanged.", + "tags": [ + "writeback-result" + ], + "title": "Does cross-fitted calibration-aware screening selection improve independent quality over classification-only selection?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e9aa67f4068c3a44", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T00:20:42+00:00", + "result_id": "question_0e8b9b1d86ee0406d30057af3a83a08f8db8ece33f0652b53d3801d7aa1627e6", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:030497c1001936f5", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Owned generated reliability controls and graph fixtures;fixed primary predictions. Six fresh full solves compare42with10. Initial geometry and new tie confirmations retained separately.", + "title": "Does preserving equal-margin ties repair validation reliability binning and improve fixed-classifier calibration on owned graph probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "a73b86eeac9dc969", + "event_kind": "finding", + "evidence_refs": [ + "experiments/tied_calibration_analysis.json", + "experiments/tie_aware_geometry_probe.json", + "experiments/tied_calibration_full_path.json", + "experiments/tied_calibration_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_0e8b9b1d86ee0406d30057af3a83a08f8db8ece33f0652b53d3801d7aa1627e6", + "recorded_at": "2026-10-15T00:20:42+00:00", + "result_id": "finding_5c04d69c03a90f34d18bc3e0c2fa082186b639671d91685571ad3b9f0d512aa8", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:42d65980a94efea8", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: f204dd2051a2cbeda4aa8d2bf917eb9ee38e8f5f40a070ef03d04b6f206a6c6a Applicability: Owned generated reliability controls and graph fixtures;fixed primary predictions. Six fresh full solves compare42with10. Initial geometry and new tie confirmations retained separately. Observation: Known80% constant-margin reliability becomes20% under legacy bins;tie repair returns80%.42passes10interface/4graph checks;6fresh solves preserve predictions/choices with2quality gains0.00146/0.00031 and4unchanged.24ordinary pairs retained. Interpretation: The independent count reference confirms a binning defect on exact ties. Extending bin boundaries through ties preserves original smoothing,PAV and interpolation;untied controls are exactly unchanged. Four-arm diagnostic ablation isolates this repair from confidence-geometry choice. Earlier severe constant-geometry losses improve,but a structural case and new controls still lose under geometry selection,so it is not promoted. Actual42 contains only the tie correction on best10;full-path quality gains are calibration-only and do not certify native generalization. Timing and instrumentation-error repair are retained separately. Keep exact37 staged for its isolated native comparison,then consider42without combining the unproved geometry rule.", + "tags": [ + "writeback-result" + ], + "title": "Does preserving equal-margin ties repair validation reliability binning and improve fixed-classifier calibration on owned graph probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "dff03a1f2ee6b1d8", + "event_kind": "finding", + "evidence_refs": [ + "experiments/partial_tie_analysis.json", + "experiments/partial_tie_counterexample.json", + "experiments/partial_tie_reliability.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_0e8b9b1d86ee0406d30057af3a83a08f8db8ece33f0652b53d3801d7aa1627e6", + "recorded_at": "2026-10-15T00:27:40+00:00", + "result_id": "finding_fa944d6c9a347d734762e8d9ccfd02f6345e838a9afc82caec80bcbc4cd0b0b6", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:36ed893d2a9b4341", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: f204dd2051a2cbeda4aa8d2bf917eb9ee38e8f5f40a070ef03d04b6f206a6c6a Applicability: Owned generated reliability controls and graph fixtures;fixed primary predictions. Six fresh full solves compare42with10. Initial geometry and new tie confirmations retained separately. Observation: 336fresh reliability controls:84untied exact;756tied row permutations stable for42. Among252tied cases118gain/114lose. Worst quality loss0.009443and Brier+0.000150replay with independent Python-double sums;shifted controls have negative mean. Interpretation: The tie defect repair and row-order stability remain supported,while heldout calibration utility is conditional. Partial ties change bin boundaries,and monotonicity or distribution-shift assumptions can fail. Worst-case confidence crosses a15-bin ECE boundary;alternative bin counts and Brier are retained as sensitivity diagnostics,not policy-selection targets. Native Squirrel utility is unknown. Preserve the unchanged42checkpoint,its earlier six full solves and all counterevidence. Keep37staged for its isolated native comparison before considering any42native test;avoid tuning to observed ECE bin edges.", + "tags": [ + "writeback-result" + ], + "title": "Does preserving equal-margin ties repair validation reliability binning and improve fixed-classifier calibration on owned graph probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "c1bb86dd73562f09", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T00:37:24+00:00", + "result_id": "question_77fcb1349bbfab7538bc8c1a0b8381de29e35774e15d47efd115d2db9c745af1", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:e0559b4588d767c0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source-matched native public full-ten-split comparison of37and10;39is a separate combined-change reference. Local two-hop prototype results are separate.", + "title": "Did isolated sparse training-row forward37 improve native final_score over native-best10 while preserving quality?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e3d7cb984f786d48", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v37_result.json", + "experiments/native_v37_analysis.json", + "experiments/native_v37_readback.log" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_77fcb1349bbfab7538bc8c1a0b8381de29e35774e15d47efd115d2db9c745af1", + "recorded_at": "2026-10-15T00:37:24+00:00", + "result_id": "finding_7c066e237481a344e5fc6b398b5cb6f0e8785d66bb00b2122e0faa7ebfaabfd9", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:2fe18131b9d52fa4", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 0e9057705b3aa3e5c1f24476bb7d0967f1e66cb4c80f25d2d028c53b02430e05 Applicability: Source-matched native public full-ten-split comparison of37and10;39is a separate combined-change reference. Local two-hop prototype results are separate. Observation: 37passed4/4at0.733936versus10at0.735142. Class metrics,ECE and overall quality match10exactly;mean time7.456vs6.681s and log-time penalty rose. ActualIDaa45953a2e74/agent-15,immutable source and free readback match. Interpretation: Restore source-verified10and retain37. This isolated native run did not convert local sparse-forward savings into a better final_score. Native39had lower aggregate quality after adding shifted screening,while37does not reproduce that loss. That comparison narrows future investigation,but aggregate equality does not prove identical per-node predictions or expose selected models,and sequential timings cannot identify a native hotspot or unique cause. Do not restage unchanged37solely from the earlier local cost result. The independently checked two-hop profile operator is a distinct research route;its standalone development models remain below full10and are not promoted.", + "tags": [ + "writeback-result" + ], + "title": "Did isolated sparse training-row forward37 improve native final_score over native-best10 while preserving quality?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "13e02f712b2a1eb5", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T00:59:26+00:00", + "result_id": "question_d7d8b96c2c3e85431a76995089fd9038746af17b46215707700d534ac52f688f", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:f59c381257365501", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source43versus10;six preregistered fresh compressed/wide-noisy/structural inputs and one separate hub-noise development diagnostic.No native evidence.", + "title": "Does a fixed validation-gated two-hop profile score blend improve the full10classifier on owned graph probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "730f303f9cd9f236", + "event_kind": "finding", + "evidence_refs": [ + "experiments/two_hop_blend_analysis.json", + "experiments/validation_two_hop_blend.json", + "experiments/two_hop_blend_full_path.json", + "experiments/two_hop_hub_diagnostic.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_d7d8b96c2c3e85431a76995089fd9038746af17b46215707700d534ac52f688f", + "recorded_at": "2026-10-15T00:59:26+00:00", + "result_id": "finding_ce96a6fe8e88edd339ad48f02db12998838d2b81ce8824629f49c4918c1fcdb3", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:c6ca9fdf117631f3", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 7c26566303c58f24774f785a599a97fd8ac1f3d3f4ba12cdf98fdd73cb17cca5 Applicability: Source43versus10;six preregistered fresh compressed/wide-noisy/structural inputs and one separate hub-noise development diagnostic.No native evidence. Observation: All6confirmations and1declared hub diagnostic leave mix0;outputs exact10.54augmented ridge references and10interface checks pass.No final profile refit or broad timing executed. Interpretation: Fixed43does not demonstrate full-model utility in this scope. Retain native-best10and runnable43;do not infer selected-path invariance or speed from skipped checks. The single changed diagnostic tests stronger hub suppression and remains development evidence;it does not replace the6confirmations or justify lowering the gate. Prior normalization controls were mixed. Before extending score blending,inspect actual train-only validation error overlap and score scales on2already-used noisy inputs to distinguish missing complementarity from a scale mismatch. No universal ban on two-hop methods or native-generalization claim follows.", + "tags": [ + "writeback-result" + ], + "title": "Does a fixed validation-gated two-hop profile score blend improve the full10classifier on owned graph probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "99da2e1e3e87daf7", + "event_kind": "finding", + "evidence_refs": [ + "experiments/profile_overlap_analysis.json", + "experiments/profile_overlap_diagnosis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_d7d8b96c2c3e85431a76995089fd9038746af17b46215707700d534ac52f688f", + "recorded_at": "2026-10-15T01:06:38+00:00", + "result_id": "finding_1903a018333cc682f8f56c3d8cf8324060f798a9d5e80c97b32aa0de110cb1f6", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:b18f21f0f3aa63b6", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 7c26566303c58f24774f785a599a97fd8ac1f3d3f4ba12cdf98fdd73cb17cca5 Applicability: Source43versus10;six preregistered fresh compressed/wide-noisy/structural inputs and one separate hub-noise development diagnostic.No native evidence. Observation: On2reused noisy validation sets,profiles correct4of262baseline errors but lose12correct rows.Both raw and RMS-matched fixed blends lose quality.No test-truth tuning. Interpretation: Outcome: replan retained: - Native-best10 - Runnable43and scoped counterevidence changed: - Verify label-free fixed-size cubic random features before full-model integration stopped: - Scale-only43expansion and unchanged43staging", + "tags": [ + "writeback-result" + ], + "title": "Does a fixed validation-gated two-hop profile score blend improve the full10classifier on owned graph probes?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2ae8b24065b54251", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T01:14:39+00:00", + "result_id": "question_3cb4365f6cf1324e10e3b51577bc3f719a8f6b0a15af1dedf88791004c7e2639", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:8ad1ff57d0b7a0d9", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed512/2048features,seeds61/73,on2owned5010-node directed cubic controls.Independent map references;no classifier fitted or native evidence.", + "title": "Can a label-free degree3TensorSketch preserve cubic graph similarity at lower representation-construction cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8888017e12561a92", + "event_kind": "finding", + "evidence_refs": [ + "experiments/cubic_tensorsketch_analysis.json", + "experiments/validation_cubic_tensorsketch.json", + "experiments/validation_tensorsketch_float.json", + "experiments/cubic_tensorsketch_probe.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_3cb4365f6cf1324e10e3b51577bc3f719a8f6b0a15af1dedf88791004c7e2639", + "recorded_at": "2026-10-15T01:14:39+00:00", + "result_id": "finding_103cfb799a30691ab1c237829c9439a88c966f84118d4577620a518479740508", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:788af8e362e8e954", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 2e77870e054a267813af53f32f8934bebf71d5ea732225f3dd1e36533b1fbe80 Applicability: Fixed512/2048features,seeds61/73,on2owned5010-node directed cubic controls.Independent map references;no classifier fitted or native evidence. Observation: 18double tensor references max4e-15;18float max1.3e-7.32ordinary pairs:2048active kernel error6.9-9.6%,construction ratio0.42-0.47;512error25-30%.No full-model quality/cost measured. Interpretation: Retain10and the label-free prototype. Both2048maps improve similarity fidelity over512and reduce local representation-construction time,which supports testing a fixed2048seed61single cubic challenger. This is not full-solve speed or native utility:random-feature fitting and applications were excluded,finite kernel error remains,and output bytes are not peak process memory. Prior basis23failed local score and native28/35retained quality but missed total score;the new representation changes that mechanism rather than validating old variants. Require independent fit accuracy,permitted-target ownership,fresh visible-label quality,executed selected-path invariances and ordinary complete-solve costs before staging.", + "tags": [ + "writeback-result" + ], + "title": "Can a label-free degree3TensorSketch preserve cubic graph similarity at lower representation-construction cost?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fbef8a00dd5d4444", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T01:39:43+00:00", + "result_id": "question_29c8d459a978d7f50891ffbb7d2eb46eaf9f959fa4d969b4332fc096a40d8049", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:480c0df7e75cc7a0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source-matched public full-ten-split45versus10;45combines sketch mapping,direct smaller-system fitting andcanonical input order.Local44failure and45repair are separate evidence.", + "title": "Did fixed2048cubic-sketch challenger45 improve native full10final_score over native-best10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "cb77fd8c850f7963", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v45_result.json", + "experiments/native_v45_analysis.json", + "experiments/native_v45_readback.log", + "experiments/after_native45_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_29c8d459a978d7f50891ffbb7d2eb46eaf9f959fa4d969b4332fc096a40d8049", + "recorded_at": "2026-10-15T01:39:43+00:00", + "result_id": "finding_9480a6a4d431899210e7e90e70eba26ddd8c124d0a592f082f1a7d8efaf0a707", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:1a676c32eb2d4ebf", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: f12b1b92928b50c96198af9a34ff45e138525917fb5590066f85103d51f7badf Applicability: Source-matched public full-ten-split45versus10;45combines sketch mapping,direct smaller-system fitting andcanonical input order.Local44failure and45repair are separate evidence. Observation: 45passed4/4at0.731336vs10at0.735142.Aggregate class/ECE/quality match10;mean time9.635vs6.681s,geometric ratio1.463.ID1e69e93e19a8/agent-16andfree readback match exact source;restore10. Interpretation: Retain source45and scoped local evidence;do not restage unchanged45from synthetic gains. Mathematical and interface checks did not establish native utility. Public aggregates cannot prove selection decisions,nodewise parity or a timing stage;45combines map approximation,direct fitting andcanonical ordering. Native26/28/35had a small quality gain with excessive log-time cost;45has no reported aggregate gain. This supports a different feasibility question:reuse already-computed graph/word validation experts andtest one fixed local support gate onaligned,reversed anduninformative reliability controls before any full-model integration. Keep the native-best10and original constraints.", + "tags": [ + "writeback-result" + ], + "title": "Did fixed2048cubic-sketch challenger45 improve native full10final_score over native-best10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b6a657117aa32ad9", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T01:50:16+00:00", + "result_id": "question_522f6b21fbb50655d00d47dde0c4cf41c035e138729ad60c2055455cb7b57e6a", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:07d2cd940181abfd", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Original10experts,zero twinweight,word weight8/(8+known-neighbor-count);six new owned aligned/reversed/uninformative controls,seeds605837/705841.No native data.", + "title": "Does fixed8support gating improve validation quality over full10on both preregistered aligned-reliability controls?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0b983ebf0dc9fb88", + "event_kind": "finding", + "evidence_refs": [ + "experiments/support_gate_analysis.json", + "experiments/support_gate_probe.json", + "experiments/validation_support_gate.json", + "experiments/after_support_gate_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_522f6b21fbb50655d00d47dde0c4cf41c035e138729ad60c2055455cb7b57e6a", + "recorded_at": "2026-10-15T01:50:16+00:00", + "result_id": "finding_7c971d0bfefbb7fa999225cfb53afce041b61de921113f024d0c74698e40a522", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:72d305ca28eaafed", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 9fbcad36a9454f86d8f2a485e69aead8d7b06c52a363901b4c1546ea8147873e Applicability: Original10experts,zero twinweight,word weight8/(8+known-neighbor-count);six new owned aligned/reversed/uninformative controls,seeds605837/705841.No native data. Observation: Both aligned validation deltas negative(-.01205,-.02331),despite gains over single experts.Reversed rejected.One uninformative case passes+.004gate;two final refits gain.01128heldout quality.Audit54quality comparisons passes. Interpretation: Stop this fixed rule because complete10already covers both intended aligned cases. Retain the positive uninformative result;it neither proves support predicts reliability nor refutes all local mixtures. Only selected case has train+valid expert refits;other heldout expert scores are development evidence.Operator invariance does not certify a new integrated solver. Next verify transferable neighborhood-word context independently from adjacency identities;require negative controls andfull10comparison before staging.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed8support gating improve validation quality over full10on both preregistered aligned-reliability controls?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2bd6ac567b0f32f4", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T01:59:19+00:00", + "result_id": "question_c0e55b98afdad8a3b4d1ab5e02b2304a9c711f25575270688757ff2d79230353", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:00fc9d187b644202", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source46 vs10 on six new owned directed vocabulary-transfer/shuffled/noisy controls,805853/905887;train-only selection,known-label final refits. No native data or cost claim.", + "title": "Can fixed neighborhood-word context transfer across disjoint neighbor identities beyond incumbent10?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7b42d69841efd90a", + "event_kind": "finding", + "evidence_refs": [ + "experiments/neighbor_context_analysis.json", + "experiments/neighbor_context_full_path.json", + "experiments/validation_neighbor_context.json", + "experiments/neighbor_context_order_diagnosis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_c0e55b98afdad8a3b4d1ab5e02b2304a9c711f25575270688757ff2d79230353", + "recorded_at": "2026-10-15T01:59:19+00:00", + "result_id": "finding_855d5598a70f3cdbb8d46a3256e3aab42580393c802659521fb076a55e027d06", + "run_id": "checkpoint:a17aab78bdd8c3c755ca07ff8e028397a44e7b9fe80903be6ee607a248a71da7:dd1ce7b7dd281fa1", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: 1de94794ac4b7f2a00589b226e2ec831528d1f91b37bdd770c8fd50947e70151 Applicability: Source46 vs10 on six new owned directed vocabulary-transfer/shuffled/noisy controls,805853/905887;train-only selection,known-label final refits. No native data or cost claim. Observation: Transfer quality gains .73058/.72307; noisy .27679/.40127; shuffled unselected and exact10. Non-flat row confidence delta .000769 fails .0002; actual rho.02 screen errors .00114/.00089 fail .0005. Keep10. Interpretation: Twelve path-sum oracles,18small fit references and10interface checks pass;four flat-confidence selected invariances pass. The added non-flat check exposes a limit: same-input calibration curves differ3.07e-5,while final-score changes under one fixed curve move confidences7.47e-4. Actual selected rho2 fits pass,but low-ridge screens do not. Retain positive andnegative results without staging46. Next repair context-only precision/convergence,not thresholds or baseline arithmetic,then replay actual systems and fresh confirmations before full-size cost. These designed controls do not establish native utility.", + "tags": [ + "writeback-result" + ], + "title": "Can fixed neighborhood-word context transfer across disjoint neighbor identities beyond incumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "60f13d8b56805829", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T06:49:11+00:00", + "result_id": "question_cee67ee53f286da7793e216d6c0e1b01aca930d9f1aa6a2b5d6a00f577ee3150", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed source47versus10,public destination full10native agent-17;local numerical/utility/cost evidence is distinct.", + "title": "Does numerically repaired neighborhood-word context47 improve native full10final_score beyond incumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "35d8ba46509dc80d", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v47_result.json", + "experiments/native_v47_analysis.json", + "experiments/context_precision_analysis.json", + "experiments/context_caps_full_probe.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_cee67ee53f286da7793e216d6c0e1b01aca930d9f1aa6a2b5d6a00f577ee3150", + "recorded_at": "2026-10-15T06:49:11+00:00", + "result_id": "finding_4076f6182c219915c6e5dc85e57f7cec41acf71c94af1c657dd8c67571de21f8", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 1770bfe1430beb0c7c59bb801db8973c069811edca98440af686ccc1f4695246 Applicability: Fixed source47versus10,public destination full10native agent-17;local numerical/utility/cost evidence is distinct. Observation: 47valid4/4at.715421vs10at.735142.Aggregate class/ECE match;mean47.275vs6.681s.ID09896f102880/agent-17,immutable source1770bfe1430b.Free readback confirms;restore10. Interpretation: 47fixes46actual-fit and non-flat order failures and gains on designed transfer controls. Native47has no reported aggregate quality gain and substantial time cost. This refutes adopting this fixed integration,not every context mechanism. Public aggregates do not identify branch selections,nodewise predictions or a timing stage. Full-size110/160cap ablation passes observed references,but47runs stop12to75iterations;mixed ordinary timing ratios provide no causal cap speedup. Keep47/48evidence without restaging. Compare a materially different context execution route with separate tie-only42under a fresh current replan;preserve original objective and best10. Next proposal is a fixed128signed context sketch;its cost source and native benefit remain unproven.", + "tags": [ + "writeback-result" + ], + "title": "Does numerically repaired neighborhood-word context47 improve native full10final_score beyond incumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "c94ff868847f54f3", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T07:10:18+00:00", + "result_id": "question_87bea16e5bddbef0ff63437978333c1ff969a45c67d179a57b315f0ee331818c", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed49versus10and47on owned operator/fit references,6replays,2fresh controls andfour5201/2089cost cases. No native49result.", + "title": "Can fixed128signed neighborhood-word context49 retain controlled transfer utility while reducing complete cost versus dense47?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "daa39df649afef0c", + "event_kind": "finding", + "evidence_refs": [ + "experiments/context_sketch_analysis.json", + "experiments/validation_sketched_words.json", + "experiments/validation_context_sketch.json", + "experiments/context_sketch_full_cost.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_87bea16e5bddbef0ff63437978333c1ff969a45c67d179a57b315f0ee331818c", + "recorded_at": "2026-10-15T07:10:18+00:00", + "result_id": "finding_b920f5a3fd18f3fd73d20b165a0a734364747f237c8abab04b165fc56bd243dc", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: b97a0f396db8ac983e79c100a281d47ea960d968aae34335a95ceb1cdb68101c Applicability: Fixed49versus10and47on owned operator/fit references,6replays,2fresh controls andfour5201/2089cost cases. No native49result. Observation: 49passes numeric/non-flat gates.Fresh Qgains vs10:.73494/.30500;shuffled rejected exact10.32ordinary pairs:geometric49/47time.635.49still costs6to30percent vs10;noisy replay Qlosses vs47:.05341/.02337.No native49submission. Interpretation: One fixed128signed sketch changes only context representation;baseline10and checked47fitter/gates/calibration stay unchanged. Twelve scalar operator references,18small and8actual Gram comparisons pass;row/mask/duplicate/class checks include a non-flat selected case. Full-size paired costs support a local computation reduction,not a native hotspot or transfer claim. Some noisy quality is lost versus47;47native quality/ECEflat andtime regression remain counterevidence. Keep10best;stage immutable49forone eligible source-bound native decision after history reconciliation. Native path waits on a dated window with finite monitor;independent tests ask whether context adds utility to a stronger own-feature baseline. No width grid or unchanged47/48retry.", + "tags": [ + "writeback-result" + ], + "title": "Can fixed128signed neighborhood-word context49 retain controlled transfer utility while reducing complete cost versus dense47?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d71ae27cef51429f", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T07:22:55+00:00", + "result_id": "question_ffcc254f047fadbd5c380581c1e043095c2fc95b3112e080666def1563a56fb4", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Four fresh owned5201/2089controls,seeds2006029/2106031,65percent informative own codes,stable20percent-coherent versus shuffled anchor vocabularies;immutable10/47/49.", + "title": "Does immutable49add complementary signal when10hasinformative own-word features?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ab718bd005a4e439", + "event_kind": "finding", + "evidence_refs": [ + "experiments/context_complementarity_analysis.json", + "experiments/context_complementarity.json", + "experiments/after_complementarity_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_ffcc254f047fadbd5c380581c1e043095c2fc95b3112e080666def1563a56fb4", + "recorded_at": "2026-10-15T07:22:55+00:00", + "result_id": "finding_927c0ec3785c63fa744ba79fc4df23633c57bf4ef5e2bc403f77260680df7a00", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: b97a0f396db8ac983e79c100a281d47ea960d968aae34335a95ceb1cdb68101c Applicability: Four fresh owned5201/2089controls,seeds2006029/2106031,65percent informative own codes,stable20percent-coherent versus shuffled anchor vocabularies;immutable10/47/49. Observation: 49unselected andexact10onall4.Dense47stable quality gains.21499/.19146withclass gains.21497/.19749;second ECE worsens.06038.Shuffled47exact10.All12ownership/independent metric checks pass. Interpretation: Scoped fixed49complementarity claim fails onthese4controls;dense47shows headroom,so chance-level baseline is notnecessary for useful context. This does not refute all projections or establish native activation. Retain earlier49gains andcost reductions,47native quality-flat/time regression,andfixed-feature/synthetic limits. No threshold tuning or source changes. Next one fixed train-only supervised coordinate selection route withindependent leakage/numeric gates,fresh confirmations andordinary cost. Existing49native task remains subject tocurrent review andsource/history reconciliation.", + "tags": [ + "writeback-result" + ], + "title": "Does immutable49add complementary signal when10hasinformative own-word features?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8111e8bc95102233", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T07:36:28+00:00", + "result_id": "question_b30c54fdc3c2a61f750ac7b157ea30c34e4b25f32a3fffbbb893630cd3ef96f7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source50, four full-size complementarity replays and two fresh5201/2089 stable20/own65 controls2206049/2306053; diagnostic replay2306053 compares frozen47/49.", + "title": "Can fixed supervised128 context coordinates retain complementary utility through independent stronger-baseline confirmations?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "159c06257d87a146", + "event_kind": "finding", + "evidence_refs": [ + "experiments/supervised_context_analysis.json", + "experiments/validation_supervised_context.json", + "experiments/supervised_context_failure_probe.json", + "experiments/after_supervised_context_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_b30c54fdc3c2a61f750ac7b157ea30c34e4b25f32a3fffbbb893630cd3ef96f7", + "recorded_at": "2026-10-15T07:36:28+00:00", + "result_id": "finding_2c8b9ee09d558a941c855183b91d70d1d4805c7c42ff76a3576d7af3d4956fd8", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 4d14e18e98edefd7b0c48b0e090f7ae7185ccfd88397df42c9e731c2199ddde4 Applicability: Source50, four full-size complementarity replays and two fresh5201/2089 stable20/own65 controls2206049/2306053; diagnostic replay2306053 compares frozen47/49. Observation: 12selector,18small/8actual fit,10interface and4non-flat checks pass. Fresh Qdelta+.22232/0; second falls back exact10. Diagnostic replay:10/50=.71840,47=.99616,49=.95495.50not staged;cost skipped. Interpretation: Fixed50 fails its declared fresh utility gate despite numerical and label-ownership checks. Two stable replays gain.18752/.11372; shuffled controls are exact10. Keep all outcomes and stop this fixed route without tuning. The failure replay is post-hoc,not fresh confirmation: it shows headroom and that49can help on this stronger-baseline case,without proving native activation. Best10 and active49 remain exact. Preserve existing49native opportunity for eligible source/history review; compare a separate fixed label-free128 subspace proposal with its own gates and fresh inputs. No native50evaluation or timing claim.", + "tags": [ + "writeback-result" + ], + "title": "Can fixed supervised128 context coordinates retain complementary utility through independent stronger-baseline confirmations?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b60cfeb4ab3f122c", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T07:43:07+00:00", + "result_id": "question_e2c3fd2ef8855e5b7c7da8fa646b2ea5ff42510b2db03345439469e2b6c09cb9", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source51operator stage only:9independent covariance/SVD references,4zero/empty and1rank-deficient case. Full-solve,utility andcost evidence is pending.", + "title": "Can a fixed label-free128 orthogonal context subspace retain utility at materially lower cost than dense47?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7e835bea9f34c71c", + "event_kind": "finding", + "evidence_refs": [ + "experiments/label_free_context_operator_analysis.json", + "experiments/validation_label_free_operator.json", + "experiments/label_free_context_review.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_e2c3fd2ef8855e5b7c7da8fa646b2ea5ff42510b2db03345439469e2b6c09cb9", + "recorded_at": "2026-10-15T07:43:07+00:00", + "result_id": "finding_57d3da65845c041237d62335db2c4dcd6d828f52a6e5d72409ad755deac2e48e", + "schema_version": "loopx_explore_result_event_v0", + "status": "tentative", + "summary": "Input revision: de5bc9ce6c5579441d49e5ff385d281f2607a22c956c3eaf6db30737f03c474e Applicability: Source51operator stage only:9independent covariance/SVD references,4zero/empty and1rank-deficient case. Full-solve,utility andcost evidence is pending. Observation: 14operator checks pass;projector androw-order errors<1e-8,orthogonality<1e-10. All47methods except context features unchanged.51not staged.49native window is eligible;remaining51gates stay open. Interpretation: This establishes the fixed range-finder implementation onowned inputs,not utility or native improvement. Preserve50fresh-gate failure andall49counterevidence. Save the operator stage to prioritize the now-eligible49native decision underfresh current authority;do not substitute51. Keep51task open foractual-fit/non-flat checks,six explicit replays,two fresh controls andconditional ordinary costs,within its remaining active budget. Basis takes features only;full-solve masking/basis checks remain pending.", + "tags": [ + "writeback-result" + ], + "title": "Can a fixed label-free128 orthogonal context subspace retain utility at materially lower cost than dense47?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0f848a8c13a7fafe", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T07:55:50+00:00", + "result_id": "question_b6f0f9a175154f249cb3efbb8d031dd2013d1f1d0763d6452da63c70479c637c", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Immutable49 b97a0f396db8 versus native10incumbent, public full10 agent-18 result2460a42e064d. Local full-size replay remains separate.", + "title": "Does exact fixed128signed context49 improve valid native full10score beyond incumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3349fd974d010204", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v49_result.json", + "experiments/native_v49_analysis.json", + "experiments/validation_native49_full_control_normalized.json", + "experiments/native_v49_source_binding.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_b6f0f9a175154f249cb3efbb8d031dd2013d1f1d0763d6452da63c70479c637c", + "recorded_at": "2026-10-15T07:55:50+00:00", + "result_id": "finding_721ad342275067d3dbce0e63226133dc968adf3f3e93d34ebbee76a760b386aa", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: b97a0f396db8ac983e79c100a281d47ea960d968aae34335a95ceb1cdb68101c Applicability: Immutable49 b97a0f396db8 versus native10incumbent, public full10 agent-18 result2460a42e064d. Local full-size replay remains separate. Observation: 49valid4/4 full10score.729531vs10at.735142;aggregate class/ECE exact10and47. Mean11.651vs6.681s,penalty.024233vs.018622. Free readback confirms. Restore10.3440-target reference error2.76e-7;4non-flat full-size checks pass. Interpretation: Fixed49reduces reported native time versus47butadds overhead versus10without aggregate quality gain. This refutes adopting this fixed source,not all context. Public equality does not identify branch activation,nodewise parity or a timing hotspot. Preserve mixed local49utility,50fresh-gate failure andunqualified51operator evidence. Full-control first harness failure was a tuple/list comparison bug;retain it andthe corrected passing run without changing49. Best10is restored. Retire fulfilled window monitoring;review native cost/utility evidence andremaining51budget before further qualification,staging oranother native decision.", + "tags": [ + "writeback-result" + ], + "title": "Does exact fixed128signed context49 improve valid native full10score beyond incumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "c11226d93fd1779b", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T08:15:24+00:00", + "result_id": "question_81dfa5396101dfe71997cc381bb478e70ddad1d0b0e6f2017117b5eb1d0e9018", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed source51 de5bc9ce6c55;6explicit50replays and2fresh5201/2089stable20/own65controls2406059/2506067. Numeric/interface/basis/nonflat checks separate;cost conditional andunrun.", + "title": "Does fixed51 satisfy its predeclared full-solve utility qualification on six replay and two fresh controls?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "acbb69b45bd6d1c4", + "event_kind": "finding", + "evidence_refs": [ + "experiments/label_free_context_analysis.json", + "experiments/validation_label_free_context.json", + "experiments/validation_label_free_context_settlement.json", + "experiments/after_label_free_context_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_81dfa5396101dfe71997cc381bb478e70ddad1d0b0e6f2017117b5eb1d0e9018", + "recorded_at": "2026-10-15T08:15:24+00:00", + "result_id": "finding_172ff60a6f5b264b4f03a0eac116d3562d3266b30100062ce4b87e233b74182e", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: de5bc9ce6c5579441d49e5ff385d281f2607a22c956c3eaf6db30737f03c474e Applicability: Fixed source51 de5bc9ce6c55;6explicit50replays and2fresh5201/2089stable20/own65controls2406059/2506067. Numeric/interface/basis/nonflat checks separate;cost conditional andunrun. Observation: 18small/8actual references pass,max1.23e-6;10interface/4nonflat and2basis-label checks pass. Five stable controls gain,but2206049replay rejects context andreturns exact10. Both fresh Qgains.206737/.085599. No51cost,staging ornative. Interpretation: Fixed51fails its declared all-stable utility gate. Retain its five stable gains andtwo exact10shuffled controls;this is not a numerical or label-ownership failure anddoes not refute every subspace ornative utility. Stop costs/staging without retuning.265.456active local seconds total. Best10remains exact. Prefer next source-bound native review ofisolated42calibration over unchanged context retries;preserve42shifted quality losses andnative cooldown. Prior operator-only finding remains valid within its scope.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed51 satisfy its predeclared full-solve utility qualification on six replay and two fresh controls?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "28b4c2707665fb13", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T08:41:40+00:00", + "result_id": "question_b1ba9d9c58377e4927d839c0b4437e2dc118f6e6db9eacbd7277e409e42bdee3", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "52f96fe30b82ac versus10:18small/12reused broad references,4fresh5201/2089controls2606071/2706077,10interface/4nonflat checks. Conditionalcost unrun;no native52.", + "title": "Does fixed independent-CG panel52 meet its frozen numerical,whole-solve andcost qualification?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "86cb075bf1bd6aff", + "event_kind": "finding", + "evidence_refs": [ + "experiments/independent_cg_panel_analysis.json", + "experiments/validation_independent_cg_panel_nonflat.json", + "experiments/validation_independent_cg_panel.json", + "experiments/after_independent_cg_panel_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_b1ba9d9c58377e4927d839c0b4437e2dc118f6e6db9eacbd7277e409e42bdee3", + "recorded_at": "2026-10-15T08:41:40+00:00", + "result_id": "finding_c1f693ff1245fdcd799a10cfcbf52c95623fb6e9138245967268a6ef8d382c57", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: f96fe30b82ac90fd059781dfad9df24577c4c15be1b29ce34c0fe8783420f631 Applicability: 52f96fe30b82ac versus10:18small/12reused broad references,4fresh5201/2089controls2606071/2706077,10interface/4nonflat checks. Conditionalcost unrun;no native52. Observation: 18small,4fullcontrols,10interface/4nonflat pass. Four rho.02broad references fail5e-4inboth52and10;52errors.034-.171andpanel/base delta<=.21464. Fullclasses/choices match;maxconfidence delta2.15e-6. No52cost/staging/native. Interpretation: Fixed52fails its declared absolute numerical qualification. The4failures are shared,not candidate-only;they do not prove thatbatching worsens native quality. Preserve every local parity result anddifference. Independentstates still change floating arithmetic;separate matrix-product width from norm-reduction width before anydistinct repair decision. No speed claim from unrun costs. Initialnonflat harness incorrectly chose span.04793;retain failure andfinish unchangedthreshold checks onexisting span.15252control. Native42remains exactandprioritized;best10intact. Native39cause andnative52utility remainunknown.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed independent-CG panel52 meet its frozen numerical,whole-solve andcost qualification?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9d26464fc723b668", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T08:50:32+00:00", + "result_id": "question_048edab76534765e21a8497e277f0d12bbe5445cca080154a9948c1333be0b6c", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Float32 fullblock CG,width5/15 xcontiguous norm/dotwidth5/15;960nodes/720targets,seeds89/8119reused and2806081/2906089fresh;rho2/.2/.02,caps110/160. No native/cost inference.", + "title": "Do product and reduction width independently perturb independent-CG52 versus10 on fixed broad stress controls?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "59565fccfa3c5d1b", + "event_kind": "finding", + "evidence_refs": [ + "experiments/cg_width_attribution_analysis.json", + "experiments/cg_width_attribution.json", + "experiments/attribute_cg_widths.py" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_048edab76534765e21a8497e277f0d12bbe5445cca080154a9948c1333be0b6c", + "recorded_at": "2026-10-15T08:50:32+00:00", + "result_id": "finding_919cab48a7c402f59218657c39d47fc9772c47a2fce6395964010a324b20a179", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: a9f5dcca122232a0c6a8c883792d87c0397d5df78a009f73b31754eeaf4fb715 Applicability: Float32 fullblock CG,width5/15 xcontiguous norm/dotwidth5/15;960nodes/720targets,seeds89/8119reused and2806081/2906089fresh;rho2/.2/.02,caps110/160. No native/cost inference. Observation: 32runs/96ridge results;16 exact controls pass. Reduction alone changes the initial norm; product alone changes the first product at iteration1. Both perturb trajectories. All32 ridge.02 runs cap and fail double score5e-4; higher ridges pass. Sources10/42/52 unchanged. Interpretation: Both arithmetic widths independently change floating trajectories; effects interact. Exact positive controls establish this scoped attribution. Neither mixed corner repairs the shared stress error. Ridge.2 stopping can move one iteration, but every ridge.02 run reaches its cap. Retain coefficients, scores, stopping counts, true residuals and independent double references. Deprioritize width-only repair and prioritize eligible42 native review. Native39 cause and52 utility remain unknown. Qualification and user acceptance are unchanged.", + "tags": [ + "writeback-result" + ], + "title": "Do product and reduction width independently perturb independent-CG52 versus10 on fixed broad stress controls?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1f2c5b14f865777a", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T08:57:39+00:00", + "result_id": "question_6020666a8dc9388f934555e0bc0339362ff228df62c7bfd8e96352975992672f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Immutable42 f204dd2051a2, destination native5978f280dd9f/agent-19 versus10 2b2d4c02e380/agent-6. Public ten-split aggregates; local mixed controls are separate evidence.", + "title": "Does exact tie-only42 improve valid native full10 final score beyond incumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "c713df41932e3e6a", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v42_result.json", + "experiments/native_v42_analysis.json", + "experiments/native_v42_current_review.json", + "experiments/after_native42_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_6020666a8dc9388f934555e0bc0339362ff228df62c7bfd8e96352975992672f", + "recorded_at": "2026-10-15T08:57:39+00:00", + "result_id": "finding_1b3433ff133dac4fb16e31d765e523ab39b8f4723e5dfda61f254849ef680ad3", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: f204dd2051a2cbeda4aa8d2bf917eb9ee38e8f5f40a070ef03d04b6f206a6c6a Applicability: Immutable42 f204dd2051a2, destination native5978f280dd9f/agent-19 versus10 2b2d4c02e380/agent-6. Public ten-split aggregates; local mixed controls are separate evidence. Observation: 42 valid4/4 full10 scores .73227310 versus10 .73514207. ECE improves .00045299 and class metrics match; mean time9.113s versus6.681s. Quality gains .00004530; time penalty rises .00291427. Free readback verified; exact10 restored. Interpretation: The tested source fails the native adoption criterion. Retain its positive aggregate calibration evidence, six local solves/24 ordinary pairs and336 mixed controls including shifted losses. Aggregate class equality does not prove nodewise parity. One native time observation per source does not identify a hotspot or prove repeatable calibrator overhead, especially with local paired near-parity. Stop unchanged42 retries and preserve exact10. Require a fresh semantic comparison before another primary-quality or distinct solver experiment;52 remains unqualified and no numerical gate is relaxed.", + "tags": [ + "writeback-result" + ], + "title": "Does exact tie-only42 improve valid native full10 final score beyond incumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "96fdfbe63e5c003f", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T09:27:02+00:00", + "result_id": "question_af3cb08d2f4fe246fa0ede15ee8353830994d3ece90acc343e794f462b24f0ca", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Immutable53 835244f4cc08 versus10;36small/16actual weighted references,8fresh5201/2089prior controls,10interface and4nonflat checks. No native53 orcost evidence.", + "title": "Does fixed class-balanced linear ridge53meet its numerical,utility,invariance andcost qualification?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "4c903919d8c67e43", + "event_kind": "finding", + "evidence_refs": [ + "experiments/class_balanced_ridge_analysis.json", + "experiments/validation_class_balanced_ridge.json", + "experiments/after_class_balanced_ridge_route.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_af3cb08d2f4fe246fa0ede15ee8353830994d3ece90acc343e794f462b24f0ca", + "recorded_at": "2026-10-15T09:27:02+00:00", + "result_id": "finding_3cd5053251dd70dd0bc2287cfb5a897b276037260a2373f5db32d2a4b977be57", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 835244f4cc085394d2fa8f031e453f842cfb25d5e190e0fad3bb29530b8ff4bc Applicability: Immutable53 835244f4cc08 versus10;36small/16actual weighted references,8fresh5201/2089prior controls,10interface and4nonflat checks. No native53 orcost evidence. Observation: All52math references and10interface checks pass. Eight meanquality+.10706,worst0;both balanced outputs exact10. Nonflat confidence changes .000274(roworder)/.000242(classperm) exceed2e-4;predictions equal. Fixed53stopped beforecost/staging/native. Interpretation: Refuted is scoped to frozen qualification ofthis exactsource. Six skew cases retain large classification gains;do not discard weighted-lossfamily. Unknown-label andduplicate/selfedge tests pass. Original actualfit error<=.000155doesnotprove transformed fits ordownstreamconfidence invariance. Baseline10same-control sensitivity wasunmeasured,so first-divergence attribution isunknown. Allharness failures are retained;candidate andcriteria unchanged. Require freshsemanticreview ofsource-frozen attribution before a distinct repair. Protect10andpriornegativeevidence;native validfull10score still governs adoption.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed class-balanced linear ridge53meet its numerical,utility,invariance andcost qualification?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "411a684e707f7512", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T09:42:31+00:00", + "result_id": "question_b062b1aef471a0f4d1287a6f014592c9b05691764f0346104464661b753ea5c7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Unchanged10/53;owned3006091stationary-skew5201/2089,original,row/partition andclass permutations. Six tracedsolves/12actualfit references. No cost/native inference.", + "title": "Are53nonflat confidence failures unique to weighting,anddoes selected-fit accuracy orcalibrator drift dominate?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e6ee4639a84cbc19", + "event_kind": "finding", + "evidence_refs": [ + "experiments/class_balanced_sensitivity_analysis.json", + "experiments/class_balanced_sensitivity.json", + "experiments/class_balanced_ridge_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_b062b1aef471a0f4d1287a6f014592c9b05691764f0346104464661b753ea5c7", + "recorded_at": "2026-10-15T09:42:31+00:00", + "result_id": "finding_a0e9f801c162867bd385f3a352b1ae0a897c92eaa18ecef872381acae42f522d", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: b5a2bf50ca9c15324c98708dea1f28dfbbd85806c7715236c60803616317c8c8 Applicability: Unchanged10/53;owned3006091stationary-skew5201/2089,original,row/partition andclass permutations. Six tracedsolves/12actualfit references. No cost/native inference. Observation: Exact2originalreplays;12refs pass5e-4. Both10/53fail2e-4confidence underbothtransforms;classes/settings equal. Calibrator-only delta<=4.71e-6;fixedcalibrator scorechanges retain<=2.87e-4. Double-fit pipeline delta<=2.155e-7. Sources unchanged. Interpretation: Confirmed onlyfor shared numerical sensitivity onthisownedcontrol. Tiny sharedpreprocessing perturbations are amplified in approximatefloat32CGfits;doublefits onthe samegeometry suppress sensitivity. Binvalues areexact;calibratorcenterdrift issecondary. Doesnotseparate stopping tolerance fromroundoff,prove recursive residualfailure,qualify53ortransfer toSquirrel. Retain53eight-control primarygains andfailedgates. Stopitscost/staging/native;freshsemantic comparison ofdistinctselected-fitaccuracyrepair versusnewloss mustinclude complete future qualification andfullmarginalcost.42nativecalibrationloss remainscounterevidence;nativeadoptionrequiresvalidfull10scoreabove10. Initialtuple/listdiagnostic-harnessfailure retained;originalnumericcriteria andtimer unchanged.", + "tags": [ + "writeback-result" + ], + "title": "Are53nonflat confidence failures unique to weighting,anddoes selected-fit accuracy orcalibrator drift dominate?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d4b02c18269b1d22", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T10:05:38+00:00", + "result_id": "question_a10d54e7121c9bb3186253b8c5a426859c3344bbcb753aaca0903ba04aa3c43a", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Immutable54 a7e7db9b882797d1f0a0cdec4016fc3fb4422e4fbfa65f829a362824f6501629;12owned5201/2089controls,eightnonflattransforms,62numericrefsand16ordinary10/54pairs. Synthetic/localonly.", + "title": "Does fixed selected-fit exact weighted54repair53confidence sensitivity whilepreservingprimaryutility atboundedfullcost?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "202e9df3a75d2448", + "event_kind": "finding", + "evidence_refs": [ + "experiments/selected_exact_weighted_analysis.json", + "experiments/validation_selected_exact_weighted.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_a10d54e7121c9bb3186253b8c5a426859c3344bbcb753aaca0903ba04aa3c43a", + "recorded_at": "2026-10-15T10:05:38+00:00", + "result_id": "finding_094ac6a2b2faaf7a44c6c23199f0e1d791492b103cf94b17a8dbfd34f7166fac", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: a7e7db9b882797d1f0a0cdec4016fc3fb4422e4fbfa65f829a362824f6501629 Applicability: Immutable54 a7e7db9b882797d1f0a0cdec4016fc3fb4422e4fbfa65f829a362824f6501629;12owned5201/2089controls,eightnonflattransforms,62numericrefsand16ordinary10/54pairs. Synthetic/localonly. Observation: 62refs/10interface/12controls/8nonflatchecks pass. MeanQdelta10 0.092047;maxconfidence delta 1.75e-07.16ordinarypair geomtime 1.4896,meanscoregain 0.058078. Localqualified=True. No native. Interpretation: Proceed to separate source-bound native continuation afterfreshcurrentreview;local qualification grantsnoexecution. Scope is exact54fixedqualification,not weighted-lossfamily ornativeSquirrel. Original53screen diagnostics remainexact;53failedconfidencequalification isnot rewritten. Balanced double objective ismatched without requiringold10floatbitwise outputs. Allharnessfailureevidence andold52broadCGfailures retained. Denseoracles contribute toreportedlocalprocessmemory. Native10remainsadoptionauthority;prior42/47/49nativequality/timecounterevidence stillapplies.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed selected-fit exact weighted54repair53confidence sensitivity whilepreservingprimaryutility atboundedfullcost?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "513411be087ba333", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T10:12:22+00:00", + "result_id": "question_6d9e21e493f5c485197791557adf1916b4edabbc46f592ea5061705d28a7b3c7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Immutable54 a7e7db9b8827;native62448afac7bd/agent-20 versus10 2b2d4c02e380/agent-6. Publicten-split aggregates;source-boundlocalqualification separate.", + "title": "Does exact weightedselected-fit54improve validnativefull10finalscore overincumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1f9b9c5942e42004", + "event_kind": "finding", + "evidence_refs": [ + "experiments/native_v54_analysis.json", + "experiments/native_v54_result.json", + "experiments/selected_exact_weighted_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_6d9e21e493f5c485197791557adf1916b4edabbc46f592ea5061705d28a7b3c7", + "recorded_at": "2026-10-15T10:12:22+00:00", + "result_id": "finding_1323e1e2928bbc8096aeba237e0e0ea186499c8cb60226e1a1d6b68fd64bc735", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: a7e7db9b882797d1f0a0cdec4016fc3fb4422e4fbfa65f829a362824f6501629 Applicability: Immutable54 a7e7db9b8827;native62448afac7bd/agent-20 versus10 2b2d4c02e380/agent-6. Publicten-split aggregates;source-boundlocalqualification separate. Observation: 54valid4/4full10score .72106677 versus10 .73514207. Quality-.008455;timepenalty+.005620;ECEimproves .000879. Meanseconds11.524versus6.681. Free readback and20completedhistory verified;exact10restored. Interpretation: Refuted onlyfortested54nativeadoption. Preserve62numericrefs,12ownedcontrols,8nonflatchecks and16ordinarypairs:theyvalidate numericalrepair/syntheticskewutility,notnative transfer. Nativeaggregateclassificationdeclines andaddedtime outweighssmallcalibrationgain. Publicfeedbackdoesnotisolateweightedlossversusselectedprecision,exposeselectedfamilies,nodewisepredictions orprovea hotspot. Do notdiscardallweightedlosses,rewrite53gates,retryunchanged54oroptimizeECEbinnedfeedback. Keep10;freshsemanticcomparison mustseekprimaryqualitywithcompletecost andretainprior42/47/49nativefailures. Native20cooldown affects onlyfutureevaluation;eligibilitygrantsnoexecution/capture.", + "tags": [ + "writeback-result" + ], + "title": "Does exact weightedselected-fit54improve validnativefull10finalscore overincumbent10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ef765c07c00406cf", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T10:39:28+00:00", + "result_id": "question_497445890ae118799d8a4a305a6a3b797a6e9dbd81e72880974800d47cd55f5d", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed55 f6f9f552014a:18 small references,10 interface checks,2 bypasses,first5201/2089 balanced replay,4actualfits. Laterutility/invariance/cost unrun after failedgate.", + "title": "Does fixed multinomial challenger55meet numerical,utility andcost qualification?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b4b31f43486d0ef2", + "event_kind": "finding", + "evidence_refs": [ + "experiments/multinomial_challenger_analysis.json", + "experiments/validation_multinomial_challenger.json", + "experiments/multinomial_challenger_plan.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_497445890ae118799d8a4a305a6a3b797a6e9dbd81e72880974800d47cd55f5d", + "recorded_at": "2026-10-15T10:39:28+00:00", + "result_id": "finding_409419c6079a5e884130509d799e23734219fcc50a18ea4def6e9e5c26cc7363", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: f6f9f552014a50474a418d4b29c751e0842950c41022e808d336f70d852d0c96 Applicability: Fixed55 f6f9f552014a:18 small references,10 interface checks,2 bypasses,first5201/2089 balanced replay,4actualfits. Laterutility/invariance/cost unrun after failedgate. Observation: 18small/10interface/2bypass pass. Firstbalanced Q+.01707,class+.01797,140predictions change;selectedrho2fits pass. Unselectedrho.02hits80steps;gradient1.894e-6 exceeds1e-6. Stop55;no invariance/cost/native.10unchanged. Interpretation: Refuted onlyforfrozen55qualification. Actualmanualgradient matches independentautograd;low-ridge fit reachesfixedcap,while18smallNewton/loss/gradientchecks pass. Selectedrho2train/finalfits pass. Onebalancedreplaygain suggestsprimaryheadroom,butnotfresh replication,fullutility,confidenceinvariance,cost ornative transfer. Do notdiscardalllikelihoodlosses,dropfailedrhoafterseeingresults,relaxstationarity orretry55withcap/tolerancegrid. Keepalltraces andbaseline diagnostics. Initialsupplied-validation bypassbug fixed todeclaredplan;originalsource/failure retained. Referenceautogradno_grad issue fixedwithoutcandidate/gate/timerchange. Replan a distinct conditioning/optimizer intervention versusanotherroute withcompletequality-cost coverage;10remainsbest and54nativecounterevidence remainsbinding.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed multinomial challenger55meet numerical,utility andcost qualification?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9a4b305ebb9f6afb", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T10:58:09+00:00", + "result_id": "question_98b27af7dec7acb85ee0cfab871fde2618442fb7ad6ca77e7d8603d72d354a79", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed56 c92986e1af1f versus55 f6f9f552014a;18small oracles,10interface,2bypasses,first5201/2089balanced replay and4actualfits. Local only;later qualification unrun.", + "title": "Do fixed initial-Hessian-diagonal coordinates56 qualify the likelihood challenger at unchanged gates?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "17684c9c79a921d6", + "event_kind": "finding", + "evidence_refs": [ + "experiments/preconditioned_multinomial_analysis.json", + "experiments/validation_preconditioned_multinomial.json", + "experiments/preconditioned_multinomial_plan.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_98b27af7dec7acb85ee0cfab871fde2618442fb7ad6ca77e7d8603d72d354a79", + "recorded_at": "2026-10-15T10:58:09+00:00", + "result_id": "finding_37edf4e05e02b46d5d22a21ebc5991814ed30adc0b877bfcd3340d9ca426c084", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: c92986e1af1fa29f84c6e38ff970f4dccc2ff80cc8205f79d5467a0f8a11b031 Applicability: Fixed56 c92986e1af1f versus55 f6f9f552014a;18small oracles,10interface,2bypasses,first5201/2089balanced replay and4actualfits. Local only;later qualification unrun. Observation: 18math/Hessian and10interface/2bypass checks pass. Fullrho.02 still hits80steps:originalrelativegradient7.694e-6 versus55at1.894e-6,limit1e-6. Selectedrho2 classes equal55;Qgain10+.01707 unreplicated. Stop56;no costs/native. Interpretation: Refuted onlyforfixed56qualification. Sparse duplicate/zero-column semantics,original-Hessian diagonal,coefficient mapping,chainrule andactual gradients match independent references. Both55/56low-ridge fits cap;fixed diagonal coordinates worsen this residual,not the selected classification. Selectedrho2train/finalfits pass;140test predictions differfrom10butexactlymatch55. Do notdrop theunselectedrho,relaxtruegradient/caps orrunscalegrid. Thisdoesnotrefuteallconditioning/likelihoodmethods,proveoff-diagonalcurvaturecause orqualifyNewton-CG. Preserve11unrun controls,freshbalancedreplication,8selectedinvariances,16ordinarypairs andnative asunproven. Replan a distinctcurvature-aware optimizer versusdifferentprimaryroute withfullmarginalcost;best10and54nativecounterevidence retained.", + "tags": [ + "writeback-result" + ], + "title": "Do fixed initial-Hessian-diagonal coordinates56 qualify the likelihood challenger at unchanged gates?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "91061489b9ee306b", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T11:20:36+00:00", + "result_id": "question_d5d4f9d965b4ef0f87558d3febecbf7259468756a8c8e3fea4c6aade4b98722d", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed57 b517da398307;12small references,12full5201/2089 controls,12train fits,10interface/2bypasses. Selected refit/invariance,cost andnative unrun.", + "title": "Does fixed pooled diagonal Gaussian discriminant57 qualify over native-best10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1778e1aa878d8ff1", + "event_kind": "finding", + "evidence_refs": [ + "experiments/pooled_discriminant_analysis.json", + "experiments/validation_pooled_discriminant.json", + "experiments/pooled_discriminant_plan.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_d5d4f9d965b4ef0f87558d3febecbf7259468756a8c8e3fea4c6aade4b98722d", + "recorded_at": "2026-10-15T11:20:36+00:00", + "result_id": "finding_0075e5588198b6da61e2a452dcbcbd380bf518bcfbce5a7b4f65676e77ad0db5", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: b517da398307117a43aea39e13e8ccbbbc0d5f73e4a60ea348a2581be04db1ad Applicability: Fixed57 b517da398307;12small references,12full5201/2089 controls,12train fits,10interface/2bypasses. Selected refit/invariance,cost andnative unrun. Observation: All moment references pass;all12 full controls reject57. Best validation proposal delta ranges -.22766 to-.10248. Exact10 fallback;mean quality andbothfreshbalanced classgains0. Stop57;no selected invariance,cost ornative. Interpretation: Refuted only for fixed57 qualification. Sparse duplicate aggregation,repeated-row multiplicity,missing classes,rank-deficiency,constant/empty features andempty targets pass independent two-pass/density references. All12 actual fits respect train ownership andignore poisoned outside labels;baseline search/calibrator input andfallback areexact10. Screening loses substantially on all8 replays and4fresh controls,so required primary replication/activation fails despite numerical correctness. This doesnotrefute all generative models or covariance estimators,norqualify selected known-row refits,8selected invariances,16whole-solve pairs ornative transfer. Keep55/56 distinct:one unreplicated balanced likelihood gain andunchanged convergence failures remain. Preserve54 native quality/time loss. Replan a distinct primary mechanism versus curvature-aware likelihood optimization;no fixed57 shrinkage,floor,prior,threshold orfixture grid. Both harness adapter errors retained;candidate,gates andoriginal25minute start unchanged.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed pooled diagonal Gaussian discriminant57 qualify over native-best10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "91e66b7005bc83b8", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T12:00:28+00:00", + "result_id": "question_7a5a65f26ac4e5a5dbff7cc9cd2c8c11486775fb6341b5355ae9f8864b181655", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed58 f73a29b55ab9;18small/HVP references,10interface/2bypasses,4full5201/2089replays and16actualfits.8remainingcontrols,invariance,cost/native unrun.", + "title": "Does fixed dynamic Newton-CG58 qualify all likelihood fits,primary utility andcost over10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3c752009272e6937", + "event_kind": "finding", + "evidence_refs": [ + "experiments/newton_cg_multinomial_analysis.json", + "experiments/validation_newton_cg_multinomial.json", + "experiments/newton_cg_multinomial_plan.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_7a5a65f26ac4e5a5dbff7cc9cd2c8c11486775fb6341b5355ae9f8864b181655", + "recorded_at": "2026-10-15T12:00:28+00:00", + "result_id": "finding_34ca13994395d3c420ee0f4b42c322c698c94a9ab48e72a1279a12015d56020f", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: f73a29b55ab97e88f1d335a5d5fc6ccb70b20eddea34ce4af317f6d64d1ebf58 Applicability: Fixed58 f73a29b55ab9;18small/HVP references,10interface/2bypasses,4full5201/2089replays and16actualfits.8remainingcontrols,invariance,cost/native unrun. Observation: 58solves55/56firstlow-ridge failure;4replayQ gains .01707/.10816/.15905/.14138.15/16fits pass;reversedrho.02innerCG caps60,residual.002569>.001 atB0,gradientratio1. HVP/diagonalrefs pass. Stop58;no costs/native. Interpretation: Refuted onlyforfixed58qualification. Independent original-gradient,HVP,diagonal andloss checks agree,includingfailedfit;true/recursive inner residuals differ2.26e-17. The failedtrain system takesnoNewtonstep after60CGiterations;unselectedrho cannotbedropped. Firstbalancedrho.02 nowgradient1.05e-9 versus55at1.89e-6/56at7.69e-6,while selectedclasses equal55/56. Preserve15converged fits,18small/duplicate/degenerate checks,10interface/2bypasses,ownership/poison controls andall4selectedclass/quality gains. These areexplicitreplays,notfreshbalancedreplication,8selectedinvariances,16ordinarycostpairs ornativetransfer. Do notextendcaps,relaxinner/outer gates orrunforcing/preconditioner/rho grids. ThisdoesnotrefuteNewton/likelihood universally orprovea causal spectrum/hotspot. Retain54native quality/time loss and57nonactivation. Replan a distinctprimary route versus independentlymotivated structuralcurvature usingfullmarginalcost. Builder-marker andanalysis-import failures areseparately retained;candidate/gates/start unchanged.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed dynamic Newton-CG58 qualify all likelihood fits,primary utility andcost over10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "24f9e0e93ee86c96", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T12:33:18+00:00", + "result_id": "question_5b8da9ce1a245862cb153477f33edf53962a8349c8ba02764a1de587aa5f65cc", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed59 14dd243baaf6;18math/6Adam,12full5201/2089controls and12trainfits,10interface/2bypasses. Selectedrefit/invariance,cost/nativeunrun.", + "title": "Does fixed32-tanh head59 qualify primary utility andcompletecost overnative-best10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "95afbfb15059bc63", + "event_kind": "finding", + "evidence_refs": [ + "experiments/fixed_mlp_analysis.json", + "experiments/validation_fixed_mlp.json", + "experiments/fixed_mlp_plan.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_5b8da9ce1a245862cb153477f33edf53962a8349c8ba02764a1de587aa5f65cc", + "recorded_at": "2026-10-15T12:33:18+00:00", + "result_id": "finding_7725e1cf12f44df269fd8f4fb52b91dbe93a8e1def811ff3c85b43b1ffcc226a", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 14dd243baaf62dd7c1e8e5cac85fd51f154280c8fbb659e22f09b5b428a0f88e Applicability: Fixed59 14dd243baaf6;18math/6Adam,12full5201/2089controls and12trainfits,10interface/2bypasses. Selectedrefit/invariance,cost/nativeunrun. Observation: Allnumeric/trainingchecks pass,butall12controls reject59. Bestvalidation classdeltas -0.20234to-0.07375;exact10fallback,meanQ/bothfreshbalancedgains0. Noqualifyingselectedcase. Stop59;nocost/native. Interpretation: Refutedonlyforfixed59qualification. Independentforward/loss/analyticgradient,2directionalFD,6Adamparameter/momentalgebra andallactual120-stepreferences agree. Actualfits ignorepoisonedoutside labels,resetstates,anddecreaseobjective. Canonical label-freegraphcolumns/word/degreeconstruction hasexactpositivecontrols;original10helpers/search/fallback/calibrator remainverified. All8replays and4freshcontrols losevalidationclassification,includingbothfreshbalancedseeds;noactivationorfreshprimarygain. Selectedknownrefits,8selectednonflatinvariances,16ordinarycostpairs andnative transferremainunproven;fallbackequalitydoesnotprove59invariance. Stopwithout width/seed/lr/epoch/L2/activation/threshold/fixturegrid. ThisdoesnotrefuteallMLPs,certifyaglobaloptimum orproveoverfitcause. Preserve55/56/58numericalcounterevidence and54nativequality/time loss. Replan distinctprimarymechanismsversus independentlymotivatedstructuralconditioning withfullmarginalcost.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed32-tanh head59 qualify primary utility andcompletecost overnative-best10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "24ededaa99c27c87", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T12:55:38+00:00", + "result_id": "question_1492e1dc25bd1183f1a2b1a3824da7308e4cab7535e2009d3cf1048ba410c47f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed60 c4086b708b07;18small block/HVP/Newton references,10interface/2bypasses,1full5201/2089replay and4actualfits.11remainingcontrols,invariance,cost/native unrun.", + "title": "Does fixed class-block Newton60 qualify all likelihood fits,primary utility andcomplete cost over10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "aef2d63611b2d390", + "event_kind": "finding", + "evidence_refs": [ + "experiments/class_block_newton_analysis.json", + "experiments/validation_class_block_newton.json", + "experiments/class_block_newton_plan.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_1492e1dc25bd1183f1a2b1a3824da7308e4cab7535e2009d3cf1048ba410c47f", + "recorded_at": "2026-10-15T12:55:38+00:00", + "result_id": "finding_ba55ad11dd9e5541810f526590574cf01d523943fc55d6609b1d82527761e192", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: c4086b708b070b473f8a6774a6bda7c80534d46e4ba8cb6fd9a9c9c25c72711c Applicability: Fixed60 c4086b708b07;18small block/HVP/Newton references,10interface/2bypasses,1full5201/2089replay and4actualfits.11remainingcontrols,invariance,cost/native unrun. Observation: 60 repairs58 trainfailure:42innersteps,gradient2.92e-9. Selectedknownrho.02 caps60 atresidual .001444,gradient7.85e-6.18small/4actualreferences pass;3fits converge;replayQ+.14236. Stop60;no cost/native. Interpretation: Refuted only for fixed60 qualification. Same source/input/prior,bitwise unchanged diagonal andold58 trace establish initial correspondence. Exact class blocks reduce the original train first solve from a failed60-step residual .002569 to42steps at.0008245;trainfit converges. Newly selectedrho.02 known3440-target fit takes7Newtonsteps thenfails its8th inner solve atunchanged60cap;true/recursive residuals agree within2.71e-16. Independent dense/autograd class blocks,HVPs,Cholesky applications,original gradients andprobabilities agree,including failedfit;ownership andpoison controls pass. Validation selection/calibrator andoriginal10search are independently verified. One replay gain is not fresh replication,numerical qualification,8selected invariances,16whole-solve pairs ornative transfer.58 had selectedrho.2 andnever ran thisknownrho.02 system;do notclaim60 worsens anunobserved58fit. Stop without block/cap/forcing/rho/jitter grid. Retain59 nonactivation and54 native quality/time loss;require fresh semantic comparison ofprimary alternatives andindependently justified curvature withfull cost.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed class-block Newton60 qualify all likelihood fits,primary utility andcomplete cost over10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "cbe60cfb64e98052", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T13:22:35+00:00", + "result_id": "question_5eaebf5f1fc8152af72e33574434ab0acaabf7e2e5aa4841f5aeec8c2251e9a9", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed61 995adb9505b3;18small optima,6kinks,8interface smokechecks and3direct110-target systems. No full5201controls,bypasses,selectedinvariance,cost/native.", + "title": "Does fixed one-vs-rest squared-hinge61 qualify all fits,primary utility andcomplete cost over10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fda3b4cc59a31951", + "event_kind": "finding", + "evidence_refs": [ + "experiments/squared_hinge_analysis.json", + "experiments/validation_squared_hinge.json", + "experiments/squared_hinge_interface_analysis.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_5eaebf5f1fc8152af72e33574434ab0acaabf7e2e5aa4841f5aeec8c2251e9a9", + "recorded_at": "2026-10-15T13:22:35+00:00", + "result_id": "finding_92cb60dcdaba7e6cc34ef861ff8316d59dd411cf4977a994ddf61d86e36387eb", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 995adb9505b33623aa77be115361a3130aa1ef73765ee97ee09ac697d5747f63 Applicability: Fixed61 995adb9505b3;18small optima,6kinks,8interface smokechecks and3direct110-target systems. No full5201controls,bypasses,selectedinvariance,cost/native. Observation: 18small/6kinks pass. Exactinterface rho.02 caps60 inthirdinner solve:residual .001716>.001,gradient .04283>1e-6,probabilityerror .31067. Independent gradients/HVPs/residuals pass;2higher-rho probes converge. Stop61;no fullcontrols/native. Interpretation: Refuted only for fixed61 qualification. Independent exhaustive active-set optima,margin-boundary,duplicate/repeated-row/zero-column anddegenerate references pass. Eightinterface smokechecks do not prove selected61 invariance oralltheirfitstationarity. Exactseeded replay preserves RNGconsumption andtheoriginal .3106704375 failure. On110targets/217features,rho.02 takes2Newtonsteps thenthirdinner solve caps60;oneclass residual .001716,true/recursive gap9.07e-16. Independent dense loss/gradient,active masks,HVPs,diagonal andeveryinner true residual agree;outside-target poison isexact. Twohigher-rho systems converge. No thirdNewtonstep accepted. This is scoped solvercounterevidence,not universal margin refutation,primary losscomparison ornative transfer. Remaining2interface checks,explicit2bypasses,12fullcontrols/freshprimaryreplication,8selected transforms,16whole-solvepairs andnative remainunproven. Keep58/60partialrepairs,59nonactivation and54native loss. Replan beforeanotherdistinctroute;no61cap/rho/margin/forcing/threshold/fixturegrid.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed one-vs-rest squared-hinge61 qualify all fits,primary utility andcomplete cost over10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "afae8079d5812952", + "event_kind": "node", + "goal_id": "graph-warm-on-20261015-02", + "node_kind": "question", + "recorded_at": "2026-10-15T13:45:49+00:00", + "result_id": "question_140ba66b10a0a6e7c08dbd2ca56e5e23c2dba6f7b2721bce9cf7865e5c2e995b", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Fixed62 461a644f634f;saved61 failed coefficient andone110-target fitfromzero,5outer systems/25class factorizations. No full-size utility/cost/native.", + "title": "Does fixed active-target kernel squared-hinge62 qualify every fit,primary utility andcomplete cost over10?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "48d84491f7c8af81", + "event_kind": "finding", + "evidence_refs": [ + "experiments/active_kernel_hinge_analysis.json", + "experiments/validation_active_kernel_hinge.json", + "experiments/active_kernel_hinge_plan.json" + ], + "goal_id": "graph-warm-on-20261015-02", + "node_id": "question_140ba66b10a0a6e7c08dbd2ca56e5e23c2dba6f7b2721bce9cf7865e5c2e995b", + "recorded_at": "2026-10-15T13:45:49+00:00", + "result_id": "finding_7296be0092dfc1265511f4590ca04fb5dcede17569185ac13050c86c5f24ce0e", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 461a644f634fb249806a60ecd03a5578992df0de46a38150bf893b0e0def4e12 Applicability: Fixed62 461a644f634f;saved61 failed coefficient andone110-target fitfromzero,5outer systems/25class factorizations. No full-size utility/cost/native. Observation: Saved61 endpoint passes independent Gram/primal checks.62fromzero accepts4steps;fifth fails only solvedclass4:gradient~1e-14,relative residual~1,Gram~6e-16. Wholefit gradient.01723,probabilityerror.06533. Stopfixed62;othercontrols unrun. Interpretation: Refuted only for fixed62 qualification under unchanged every-nonzero-class forcing. At61 failed coefficient,forcing1.32e-13 andprimalendpoint error5.71e-14 establish scoped repair. Newfit Gram/system/factor errors<=1.78e-15,dense primal endpoints<=1.18e-13;independent gradients/HVPs agree. On fifth solve,class4 mask/system/rhs/factor/dual repeat bitwise andpreviousendpoint matchescurrentB to roundoff;B-endpoint cancellation leaves tiny direction against roundoff-scale nonzero gradient. Otherclasses remain nonstationary;no failedstepaccepted. Absolute residual near1e-14 doesnotpass fixedrelative1e-3gate. This doesnot refute exact endpoint identity,certify wholefit convergence orprove primaryutility/native gains. Outside-target poison exact. Twohigher-rho positives,18small/6newboundaries,10interfaces/2bypasses,12fullcontrols,8selectedchecks and16costpairs unrun. No gate/solver/rho/fixture grid;freshsemanticreplan distinguishes numericaltermination fromprimary alternatives. Keep58/60partialrepairs,59nonactivation and54native loss.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed active-target kernel squared-hinge62 qualify every fit,primary utility andcomplete cost over10?" + } + ] +} diff --git a/apps/presentation/site/public/lab/time-machine/data/portfolio.json b/apps/presentation/site/public/lab/time-machine/data/portfolio.json new file mode 100644 index 0000000000..2a7a2cbf88 --- /dev/null +++ b/apps/presentation/site/public/lab/time-machine/data/portfolio.json @@ -0,0 +1,1952 @@ +{ + "description": "Unchanged Explore history from a real, sealed research checkpoint later restored by follow-up experiments. Original text, timestamps, identities and references; no invented relationships.", + "source_sha256": "b876dd7660e0aab19db717f60e86865494a74b8fa27b42a5a67ee295ef524397", + "source_bytes": 104922, + "goal_id": "portfolio-warm-on-20261015-01", + "display_name": "Portfolio · 近期研究 checkpoint", + "provenance": { + "case_id": "portfolio_risk_calibration", + "source_run_id": "portfolio-warm-stride-on-20261015-01", + "checkpoint_id": "7fa63b57b1d3472496ed917d3acdba91", + "checkpoint_sha256": "2a05eea1753ea269bcfdfcf14cc5a4fc796bfad735d1adf0e99d889343ea1313", + "captured_at": "2026-10-15T13:45:34.711371+00:00", + "captured_native_turns": 58, + "selection": "Use Portfolio58-turn pre-replan point after its full real receiver passed; newer65-turn point timed out on repeated native thread/turns/list during fresh preparation. Select by recovery quality, not score. Preserve newer-point failure evidence.", + "restored_by": [ + "portfolio-replan-fresh-20261016-02", + "portfolio-replan-retained-20261016-02" + ], + "redactions": [], + "claim_scope": "diagnostic_only", + "score_countable": false, + "trajectory_publication": { + "status": "uploaded", + "repository": "https://huggingface.co/datasets/huangrt01/loopx-edgebench-research-checkpoints-20261015", + "revision": "c858f51021dc1c1cf08263f9c04b692a5e29b40b", + "archive": "portfolio_risk_calibration.zip", + "archive_sha256": "fabc92479269b5f08eac0ee10f4a89aba4f75efef27f06e2b6b6d981d4bb4612" + }, + "boundary": "Captured Explore knowledge only; original native execution state, credentials, evaluator and models are not imported." + }, + "source_counts": { + "event_count": 76, + "event_kinds": { + "node": 38, + "finding": 38 + }, + "node_kinds": { + "question": 38 + }, + "statuses": { + "open": 32, + "refuted": 32, + "confirmed": 6, + "resolved": 6 + }, + "first_recorded_at": "2026-10-14T20:30:26+00:00", + "last_recorded_at": "2026-10-15T13:44:53+00:00", + "edge_count": 0, + "parent_records": 0 + }, + "events": [ + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8e2fabc9b15a87f1", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T20:30:26+00:00", + "result_id": "question_97e36b6161239743d643c33214e574370f75f5e9b470341790869ed4c4800a83", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:8e17428b37a871b8", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2024 cold start; forecast-v4; one-session variance QP; inertia0/0.00025/0.0005/0.001; exact fees/caps.", + "title": "Do the tested cost-gated one-session executions improve the frozen forecast strategy?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "c595691adcd20d3f", + "event_kind": "finding", + "evidence_refs": [ + "evidence:opportunity-cost-gates" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_97e36b6161239743d643c33214e574370f75f5e9b470341790869ed4c4800a83", + "recorded_at": "2026-10-14T20:30:26+00:00", + "result_id": "finding_3cca24bb14a9ed88c330629d36574282fa05ea255c47319caf58042f4ecf5f22", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:48c383ed164d960e", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 3bd3bcab016b03ff1b46c6f0ad7238f1d86a022f9953fcf79893b647f2e6cfc3 Applicability: Public2024 cold start; forecast-v4; one-session variance QP; inertia0/0.00025/0.0005/0.001; exact fees/caps. Observation: Public2024 control Sharpe3.30/turnover145%; daily trials1.53-2.99/199-719%, with ledger, caps and model clocks valid. Chronological comparisons support retaining the existing risk/prior settings. Interpretation: Outcome: continue retained: - Forecast-v4, exact ledger and risk clocks; native incumbent and open validation Todo. stopped: - Tested daily opportunities and direct-Sharpe defaults; additional trend-risk rule.", + "tags": [ + "writeback-result" + ], + "title": "Do the tested cost-gated one-session executions improve the frozen forecast strategy?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8863964644ccf55c", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T20:54:55+00:00", + "result_id": "question_1ae72ed2059bc3f1b2812396832d04b3707791e72cd99cca5c2b01157ae13383", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:3f7a8fdd1434c64e", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2022-2024 cold starts; per-ETF ridge100; recent2yr/EWMA252; signal0.4; V3/V4 cost policies.", + "title": "Do recent training windows improve robust forecast portfolio Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f9133a99774fbf96", + "event_kind": "finding", + "evidence_refs": [ + "evidence:forecast-recency" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_1ae72ed2059bc3f1b2812396832d04b3707791e72cd99cca5c2b01157ae13383", + "recorded_at": "2026-10-14T20:54:55+00:00", + "result_id": "finding_1ad4affdeb26e0b2eb765c6f051ca32c1b02aae69a9c868f64954f8abba0def9", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:19d84bac4c37ff14", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 3e0f74241be0f443dffe39ce3e6d3ccf0fc388991d6340fd4db23990673f811e Applicability: Public2022-2024 cold starts; per-ETF ridge100; recent2yr/EWMA252; signal0.4; V3/V4 cost policies. Observation: Three-fold mean Sharpe falls under both cost policies: expanding1.48/1.89,recent0.71/1.05,EWMA1.22/1.42.30 valid cases; production unchanged. Interpretation: Outcome: continue retained: - Native-best V3, exact ledgers/costs/clocks; full acceptance. stopped: - Tested two-year/EWMA252 ridge and price-trend defaults.", + "tags": [ + "writeback-result" + ], + "title": "Do recent training windows improve robust forecast portfolio Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3f0efb61862a9c31", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T21:37:18+00:00", + "result_id": "question_1cc6be5c16dc8e983dc2689711a727f15ccd08583cbdd2f3d5061437b63b2803", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:62f2fa63ac7c994a", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; ridge100 fit only earlier years; gamma40, decay60, blend0.4, V4 execution; original costs/caps.", + "title": "Do5/20-session return targets raise fivefold mean portfolio Sharpe under decay-v6?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "12f6fbeb6263e35a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:forecast-horizons" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_1cc6be5c16dc8e983dc2689711a727f15ccd08583cbdd2f3d5061437b63b2803", + "recorded_at": "2026-10-14T21:37:18+00:00", + "result_id": "finding_b862f54f096483725e319ddf9e8d1c4900c868221f7a098813b843d5444194b1", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:5a87324c027c48ec", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: fc66847579409493603bdfef855d366f3ec3ecc76c1073788bcb958120a49dac Applicability: Public2020-24 cold starts; ridge100 fit only earlier years; gamma40, decay60, blend0.4, V4 execution; original costs/caps. Observation: Mean Sharpe falls from1.77 at horizon1 to1.20/1.33 at5/20; worst0.49 to-0.12/0.21. Ten strategies pass ledgers, caps, events and exact80-session prefixes. Interpretation: Retain horizon1 within this recipe. Some individual folds improve; this is scoped negative mean-Sharpe evidence, not a native or universal result.", + "tags": [ + "writeback-result" + ], + "title": "Do5/20-session return targets raise fivefold mean portfolio Sharpe under decay-v6?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1325576bc92a9f8d", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:01:12+00:00", + "result_id": "question_ea37a63eb08965f37555551f29baa50def325120ac37c52d72fd8d520ceec352", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:eb5f893ca9c35a5a", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; V3 gamma40; earlier-year winsorized means, HAC20, global/class pooling; original costs/risk/hedges.", + "title": "Do the tested uncertainty-weighted priors beat the fixed-half blend on fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7ca24279d09e4809", + "event_kind": "finding", + "evidence_refs": [ + "evidence:uncertain-priors" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_ea37a63eb08965f37555551f29baa50def325120ac37c52d72fd8d520ceec352", + "recorded_at": "2026-10-14T22:01:12+00:00", + "result_id": "finding_99c1d552520eda88c201f286543b0bf05fd7e6fd2834a3575458d9a2b739d952", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:c6c7ee8a5291e92f", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 38861559af1b30cf9700f45d7dd9966e3f18b041d5d1fe0652b80c5e39c4cc6c Applicability: Public2020-24 cold starts; V3 gamma40; earlier-year winsorized means, HAC20, global/class pooling; original costs/risk/hedges. Observation: HAC20 global/class priors yield mean/worst Sharpe1.313/0.247 and1.333/0.052 vs fixed-half1.394/0.402. All20 ledgers/caps/events/prefixes pass. V7 passes three suites; mean return falls0.41pp vs V3. No native transfer claim. Interpretation: Outcome: replan retained: - Native-best V3 and all original costs, clocks, caps, risk, shorts and target priorities. changed: - Prepared fixed-half V7 separately; investigate which class drives its modest public benefit. stopped: - Adopting the two tested uncertainty-weighted prior estimators.", + "tags": [ + "writeback-result" + ], + "title": "Do the tested uncertainty-weighted priors beat the fixed-half blend on fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8b98f6b538f5c94d", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:08:36+00:00", + "result_id": "question_661547742b965975f7ee06c77f08526c26486792b08e869718518eeeca9e588b", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:babb66b2893d9be6", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; V3; negative20-session equity mean gates50% shrinkage, with/without Commodity half; earlier-year priors and original costs/clocks.", + "title": "Does tested conditional equity-prior shrinkage beat fixed-half mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "17027503bda9eb5c", + "event_kind": "finding", + "evidence_refs": [ + "evidence:class-prior-followup" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_661547742b965975f7ee06c77f08526c26486792b08e869718518eeeca9e588b", + "recorded_at": "2026-10-14T22:08:36+00:00", + "result_id": "finding_51ab57b26c0b7207894b75da767467ec30bf38a5d7adc76d86d895cc982eecb1", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:019e8610fecfa0a0", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: d1c692b810ae55a1b46847417a3bb7ae19da7e733a66209ad510df01f590c59f Applicability: Public2020-24 cold starts; V3; negative20-session equity mean gates50% shrinkage, with/without Commodity half; earlier-year priors and original costs/clocks. Observation: Conditional rules yield mean/worst Sharpe1.316/-0.131 or1.337/-0.038 vs fixed-half1.394/0.402. Ninety strategies pass contracts/prefixes. V7 gain is mostly2022 equity shrinkage; other-year mean gain is negative. No native conclusion. Interpretation: Outcome: continue retained: - Native-best V3, exact V7 and all original costs, risk, hedge, causal and performance requirements. changed: - Investigate opening exposure versus later shrinkage; recognize fragile V7 local support. stopped: - Adopting the tested negative20-session conditional-prior rules.", + "tags": [ + "writeback-result" + ], + "title": "Does tested conditional equity-prior shrinkage beat fixed-half mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "270e1bf6a8522a5a", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:14:27+00:00", + "result_id": "question_9394687904fa9a731906bfb317764b85afdc6b0fe6c1abb99fb96b275aacac5f", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:d2e6708e60b47ae9", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; V3 risk/hedge/cost rules; earlier-year priors; fixed20-session opening/later half-blend factorial.", + "title": "Does opening shrinkage dominate V7 fivefold mean Sharpe benefit?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7bed4c73953ae985", + "event_kind": "finding", + "evidence_refs": [ + "evidence:prior-windows" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_9394687904fa9a731906bfb317764b85afdc6b0fe6c1abb99fb96b275aacac5f", + "recorded_at": "2026-10-14T22:14:27+00:00", + "result_id": "finding_f65d6147cd606e3144efc113b0d3f1dc2aedb6c1f3a3d94f03070569f3917dcf", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:7846d33c5bd58bcc", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 491df0a2eb51d28abf42c4e1ef25262b88deda30641e04f3753409bd68e2a34c Applicability: Public2020-24 cold starts; V3 risk/hedge/cost rules; earlier-year priors; fixed20-session opening/later half-blend factorial. Observation: Mean opening/later Sharpe effects are-0.036/+0.116. Delayed V8 beats V3 in4/5 years with mean1.426, return2.93%, turnover114.6%; worst0.080 trails V7=0.402. All20 comparisons and V8 three suites pass. No native transfer claim. Interpretation: Outcome: replan retained: - Native-best V3, exact V7 and all original costs, risk, hedge, causal and performance targets. changed: - Qualified exact V8 separately; prioritize cold-start robustness evidence over more prior tuning. stopped: - Treating opening shrinkage as the chief fivefold mean Sharpe benefit.", + "tags": [ + "writeback-result" + ], + "title": "Does opening shrinkage dominate V7 fivefold mean Sharpe benefit?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "564a6f8df4105d14", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:21:13+00:00", + "result_id": "question_aac0a391ac54367e29a0ab57c007762d64dc98d5b021f1dac88c55e8716c3a64", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:3aaa020bd9f87035", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Seventeen public242-session starts2020Q1..2024Q1; exact V3/V7/V8; priors fit before start year; original costs/caps/risk.", + "title": "Does V8 exceed V7 mean and worst Sharpe on the quarterly cold starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "774a4d8f0c556de5", + "event_kind": "finding", + "evidence_refs": [ + "evidence:quarterly-cold-starts" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_aac0a391ac54367e29a0ab57c007762d64dc98d5b021f1dac88c55e8716c3a64", + "recorded_at": "2026-10-14T22:21:13+00:00", + "result_id": "finding_f42022e6f57f2aab86aaaf92b3bf968a0d7b5f5545204f2576522345f4ce7194", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:13f19423d407a5ca", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 0ba8f2b7ed7d726b1df2ee7362195163860c2a1becffbca90c5defcb242a2da7 Applicability: Seventeen public242-session starts2020Q1..2024Q1; exact V3/V7/V8; priors fit before start year; original costs/caps/risk. Observation: V7 mean/worst Sharpe1.506/0.031 exceeds V8=1.450/-0.229 across17 overlapping242-session starts. V8 wins8/17 vs V7, retains more return and better reported-VaR range coverage. All51 ledgers/caps/events/prefixes pass. No native transfer claim. Interpretation: Outcome: replan retained: - Native-best V3, exact V7/V8 and all original costs, risk, short, causal and performance requirements. changed: - Prioritized V7 for next eligible evaluation; preserve V8 as an alternative with different tradeoffs. stopped: - Assuming V8 mean-Sharpe superiority from January-only folds.", + "tags": [ + "writeback-result" + ], + "title": "Does V8 exceed V7 mean and worst Sharpe on the quarterly cold starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ef2c0279ff7d432b", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:29:36+00:00", + "result_id": "question_b95b950c28c113a5fb96a3526d034e840a4c6ab9fa2d74c24189bb77aed436a5", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:675ab2db4134469f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "17 public242-session quarterly starts; prior weights, samplecov60, normal95%, rounding6; ridges1e-8/1e-10 and missing/zero first-return conventions.", + "title": "Does V7 improve3-7% proxy coverage over V3 under all four declared definitions?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fce9a9ac3b680633", + "event_kind": "finding", + "evidence_refs": [ + "evidence:public-var-proxy" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_b95b950c28c113a5fb96a3526d034e840a4c6ab9fa2d74c24189bb77aed436a5", + "recorded_at": "2026-10-14T22:29:36+00:00", + "result_id": "finding_bd77c5591f823c1c4036f21130b539a6e1f81a4284d7fac8359f8d3088bfe9ad", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:fd458d83057f4995", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 48d87fe77cc2e8e74b6ea3d61165f0450146238737ac0f98f111ad44aa44c9a1 Applicability: 17 public242-session quarterly starts; prior weights, samplecov60, normal95%, rounding6; ridges1e-8/1e-10 and missing/zero first-return conventions. Observation: Public3-7% proxy coverage is V7=16/17 vs V3=14/17 and V8=17/17 under all four declared definitions. All51 replay/causal/cost checks pass. Server history has four completed submissions, best17.24. Exact V7 attempt is prepared, not submitted. Interpretation: Outcome: continue retained: - Native-best V3, exact V7/V8 and every original target, cost, risk, hedge and causal contract. changed: - Prepared V7 one-shot ID retention; investigate online forecast adaptation before eligibility.", + "tags": [ + "writeback-result" + ], + "title": "Does V7 improve3-7% proxy coverage over V3 under all four declared definitions?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8371089e0544c41b", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:42:26+00:00", + "result_id": "question_1048561881782711f80a6e1ee7dd4329ecca6a1273dcc5c453b2dd877fc84712", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:0b739c89cc4b69d9", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; frozen8-feature V4 prior, ridge100 EWLS corrections with60-session half-life, completed targets clipped+-3; same V4 costs/risk/constraints.", + "title": "Does the tested online coefficient updater improve fivefold mean and worst Sharpe over frozen forecasts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0a62dd2c63cfd324", + "event_kind": "finding", + "evidence_refs": [ + "evidence:online-coefficients" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_1048561881782711f80a6e1ee7dd4329ecca6a1273dcc5c453b2dd877fc84712", + "recorded_at": "2026-10-14T22:42:26+00:00", + "result_id": "finding_94a0f92d2342af661f0d634edacb7554dd9a731c99fb028d98c0366ee93525e7", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:d855d9936cddcb35", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: fcb61f195190501ca7a0bafb1ca1582228098c1d26636504083e0538e2ebf255 Applicability: Public2020-24 cold starts; frozen8-feature V4 prior, ridge100 EWLS corrections with60-session half-life, completed targets clipped+-3; same V4 costs/risk/constraints. Observation: Online EWLS mean/worst Sharpe1.727/0.124 trails frozen1.821/0.441; turnover146% vs156%. All10 physical cases and prefixes pass; matched controls repeat exactly. Interpretation: Outcome: continue retained: - Native-best V3, exact V7/V8 and all original acceptance. stopped: - Adopting the tested ridge100/half-life60 online coefficient updater.", + "tags": [ + "writeback-result" + ], + "title": "Does the tested online coefficient updater improve fivefold mean and worst Sharpe over frozen forecasts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b63bc6cb0901c0b9", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:49:24+00:00", + "result_id": "question_2ba2bdeeca0da1fa1b56ae44d1fe9ee00cf84f3f07276d9752cdb63461989486", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:9951c915ce90d10a", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact prior-v7 versus risk-v3 on the same task: native submissions b36facd55fbf/3b4ab368dff2; score-first selection.", + "title": "Does exact V7 exceed retained V3 on native score?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d385b756d9c547c5", + "event_kind": "finding", + "evidence_refs": [ + "evidence:native-v7-adoption" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_2ba2bdeeca0da1fa1b56ae44d1fe9ee00cf84f3f07276d9752cdb63461989486", + "recorded_at": "2026-10-14T22:49:24+00:00", + "result_id": "finding_3ceec468e2f64be66172090b00f08d76482ff40070b458b7a3ab546cc1d616d7", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:bb801afd94ed7bec", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 0a805547afcc26cc5e5bd2b6b88697749758df850d0eb6bba6da39c3b97865be Applicability: Exact prior-v7 versus risk-v3 on the same task: native submissions b36facd55fbf/3b4ab368dff2; score-first selection. Observation: V7 score20.53 exceeds V3=17.24, but Sharpe/excess regress. Public V7 class allocation explains-15.15pp, selection+0.75pp and costs-0.69pp of excess; signed equity exposure averages-1.72%. Interpretation: Outcome: replan retained: - Exact V3/V7/V8 checkpoints, original priority targets and all physical/causal constraints. changed: - Adopted exact V7 by native score-first policy. - Investigate allocation exposure with a matched objective intervention.", + "tags": [ + "writeback-result" + ], + "title": "Does exact V7 exceed retained V3 on native score?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "aa35a878a0328b1b", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T22:56:44+00:00", + "result_id": "question_8888aa6d82f796d2ed32a7210e159ce5d12a2ab37b561151f1b8e0cc64b3b020", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:0d710684645ddc49", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; V7 prior/risk/cost rules, gamma40; penalty gamma*equity-class variance*(net equity-0.10)^2; one unconditional target.", + "title": "Does the fixed0.10 soft net-equity objective improve V7 fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "389eedde0dbaa4e1", + "event_kind": "finding", + "evidence_refs": [ + "evidence:soft-equity-objective" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_8888aa6d82f796d2ed32a7210e159ce5d12a2ab37b561151f1b8e0cc64b3b020", + "recorded_at": "2026-10-14T22:56:44+00:00", + "result_id": "finding_cc5357649f60ab011a23d909989a58d994f3ec112f53057e4781bd1bd633180a", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:672571aca5531184", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 570e3494e0154d41fcfba64b35f61c70086b49235c904f5206c728c8ef69dfb8 Applicability: Public2020-24 cold starts; V7 prior/risk/cost rules, gamma40; penalty gamma*equity-class variance*(net equity-0.10)^2; one unconditional target. Observation: Soft target raises mean return2.67% to2.97% and net equity2.1% to7.3%, but mean/worst Sharpe1.384/-0.001 trail V7=1.394/0.402. All10 physical/prefix checks pass; disabled controls match unmodified V7. Interpretation: Outcome: continue retained: - Exact native-best V7, V3/V8 checkpoints and all original targets/constraints. stopped: - Adopting the tested unconditional0.10 equity soft target.", + "tags": [ + "writeback-result" + ], + "title": "Does the fixed0.10 soft net-equity objective improve V7 fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "9bc2eb472164b2dc", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:03:48+00:00", + "result_id": "question_959c8a5d07c0f5390c238f97e60d1e99680a095e383b0f5f06864cfb217c2e1f", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:70e2806656ed9579", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; gamma40/V7 rules;0.10 soft target after20 observations only if trailing60-session Equity mean is positive; same costs/risk/event clocks.", + "title": "Does the tested trend-conditioned equity objective improve both V7 fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d41adc9292ff0419", + "event_kind": "finding", + "evidence_refs": [ + "evidence:conditional-equity-objective" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_959c8a5d07c0f5390c238f97e60d1e99680a095e383b0f5f06864cfb217c2e1f", + "recorded_at": "2026-10-14T23:03:48+00:00", + "result_id": "finding_e66773b72a056f22316cb2daf23ddfbe22d1ad68fa4a59a71383283578eb0284", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:276a62106c8860b6", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 3ea1a36b6a08367bba5411717d366adbf43180ec693fb84157e461ad403d62fd Applicability: Public2020-24 cold starts; gamma40/V7 rules;0.10 soft target after20 observations only if trailing60-session Equity mean is positive; same costs/risk/event clocks. Observation: Conditional mean/worst Sharpe1.431/0.164 versus V7=1.394/0.402 and unconditional1.384/-0.001.2024 dominates gains; all15 physical/policy prefixes pass and controls reproduce exactly. Interpretation: Outcome: continue retained: - Exact native-best V7, V3/V8, both prototypes and all original constraints/targets. changed: - Preserved conditional prototype; require broader start-date evidence before integration.", + "tags": [ + "writeback-result" + ], + "title": "Does the tested trend-conditioned equity objective improve both V7 fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ef80a49649bcead6", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:10:54+00:00", + "result_id": "question_83fbf99da875487b456b4e30b55e3ae6b6afd003ebf1510dc3feb91bbcdeea24", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:0ab90e890df5a57f", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public17 matched242-session starts2020Q1..2024Q1; earlier-year priors; unchanged20/60 positive-trend0.10 soft target and V7 risk/cost/short rules.", + "title": "Does the fixed conditional objective improve V7 mean and worst Sharpe on17 quarterly cold starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e059a7984eed298d", + "event_kind": "finding", + "evidence_refs": [ + "evidence:conditional-quarterly-starts" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_83fbf99da875487b456b4e30b55e3ae6b6afd003ebf1510dc3feb91bbcdeea24", + "recorded_at": "2026-10-14T23:10:54+00:00", + "result_id": "finding_0b5bf181cb7cc89b92e287930db16c22d3afce7eccd7edd375407e1cf1c93c35", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:dca4262e34a8b727", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 22c6df4ab8b542dac51874fa75995f95b254ccaa2ca374d0c31b600ed40d1b27 Applicability: Public17 matched242-session starts2020Q1..2024Q1; earlier-year priors; unchanged20/60 positive-trend0.10 soft target and V7 risk/cost/short rules. Observation: Across17 overlapping starts, conditional wins7; mean/worst Sharpe1.492/-0.171 versus V7=1.506/0.031. Mean return/year-balanced Sharpe improve. All34 physical and policy prefixes pass. Interpretation: Outcome: replan retained: - Native-best V7, V3/V8, both objective prototypes and original acceptance. changed: - Test ETF-selection alpha with fixed zero-sum relative-return adjustments. stopped: - Preferring the conditional objective from January-only gains or tuning more exposure gates.", + "tags": [ + "writeback-result" + ], + "title": "Does the fixed conditional objective improve V7 mean and worst Sharpe on17 quarterly cold starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "918f2e760df20b54", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:18:44+00:00", + "result_id": "question_408c184bfc320a896f3449314bcb25d39d92bd0e09f5570ed3de930df1a344c5", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:516971a15df92fc7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; fixed last20 nonhedge Equity mean, demeaned, coefficient+/-0.25, uniform maxabs0.0005 cap after20 observations; V7 risk/cost/short rules.", + "title": "Does either fixed within-Equity relative-return signal improve both V7 fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "25a1bcc2c66bc4b9", + "event_kind": "finding", + "evidence_refs": [ + "evidence:relative-equity-signals" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_408c184bfc320a896f3449314bcb25d39d92bd0e09f5570ed3de930df1a344c5", + "recorded_at": "2026-10-14T23:18:44+00:00", + "result_id": "finding_0f1bb30d6070787f07ff67c502db99c29c235c0984047695bfce7b886a322399", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:d0954ae997fcea73", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 66c351c9c276a3b64f06a7d824709dd6eb9f5f3f363cfc00ead444b9d7ec7058 Applicability: Public2020-24 cold starts; fixed last20 nonhedge Equity mean, demeaned, coefficient+/-0.25, uniform maxabs0.0005 cap after20 observations; V7 risk/cost/short rules. Observation: Fivefold mean/worst Sharpe: V7=1.394/0.402; momentum=1.256/0.096; reversal=1.409/0.277. Both improve selection attribution but raise costs/exposure; reversal gains rely on2024. All15 physical and exact80 prefixes pass. Interpretation: Outcome: replan retained: - Native-best V7, earlier checkpoints/prototypes and every original target. changed: - Test fixed frozen-forecast/prior-shrinkage complementarity under V7 execution. stopped: - Adopting either tested relative-return signal or tuning its coefficients.", + "tags": [ + "writeback-result" + ], + "title": "Does either fixed within-Equity relative-return signal improve both V7 fivefold mean and worst Sharpe?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "080537eda2dff2b1", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:33:39+00:00", + "result_id": "question_ed894b182109bf3ea0537dc3bfbaba71f1fc770f4e67c098215f16703eaa49b5", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:34d2628a9ff5f1a1", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public17 overlapping242-session starts2020Q1..2024Q1; prior-year fits, prior blend0.5/forecast0.4, retained V4 inertia0.02/deadband0.004, original V7 risk/cost/short rules.", + "title": "Does the fixed half-prior/frozen-forecast hybrid improve V7 mean and worst Sharpe on17 quarterly starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "006c7e6fc05bd2f1", + "event_kind": "finding", + "evidence_refs": [ + "evidence:forecast-prior-quarterly" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_ed894b182109bf3ea0537dc3bfbaba71f1fc770f4e67c098215f16703eaa49b5", + "recorded_at": "2026-10-14T23:33:39+00:00", + "result_id": "finding_8e4e10c724f49596f550483f43f70a6e6dc44dfe2a7077ba03148129beab6ac0", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:6bed98eb08064560", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 4028d3b91604cc3e5ce5391d9bb93db254c2cbfbea6ab0caf155aed57335b4ff Applicability: Public17 overlapping242-session starts2020Q1..2024Q1; prior-year fits, prior blend0.5/forecast0.4, retained V4 inertia0.02/deadband0.004, original V7 risk/cost/short rules. Observation: All91 paths/prefixes pass. V4-execution hybrid fivefold mean/worst Sharpe1.822/0.567 improves V7;17-start1.364/-0.351 trails V7=1.506/0.031 with160% turnover. Return/year-balanced gains remain. Interpretation: Outcome: replan retained: - Native-best V7, other checkpoints/prototypes, mixed benefits and original acceptance. changed: - Test additional lagged macro information versus the frozen price-only model. stopped: - Qualifying this fixed hybrid from January-only gains or tuning its prior/execution parameters.", + "tags": [ + "writeback-result" + ], + "title": "Does the fixed half-prior/frozen-forecast hybrid improve V7 mean and worst Sharpe on17 quarterly starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "909feb8b967bc20b", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-14T23:52:28+00:00", + "result_id": "question_01e1865e170a1fcdc8a5dd3274b7c347b466d9515dd2c7bd3b00e6bab44943a8", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:60ee58ae66b21b8c", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Exact retained macro-v9 manifest0f5d8101adf52156 and sourcece3a11b856ee3f31; native agent-6/cf4c5340f9ac versus agent-5/b36facd55fbf; score_first/maximize.", + "title": "Does exact macro-v9 improve native score beyond V7=20.53 in the evaluated round?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "938577e7c83b2a77", + "event_kind": "finding", + "evidence_refs": [ + "evidence:native-macro-v9" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_01e1865e170a1fcdc8a5dd3274b7c347b466d9515dd2c7bd3b00e6bab44943a8", + "recorded_at": "2026-10-14T23:52:28+00:00", + "result_id": "finding_5b6717a721136e001cd0cd18d38ac167431bbe265445e669bac92c72d990aef2", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:765d06d035e689db", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 3b401dd4047d2db548e3bb1349c526accb61fe3672d2bc3a91bc7324c8ac8893 Applicability: Exact retained macro-v9 manifest0f5d8101adf52156 and sourcece3a11b856ee3f31; native agent-6/cf4c5340f9ac versus agent-5/b36facd55fbf; score_first/maximize. Observation: 83 local paths/prefixes and standalone3 suites pass; macro beats local price/neutral mean and worst Sharpe. Exact native cf4c5340f9ac scores13.84 versus V7=20.53, Sharpe1.0882; V7 trio restored. Interpretation: Outcome: replan retained: - V7 best, exact macro-v9, local mixed benefits and every original target. changed: - Test risk-appetite allocation under V7 fixed priors rather than another forecast revision. stopped: - Resubmitting completed macro-v9 or inferring native success from public macro gains.", + "tags": [ + "writeback-result" + ], + "title": "Does exact macro-v9 improve native score beyond V7=20.53 in the evaluated round?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "08511e6fbe370d9f", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T00:01:32+00:00", + "result_id": "question_bc655ef1e0aabbdeb47880301f56b0824b84088dc51be6bdbb318e41f8cf4e44", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:290a7057c0e8616b", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 cold starts; preyear frozen half-priors, volatility0.07, V7 cost/risk/short rules. Gamma20/40 only; earlier V6 evidence has different forecasts.", + "title": "Does gamma20 improve both fivefold mean and worst Sharpe over gamma40 under fixed V7?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "16dfbc3c16dd9813", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-risk-appetite", + "evidence:v7-risk-stress" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_bc655ef1e0aabbdeb47880301f56b0824b84088dc51be6bdbb318e41f8cf4e44", + "recorded_at": "2026-10-15T00:01:32+00:00", + "result_id": "finding_a1f9ccc9130fa926cefba8b9725fda31a0b9fab51ff3b1d7e90bafd1003dbc6b", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:ada7a43ecefa49d3", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: ce2834c44c32844317518c3bc21518005ecbe74d4d221c6bd6c868783c2ace7a Applicability: Public2020-24 cold starts; preyear frozen half-priors, volatility0.07, V7 cost/risk/short rules. Gamma20/40 only; earlier V6 evidence has different forecasts. Observation: Gamma20 loses Sharpe5/5; mean1.394->1.038,worst0.402->-0.631,turnover110%->125%,return+0.35pp.22 paths/11 controls pass; nine detectors pass but no5pp drawdown bonus. Interpretation: Outcome: replan retained: - Exact best V7, scoped return gain, all earlier evidence and original acceptance. changed: - Test fixed CSI500 hedge under V7 priors; earlier hedge study used V4 forecasts. stopped: - Qualifying gamma20 or increasing budget based only on unused capacity.", + "tags": [ + "writeback-result" + ], + "title": "Does gamma20 improve both fivefold mean and worst Sharpe over gamma40 under fixed V7?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1b5f77563588ee06", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T00:07:40+00:00", + "result_id": "question_ac953d00faecbc4d5b4d0e80b29a29347e5339f11bc742649f001445bdfb469b", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:3a06fe96c2411cb4", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "17 public242-session starts2020Q1..2024Q1; preyear frozen half-priors,gamma40,volatility0.07,original risk/cost/short/cap/event rules. Only CSI300 versus CSI500 hedge identity changes.", + "title": "Does fixed CSI500 improve both V7 mean and worst Sharpe on17 public quarterly starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "cece591af401e092", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-hedge-comparison", + "evidence:v7-hedge-quarterly" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_ac953d00faecbc4d5b4d0e80b29a29347e5339f11bc742649f001445bdfb469b", + "recorded_at": "2026-10-15T00:07:40+00:00", + "result_id": "finding_3e1a62f27991d866708fc02bebd339333bd99eea19eea362620535bc3c208abe", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:e633e8634c6e71e5", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: a8a2e4e6c34025451fa350ccc4f7ffcb744118fa63d34d64c79e275c576f608b Applicability: 17 public242-session starts2020Q1..2024Q1; preyear frozen half-priors,gamma40,volatility0.07,original risk/cost/short/cap/event rules. Only CSI300 versus CSI500 hedge identity changes. Observation: 44 paths/prefixes and44 engine controls pass. CSI500 wins4/17,mean/worst Sharpe1.068/-0.362 versus1.506/0.031; turnover/reported VaR coverage improves,2024 Sharpe2.81. Exact V7 unchanged. Interpretation: Outcome: replan retained: - V7=20.53, original targets, scoped static-hedge counterevidence and mixed benefits. changed: - Test cost-aware two-hedge selection at existing decisions using current V7 utility and costs. stopped: - Qualifying static CSI500 from2024 gains or treating fixed-hedge losses as a ban on all selectors.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed CSI500 improve both V7 mean and worst Sharpe on17 public quarterly starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e37687eca63ebb83", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T00:20:23+00:00", + "result_id": "question_c19beeed81a9643ac738bb8f29c8b655ba7ba956debf2cfe97ff4c4fd96eb791", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:aec6b1f55c2d3d13", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "17 public242-session starts2020Q1..2024Q1,preyear half-priors,gamma40,budget0.07. CSI300/CSI500 sign patterns ranked by existing utility/costs at existing decisions; original rules.", + "title": "Does the fixed two-hedge cost-aware selector improve V7 mean and worst Sharpe on17 public starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "fa295859c068e014", + "event_kind": "finding", + "evidence_refs": [ + "evidence:discrete-hedge-quarterly", + "evidence:discrete-v10-qualified" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_c19beeed81a9643ac738bb8f29c8b655ba7ba956debf2cfe97ff4c4fd96eb791", + "recorded_at": "2026-10-15T00:20:23+00:00", + "result_id": "finding_d5879bc9e7b295a95403787f52f480123c1bc95123887a62d36be64931b1a71d", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:4fb2ff2b8ccebbcc", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: ddc222c0bddd0747948538b9480ac17c68a7ed25f1811217de744558ba736db8 Applicability: 17 public242-session starts2020Q1..2024Q1,preyear half-priors,gamma40,budget0.07. CSI300/CSI500 sign patterns ranked by existing utility/costs at existing decisions; original rules. Observation: 66 paths/prefixes pass.17-start mean/worst Sharpe1.586/0.076 versus1.506/0.031; turnover coverage11/17 versus12/17. Standalone3 suites pass with one explicit eligible-hedge assertion adaptation; v10 staged. Interpretation: Outcome: continue retained: - Original acceptance, exact V7 native-best20.53, static-hedge counterevidence and fixed utility/cost selector. changed: - Qualified/staged exact discrete-v10 and prepared a separate guarded native attempt; no request during cooldown.", + "tags": [ + "writeback-result" + ], + "title": "Does the fixed two-hedge cost-aware selector improve V7 mean and worst Sharpe on17 public starts?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "71f941afb4d8bcef", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T00:31:48+00:00", + "result_id": "question_b36d4a0da1e91c3455f1e8ff7feb1be590fbba106b273644c995a2eb88685566", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:ed5f87790ade2cd3", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Source7b13935d4d11b197; synthetic seed4401,opposite Equity trends +/-0.006 or0.0005 in60/100-return blocks,302/402 sessions. Original rules unchanged; no market/native claim.", + "title": "Does exact discrete-v10 keep drawdown within4% in all four declared alternating-Equity stress scenarios?" + }, + { + "agent_id": "benchmark-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "da40f503201060e9", + "event_kind": "finding", + "evidence_refs": [ + "evidence:discrete-switch-stress", + "evidence:discrete-switch-latency", + "evidence:discrete-opening-ablation" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_b36d4a0da1e91c3455f1e8ff7feb1be590fbba106b273644c995a2eb88685566", + "recorded_at": "2026-10-15T00:31:48+00:00", + "result_id": "finding_15dc82c727b76c0d282fc64101d0e4c449d8c924e37261476466d7718d2a32a3", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:fc584caca307ec19", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: bb28cbdc0574a71ef42e9e7e226b56fc21085e223b83749ec3889d14ca763d14 Applicability: Source7b13935d4d11b197; synthetic seed4401,opposite Equity trends +/-0.006 or0.0005 in60/100-return blocks,302/402 sessions. Original rules unchanged; no market/native claim. Observation: 12 paths/24 prefixes/12 controls/future tests pass;12 paid switches/24 crossings.60-block DD31.36/32.16%,median lag59. Public frozen-opening worst Sharpe-0.362 versus full0.076; disabling switches loses robustness. Interpretation: Outcome: replan retained: - Exact staged v10, native-best V7=20.53, public gains, adverse results and original acceptance. changed: - Test opening identity/training-prior sensitivity; preserve financial stress losses. stopped: - Treating passed caps, ledgers or positive expected utility as drawdown evidence.", + "tags": [ + "writeback-result" + ], + "title": "Does exact discrete-v10 keep drawdown within4% in all four declared alternating-Equity stress scenarios?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "418ddee48450c832", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T07:35:54+00:00", + "result_id": "question_edb3e4393eec7be85575488c425591798428d02dbce6b0f5180978b8545ce540", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:a045a0ac832f3ee0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Destination agent-7/51cae2f5712c exact source7b13935d4d11b197 and manifestc2c8a610e411af76; versus retained agent-5/1c3facc64838, score_first/maximize.", + "title": "Does exact discrete-v10 exceed retained V7 native score20.53 in this destination round?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "242a09e69f4d0f81", + "event_kind": "finding", + "evidence_refs": [ + "evidence:native-v10-destination" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_edb3e4393eec7be85575488c425591798428d02dbce6b0f5180978b8545ce540", + "recorded_at": "2026-10-15T07:35:54+00:00", + "result_id": "finding_88365c364ce3fede567b6416868c1c45e337e23e7948b1a98da48482db8eef8f", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:f00c9cd6e2859e12", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: c2c8a610e411af76113ac29ea76699c8247cb88faf84b6d031b14aac27e4877e Applicability: Destination agent-7/51cae2f5712c exact source7b13935d4d11b197 and manifestc2c8a610e411af76; versus retained agent-5/1c3facc64838, score_first/maximize. Observation: Exact destination V10/51cae2f5712c scores11.43<20.53; Sharpe1.1194,DD1.46%,excess-22.97%,VaR4.1/3.3%,turnover112%. History has7 completed runs; matched current hashes permitted exact V7 restoration. Native1/1 is parseability. Interpretation: Outcome: replan retained: - Exact V7/V10 checkpoints, original priorities and scoped public/synthetic counterevidence. changed: - Restored V7 and reconciled destination IDs/cooldown; investigate prior-driven allocation and hedge-choice interaction. stopped: - Resubmitting completed V10 or treating its public gains as native adoption.", + "tags": [ + "writeback-result" + ], + "title": "Does exact discrete-v10 exceed retained V7 native score20.53 in this destination round?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f96c803a41bf1735", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T07:44:44+00:00", + "result_id": "question_f23ee48558b6102579c8d59429b7d4d00fac90c6989727904c463fbf28aa5ad5", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:96c8f6da83b26a02", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "2020-2024 public full-year cold starts; preyear priors,blend0/0.5 crossed with fixedCSI300/dynamic2hedges; exact V7/V10 rules. No native-trajectory explanation or transfer claim.", + "title": "Do frozen trained asset priors change V10 opening hedge choice on the five matched public cold starts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "0023a414ae15a384", + "event_kind": "finding", + "evidence_refs": [ + "evidence:prior-hedge-factorial" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_f23ee48558b6102579c8d59429b7d4d00fac90c6989727904c463fbf28aa5ad5", + "recorded_at": "2026-10-15T07:44:44+00:00", + "result_id": "finding_a0e1ca20f5a88621f5aecf93362e753fb2432073e0d662170def18239f40b2b0", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:f51db6ccd4bb2537", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 5d2e4b876209c17f742c08c8b0616ca4dfc81e955756e6f6c6ae392033dca9e7 Applicability: 2020-2024 public full-year cold starts; preyear priors,blend0/0.5 crossed with fixedCSI300/dynamic2hedges; exact V7/V10 rules. No native-trajectory explanation or transfer claim. Observation: 20 paths/20 prefixes/20 source controls pass. Removing trained priors changes opening hedge3/5, eliminates switches; class dynamic equals fixed decisions/costs/metrics. Class mean/worst Sharpe1.437/0.384 vs V71.394/0.402 and V101.661/0.473; return2.24% vs2.67/3.69%. Interpretation: Outcome: replan retained: - Exact V7 best20.53, rejected V10 and all original acceptance plus scoped public/native/synthetic evidence. changed: - Test selector-only prior centring; keep allocation forecasts to isolate choice effects from lost return carry. stopped: - Broadening identical class-only selectors or adopting them from mean Sharpe/turnover alone.", + "tags": [ + "writeback-result" + ], + "title": "Do frozen trained asset priors change V10 opening hedge choice on the five matched public cold starts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1f937d80d6149a6a", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T07:56:30+00:00", + "result_id": "question_5f989e01fce1c10353f133f2935b0743e381400bee5aa4a4fc197d78acce2289", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:1f7eda55bac13dd7", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Strictly preyear frozen priors;2020-2024 public full-year cold starts; original allocation forecasts/solves/risk/cost/cap/event/short rules. This is development evidence, not hidden-native causation.", + "title": "Does centring only learned Equity prior dispersion in hedge ranking improve both mean and worst Sharpe over fixed V7 on five matched public cold starts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "8265777fe4c84e7f", + "event_kind": "finding", + "evidence_refs": [ + "evidence:selector-prior-ranking" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_5f989e01fce1c10353f133f2935b0743e381400bee5aa4a4fc197d78acce2289", + "recorded_at": "2026-10-15T07:56:30+00:00", + "result_id": "finding_e5e315790f703c9c287bee4a3f79d35cbe1b09db0df560c6a9fdd7df874a759a", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:2647f611090075b6", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 044f9e78c1c4898760649a131eed485d69ee63b06704c0d8860c6955f267e231 Applicability: Strictly preyear frozen priors;2020-2024 public full-year cold starts; original allocation forecasts/solves/risk/cost/cap/event/short rules. This is development evidence, not hidden-native causation. Observation: Correction reaches181 decisions;3 same-input choices change, opening3/5. All362 candidate solves match V10. Every alternative admissible; none clears net epsilon. Centred holdings/costs/metrics equal V7 in5/5: mean/worst Sharpe1.394/0.402,excess-4.32%. Interpretation: Outcome: replan retained: - Exact V7 best20.53, all original acceptance and scoped public/native/synthetic counterevidence. changed: - Test initial covariance independence separately: fixed within-Equity/Bond correlation0.5, original updates thereafter. stopped: - Staging or broadening centred hedge ranking that reproduces V7; no native request.", + "tags": [ + "writeback-result" + ], + "title": "Does centring only learned Equity prior dispersion in hedge ranking improve both mean and worst Sharpe over fixed V7 on five matched public cold starts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "c188bd217d91cc0a", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T08:05:41+00:00", + "result_id": "question_9a33ea3df2082ee263e197f5052a1b49a43127fad3fff6d905d71a900832c7a1", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:565f1211f1637a44", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 full-year cold starts; preyear priors; only initial within-Equity/Bond correlation0.5. Original V7 updates/rules; development evidence, no native-transfer claim.", + "title": "Does the fixed within-Equity/Bond correlation0.5 initial covariance improve both fivefold mean and worst Sharpe over exact V7 diagonal initialization?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7c2568e23e576efb", + "event_kind": "finding", + "evidence_refs": [ + "evidence:initial-class-covariance" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_9a33ea3df2082ee263e197f5052a1b49a43127fad3fff6d905d71a900832c7a1", + "recorded_at": "2026-10-15T08:05:41+00:00", + "result_id": "finding_8dbbb512a17f15005983d79d46acf9c422c38ab6adf53609dc53d72d73f512c4", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:21cdb3f7bbbf4465", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 19747d718074cca02bfb2d4422edc542bb0e7f9f9a8c70064de75a0edc514703 Applicability: Public2020-24 full-year cold starts; preyear priors; only initial within-Equity/Bond correlation0.5. Original V7 updates/rules; development evidence, no native-transfer claim. Observation: 10 paths/prefixes and10 controls pass. Openings change5/5; empirical covariance identical after8-13 sessions. Mean/worst Sharpe1.492/0.368 vs1.394/0.402;return+0.24pp.147/176 later targets differ L1>0.001. No joint robustness gain. Interpretation: Outcome: replan retained: - Exact V7 best20.53, all original targets and scoped public/native/synthetic evidence, including mean-only gain. changed: - Probe one scheduled session20 inertia reset under original V7; do not combine rejected covariance alternative. stopped: - Quarterly expansion, correlation tuning or adoption based on mean Sharpe alone.", + "tags": [ + "writeback-result" + ], + "title": "Does the fixed within-Equity/Bond correlation0.5 initial covariance improve both fivefold mean and worst Sharpe over exact V7 diagonal initialization?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "bb11fc4464d81877", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T08:13:33+00:00", + "result_id": "question_7215707a00b3c51f85c9e29604db61170893a1e9fe20c5c78866ec1cc63e6b8d", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:450f9f7a26157980", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 full-year cold starts; preyear priors; original V7 except one session20 soft-penalty reset. Real costs/risk/forecasts unchanged. Development evidence, no native claim.", + "title": "Does one session20 soft-inertia reset improve both fivefold mean and worst Sharpe over exact V7?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "52b351d10089fd65", + "event_kind": "finding", + "evidence_refs": [ + "evidence:session20-inertia" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_7215707a00b3c51f85c9e29604db61170893a1e9fe20c5c78866ec1cc63e6b8d", + "recorded_at": "2026-10-15T08:13:33+00:00", + "result_id": "finding_9716fdac38ea4e5041b6eb116cec01ff37293914ffa572df6cd9550a2bb7aa12", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:e8c86c390747824c", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: e26d7b0da2354e23dfc295b3c141aea8443968609b3be3daa4c92b97755ee52a Applicability: Public2020-24 full-year cold starts; preyear priors; original V7 except one session20 soft-penalty reset. Real costs/risk/forecasts unchanged. Development evidence, no native claim. Observation: 10 paths/20 prefixes/10 controls pass; reset once20 and restoration verified. All5 solves succeed; targets change L1 0.051-0.207. Sharpe loses4/5;mean/worst1.244/0.279 vs1.394/0.402,turnover114%vs110%. Worst DD and VaR coverage improve. Interpretation: Outcome: replan retained: - Exact V7 best20.53, original targets and scoped public/native/synthetic evidence plus reset risk benefits. changed: - Test portfolio objective under V7 half-priors; earlier failed ratio test used V4 forecasts. Keep original execution. stopped: - Broadening, staging or adopting the session20 reset; global inertia changes inferred from it.", + "tags": [ + "writeback-result" + ], + "title": "Does one session20 soft-inertia reset improve both fivefold mean and worst Sharpe over exact V7?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "6254feab612ea2ef", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T08:23:05+00:00", + "result_id": "question_92aebbc5fd2ab781e9f6d4d811a0fecbb94eef298371f9e2b8da20adb8dd9bf4", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:c27682b382818d0e", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 full-year cold starts; preyear priors; exact V7 variance versus existing net-Sharpe mode,vol0.07 and original costs/risk/execution. Development evidence, no native claim.", + "title": "Does the existing net-Sharpe objective improve both fivefold mean and worst Sharpe over V7 variance utility under identical half-trained forecasts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "d9637d41e112bc99", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-objective" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_92aebbc5fd2ab781e9f6d4d811a0fecbb94eef298371f9e2b8da20adb8dd9bf4", + "recorded_at": "2026-10-15T08:23:05+00:00", + "result_id": "finding_a104132402a03d9f59d26a5febef8b3f6268b902caa4ddab127e05780c836e84", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:4573beff95ca661a", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 4b2c7f92a1f7f32d25e456096cc3613ea6b1653c7957c5aeffc610c72be3b41c Applicability: Public2020-24 full-year cold starts; preyear priors; exact V7 variance versus existing net-Sharpe mode,vol0.07 and original costs/risk/execution. Development evidence, no native claim. Observation: 10 paths/20 prefixes/15 controls pass. Ratio active181 decisions;60 gradient points pass,max error1e-9.8 feasible solver warnings,no fallback; converged2023 also loses. Sharpe loses5/5;mean/worst0.777/-1.066,DD4.84%,turnover147%. Interpretation: Outcome: replan retained: - Exact V7 best20.53, original targets, scoped ratio return/VaR gains and all public/native/synthetic evidence. changed: - Test causal downside penalty at fixed gamma40; retain independent original covariance budget and execution. stopped: - Broadening/adopting existing ratio or attributing all financial losses to solver warnings.", + "tags": [ + "writeback-result" + ], + "title": "Does the existing net-Sharpe objective improve both fivefold mean and worst Sharpe over V7 variance utility under identical half-trained forecasts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "487628a4d2710ada", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T08:50:16+00:00", + "result_id": "question_dd6024c6c360a95a1db517b94bbf3ec6a518a6ece07dea73e9742615db70063c", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:3b00e2ac722626b8", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 preyear cold starts; exact V7 except fixed2x raw portfolio semivariance after20 returns. Original covariance limits/costs/execution. Development evidence; no native claim.", + "title": "Does fixed2x causal portfolio-semivariance improve both fivefold mean and worst Sharpe over exact V7 variance utility?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "03c77f9ba43af2fe", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-downside-objective" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_dd6024c6c360a95a1db517b94bbf3ec6a518a6ece07dea73e9742615db70063c", + "recorded_at": "2026-10-15T08:50:16+00:00", + "result_id": "finding_0c6018958340a868e1e56902db891bb0ca960078be7405de697c1e998b147c83", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:447a55820c6f2505", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 3faeec6c973cb470e1a90944542628b8a275b43b094511e069354cf0fe0a64cd Applicability: Public2020-24 preyear cold starts; exact V7 except fixed2x raw portfolio semivariance after20 returns. Original covariance limits/costs/execution. Development evidence; no native claim. Observation: 10 paths/20 prefixes/15 controls pass;165 live downside decisions,60 gradient points pass. Mean/worst Sharpe1.200/0.527 vs1.394/0.402; all solves feasible/successful. Return/DD/VaR gains,turnover121%. Interpretation: Outcome: replan retained: - V7 best20.53, all original acceptance and scoped counterevidence, including downside worst-Sharpe/return/DD/VaR gains. changed: - Diagnose actual V7 forecast components and forward-horizon skill before choosing another return model. stopped: - Broader folds, staging or adoption of fixed downside penalty; risk-only gains treated as closing the alpha gap.", + "tags": [ + "writeback-result" + ], + "title": "Does fixed2x causal portfolio-semivariance improve both fivefold mean and worst Sharpe over exact V7 variance utility?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "71989954a6eaaf9b", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T09:03:20+00:00", + "result_id": "question_ca700debf231402e8c6a8831b96c308b668d26246a606a02a6e76b50d9475671", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:bcdb685eca377caf", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 preyear V7 cold starts; actual decisions, forward1/5/20 labels and zero/constant-prior references. Overlapping development data; labels diagnostic only, no native/alpha claim.", + "title": "Does live V7 mu improve five-session forecast MSE over its constant half-prior reference at actual decisions in five preyear cold starts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "3a6aa815043484ed", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-forecast-information" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_ca700debf231402e8c6a8831b96c308b668d26246a606a02a6e76b50d9475671", + "recorded_at": "2026-10-15T09:03:20+00:00", + "result_id": "finding_3591374468b32d7e30f15f3110f1e1003e186279ae0ccaffc68a2d6907959715", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:f8579491a239a35c", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 64b3741c3ee816db79f83ae0020df8c9c99fb8121bd069f6d52554d5aaa75696 Applicability: Public2020-24 preyear V7 cold starts; actual decisions, forward1/5/20 labels and zero/constant-prior references. Overlapping development data; labels diagnostic only, no native/alpha claim. Observation: 181 mu vectors,180/174/171 labels at1/5/20; all causal/source/arithmetic controls pass.5-close MSE loses to prior5/5;20-close IC negative5/5.1-close IC positive5/5; portfolio/class evidence mixed. V7 financial gap unchanged. Interpretation: Outcome: replan retained: - V7 best20.53, original acceptance, scoped mixed forecast/downside and all native/synthetic counterevidence. changed: - Test constant-half-prior mu, removing only rolling empirical update while retaining risk/hedge adaptation. stopped: - Treating pooled MSE, gross frozen-weight labels or next-day IC as proof of executable native alpha.", + "tags": [ + "writeback-result" + ], + "title": "Does live V7 mu improve five-session forecast MSE over its constant half-prior reference at actual decisions in five preyear cold starts?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1f90ee854156bc50", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T09:11:37+00:00", + "result_id": "question_9b7b2bffecdc57c8e49d9795723d616fe22cc25f3fd04edc0494a9254bc3e7df", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:bad67cf3c24831d0", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 preyear cold starts; exact V7 versus constant half-prior mu only. Original risk/hedge adaptation,costs,execution. Development evidence; no native claim.", + "title": "Does constant-half-prior mu improve both fivefold mean and worst Sharpe over exact V7 while preserving original risk and execution?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "7044d0c6f6493404", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-constant-prior" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_9b7b2bffecdc57c8e49d9795723d616fe22cc25f3fd04edc0494a9254bc3e7df", + "recorded_at": "2026-10-15T09:11:37+00:00", + "result_id": "finding_62e4436a9ba2b30fbc2d8765a026e1285aa601f7120e1187966382fd50b7c52d", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:82cede1bc8b787be", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 36a170bfddec82a42e0a797ad8a47da5831c16c87dbc5629e1974a903392bf7e Applicability: Public2020-24 preyear cold starts; exact V7 versus constant half-prior mu only. Original risk/hedge adaptation,costs,execution. Development evidence; no native claim. Observation: 10 paths/20 prefixes/15 controls pass;176 mu/target changes,60 gradient points,all solves succeed. Sharpe wins4/5;mean/worst1.476/0.306 vs1.394/0.402. Cost/DD gains,VaR coverage3/5vs4/5;2022 Equity selection/allocation decline. Interpretation: Outcome: replan retained: - V7 best20.53, original acceptance, constant-mu alternate benefits and all scoped diagnostic/native/synthetic evidence. changed: - Separate Equity and non-Equity empirical-mu channels in one fixed2x2 with exact retained endpoints. stopped: - Quarterly expansion or adoption of whole-vector freezing; MSE gains assumed to prove robust trading gains.", + "tags": [ + "writeback-result" + ], + "title": "Does constant-half-prior mu improve both fivefold mean and worst Sharpe over exact V7 while preserving original risk and execution?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b5cbcf7e0724a81f", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T09:21:50+00:00", + "result_id": "question_39ed97bf465e61b6c2923a2a1fb00ab1dc705f49bd9b4b2208cf814a46625b4d", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:45f12aa0f23e7350", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Public2020-24 preyear V7 cold starts; fixed2x2 Equity/non-Equity mu freezing. Original risk/hedge/cost/execution rules and exact retained endpoints. Development data; no native claim.", + "title": "Does either mixed Equity/non-Equity mu-freezing policy improve both fivefold mean and worst Sharpe over exact V7?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "6ed509fa9803bf5a", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-mu-class-factorial" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_39ed97bf465e61b6c2923a2a1fb00ab1dc705f49bd9b4b2208cf814a46625b4d", + "recorded_at": "2026-10-15T09:21:50+00:00", + "result_id": "finding_b30a9e4859f89d385db48c952b785f183f81d2725fa31611cd26f89d7e67fc05", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:7f6877e6cc62ed6c", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 551471a17384b18f0b30ccd4726e4bb8359416755b4dfb6359878d86f908e41a Applicability: Public2020-24 preyear V7 cold starts; fixed2x2 Equity/non-Equity mu freezing. Original risk/hedge/cost/execution rules and exact retained endpoints. Development data; no native claim. Observation: 20 paths/40 prefixes/40 controls pass,120 gradient points;all724 solves succeed. Mixed mean/worst1.452/0.310 and1.409/0.379 vs1.394/0.402; neither repairs worst. Both retained endpoints exact.7 native results still best20.53. Interpretation: Outcome: replan retained: - V7 best20.53,original targets,full-constant mean/cost/DD benefits,worst/VaR risks and all scoped counterevidence. changed: - Qualify exact full-constant alternate for one separately admitted native transfer/adoption decision. stopped: - Mixed-arm broadening/adoption and further class partitions without new evidence; every-proxy native prerequisite.", + "tags": [ + "writeback-result" + ], + "title": "Does either mixed Equity/non-Equity mu-freezing policy improve both fivefold mean and worst Sharpe over exact V7?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f0eebcd9c195b0f7", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T09:40:31+00:00", + "result_id": "question_107c793317d53b730d16dfec013da20634939deaeb40a24932419c765b7233be", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:77572e96d6adbc90", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "One destination-native evaluation,agent8,exact V7 with constant half-prior mu and truthful metadata. Original risk/execution/cost rules; public/synthetic qualification. No broad generalization.", + "title": "Do the retained constant-prior policy public mean-Sharpe/cost gains improve destination-native score over exact V7?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "e8f04bb7aaa79b1f", + "event_kind": "finding", + "evidence_refs": [ + "evidence:native-constant-prior" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_107c793317d53b730d16dfec013da20634939deaeb40a24932419c765b7233be", + "recorded_at": "2026-10-15T09:40:31+00:00", + "result_id": "finding_61173c40662498b792a132cf0aeb5fefbc6cf47c13a3fc540ff1fbc71223c546", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:30cc240ae6d9425f", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 0a253e0b2f538ab7db35ca5a6686cb74ad5c3c7c87bc4f7686d7498ecc706775 Applicability: One destination-native evaluation,agent8,exact V7 with constant half-prior mu and truthful metadata. Original risk/execution/cost rules; public/synthetic qualification. No broad generalization. Observation: One accepted agent8 completed10.65<20.53;Sharpe0.423vs1.373,excess-24.57%vs-22.64%,independent VaR2.9%vs4.5%. Source/ID matched;exact V7 restored;ledger/events pass. Public/CLI/causal qualification all passed. Interpretation: Outcome: replan retained: - V7 best20.53,original targets,constant public benefits/native losses and all scoped public/synthetic counterevidence. changed: - Diagnose rolling forecast value under controlled persistence/reversion/noise before choosing a new estimator. stopped: - Adoption/repeated native evaluation of this exact constant source;local MSE or mean-Sharpe gains treated as transfer.", + "tags": [ + "writeback-result" + ], + "title": "Do the retained constant-prior policy public mean-Sharpe/cost gains improve destination-native score over exact V7?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "26169297b0911a7f", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T09:56:08+00:00", + "result_id": "question_022b481ae43ce82d0f9f9df18f36ea80665d98a74ec15383a4f13c90a28fb02d", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:883934e0478236db", + "schema_version": "loopx_explore_result_event_v0", + "status": "open", + "summary": "Synthetic360-session paths;seeds4901-4903,3 declared drift/AR1/noise families and shared vol break. Exact V7 versus constant mu,embedded priors,original risk/cost/execution. No transfer claim.", + "title": "Does exact V7 rolling mu improve full-cost mean and worst Sharpe over constant prior in the declared persistent,reverting and noise regimes?" + }, + { + "agent_id": "checkpoint-source-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "6e3c41e99b9ba229", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-forecast-regimes" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_022b481ae43ce82d0f9f9df18f36ea80665d98a74ec15383a4f13c90a28fb02d", + "recorded_at": "2026-10-15T09:56:08+00:00", + "result_id": "finding_c04cd15aed0fce2842758adbb921ded2d64ab7802031deedf1db80b4c78a95df", + "run_id": "checkpoint:0250eb37566b03c22e770f7bf4ef594c895f75aae11b21055a1e093b242613a3:ac3a249a29977970", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: b3b013aa64c85042e74eed7b4b3170d989b0f3233676d0b890563bf1384ebc25 Applicability: Synthetic360-session paths;seeds4901-4903,3 declared drift/AR1/noise families and shared vol break. Exact V7 versus constant mu,embedded priors,original risk/cost/execution. No transfer claim. Observation: Constant wins8/9;all3 family mean/worst gains. Live class mu never negative in6 persistent cases;empirical sign lags12-80. Pooled20-close MSE wins3/3 persistent,target-weighted1/3. Hard controls pass;constant native10.65 remains below20.53. Interpretation: Outcome: replan retained: - Exact V7 best20.53,original targets,constant public benefits/native loss and all scoped counterevidence. changed: - Test one observed class-change statistic before replacing class mu mean;retain V7 within-class information. stopped: - Synthetic/pooled-MSE gains treated as native transfer;blanket empirical unshrinking or repeated constant evaluation.", + "tags": [ + "writeback-result" + ], + "title": "Does exact V7 rolling mu improve full-cost mean and worst Sharpe over constant prior in the declared persistent,reverting and noise regimes?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "1bc083df106f349d", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T12:39:46+00:00", + "result_id": "question_9194c07558f954c04f0be08afd2a4bd91bf73092d8cac9a1d5eeca8dbec4bed3", + "run_id": "explore-result:56eac0ebfd19493719de8264bed9381d98025f040a3c6efa3ae454f22fdf2fda", + "schema_version": "loopx_explore_result_event_v0", + "status": "resolved", + "summary": "Exact V7;9 fixed synthetic seed/family cases plus5 preyear public2020-24 starts. Last20/prior40 class contrast,global60-mean HAC5,3-SE gate;original costs/risk/execution. No transfer claim.", + "title": "Does the frozen whole-window HAC5 3-SE class-change rule materially change forecasts or targets on the declared synthetic/public cases?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "93d01535b271057f", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-class-change" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_9194c07558f954c04f0be08afd2a4bd91bf73092d8cac9a1d5eeca8dbec4bed3", + "recorded_at": "2026-10-15T12:39:46+00:00", + "result_id": "finding_56eac0ebfd19493719de8264bed9381d98025f040a3c6efa3ae454f22fdf2fda", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 8a7247163023a5f0dc4fb65a1b6dcc972fa124fb9ddf7cb56dfdf76f6ed53d22 Applicability: Exact V7;9 fixed synthetic seed/family cases plus5 preyear public2020-24 starts. Last20/prior40 class contrast,global60-mean HAC5,3-SE gate;original costs/risk/execution. No transfer claim. Observation: 37 paths/74 prefixes/51 controls pass,1400 solves. Gate inactive14/14;max|z|2.97065;all full paths equal V7. Power fixture changes1 mu/2 targets. Separate-window residual statistics yield18 diagnostic triggers;no executable financial claim. Interpretation: Outcome: replan retained: - Exact V7 best20.53,original targets,frozen failed activation and all scoped public/native/synthetic evidence. changed: - Test one separate-window residual covariance estimate with the same3-SE threshold and original execution. stopped: - Broadening or submitting inactive whole-window rule;posthoc threshold tuning or relabeling earlier failure.", + "tags": [ + "writeback-result" + ], + "title": "Does the frozen whole-window HAC5 3-SE class-change rule materially change forecasts or targets on the declared synthetic/public cases?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "2e430be1514df7ec", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T12:51:21+00:00", + "result_id": "question_f2faa5b71c698e1743b53e7e525983d525e8a668436023d43a48f1f803aaa312", + "run_id": "explore-result:a7eab7e4c66dbfc80cdfa54695fd2bdd24f66d6742fc304219a104a23d498de5", + "schema_version": "loopx_explore_result_event_v0", + "status": "resolved", + "summary": "Exact V7;9 fixed synthetic cases and5 preyear public2020-24 starts. Two-window residual covariance change;3-SE/SE>0,mu replacement,cost/risk/execution fixed. No transfer claim.", + "title": "Does separately centered HAC5 covariance make the frozen class-change rule useful without costly false mean-shift alarms?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "285cff5baa78c5db", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-class-change-residual" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_f2faa5b71c698e1743b53e7e525983d525e8a668436023d43a48f1f803aaa312", + "recorded_at": "2026-10-15T12:51:21+00:00", + "result_id": "finding_a7eab7e4c66dbfc80cdfa54695fd2bdd24f66d6742fc304219a104a23d498de5", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 84c0e24e5de93af8f4fbf4aec835708eab2ebd47c9c9701ac75273be01f94a1f Applicability: Exact V7;9 fixed synthetic cases and5 preyear public2020-24 starts. Two-window residual covariance change;3-SE/SE>0,mu replacement,cost/risk/execution fixed. No transfer claim. Observation: 18 triggers,13 diagnostic false alarms;14 paths/28 prefixes/28 controls pass. Persistent mean/worstSharpe improves;noise4901 loses0.3963. Public5 unchanged;negative Commodity shifts missed3/3. Exact zero-SE step missed;decimal roundoff may trigger. Interpretation: Outcome: replan retained: - Exact V7 best20.53,original targets,drift gains and scoped public/native/synthetic/failed-activation evidence. changed: - Isolate all18 observed trigger effects with matched omission replays before another mechanism or qualification. stopped: - Treating activation or pooled drift gains as native transfer;threshold tuning or hiding false alarms/zero-SE misses.", + "tags": [ + "writeback-result" + ], + "title": "Does separately centered HAC5 covariance make the frozen class-change rule useful without costly false mean-shift alarms?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "dcc5fc816860a67c", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T13:03:03+00:00", + "result_id": "question_9105953156691ba4b5f3cc0cf55b87bfa6801ef6d012b59a47f4f1e4fdd2ae2a", + "run_id": "explore-result:7e29eb5e036b2a97a9353ca235a4c9c1971ae91cd2e843dec881f08fb25812df", + "schema_version": "loopx_explore_result_event_v0", + "status": "resolved", + "summary": "All18 raw60-hash/class interventions on9 fixed synthetic cases. Omit one class replacement;other signals,risk,cost/gate/execution fixed. Conditional path effects;no transfer claim.", + "title": "Do matched omission replays show helpful,harmful and physically inactive class-change trigger effects under fixed execution?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "b5f29e1efc4f32d0", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-class-trigger-effects" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_9105953156691ba4b5f3cc0cf55b87bfa6801ef6d012b59a47f4f1e4fdd2ae2a", + "recorded_at": "2026-10-15T13:03:03+00:00", + "result_id": "finding_7e29eb5e036b2a97a9353ca235a4c9c1971ae91cd2e843dec881f08fb25812df", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 36b203de0106da185ef8b20d20a7c36adec195156eb984116ebd50c39ab48294 Applicability: All18 raw60-hash/class interventions on9 fixed synthetic cases. Omit one class replacement;other signals,risk,cost/gate/execution fixed. Conditional path effects;no transfer claim. Observation: 18 matched omissions pass;9 helpful,2 harmful,7 physically inactive. False alarms split4/2/7;all5 drift-window effects helpful.1011 solves and36 omission gradients pass. Large effects reconcile to holdings PnL;rounding/gate effects retained. Interpretation: Outcome: replan retained: - Exact V7 best20.53,original targets,residual drift gains/noise loss,zero-SE misses and all scoped counterevidence. changed: - Qualify one frozen standalone residual candidate for native score-first evidence despite mixed local proxies. stopped: - Retrospective latent-label filters,additive omission attribution,threshold tuning or synthetic gains as transfer proof.", + "tags": [ + "writeback-result" + ], + "title": "Do matched omission replays show helpful,harmful and physically inactive class-change trigger effects under fixed execution?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "f87f5bfb17221c33", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T13:15:46+00:00", + "result_id": "question_2fcabf74248ba327a22e5c13ca3b7476e36c4a453512dc2c21132052aa7c1178", + "run_id": "explore-result:8f608b5540f1cf3add50ddb39ec712c55ee162181b20518137383722464e4a77", + "schema_version": "loopx_explore_result_event_v0", + "status": "resolved", + "summary": "One destination-native agent9 evaluation563accd27791,source16a2ee7c938b. Original risk/cost/execution;14 matched source paths and strict CLI qualified. Published feedback only;no generalization claim.", + "title": "Does the frozen separate-residual HAC5 class-change candidate improve destination-native score over exact V7?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "aa8f65feb4b42722", + "event_kind": "finding", + "evidence_refs": [ + "evidence:native-class-change-residual" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_2fcabf74248ba327a22e5c13ca3b7476e36c4a453512dc2c21132052aa7c1178", + "recorded_at": "2026-10-15T13:15:46+00:00", + "result_id": "finding_8f608b5540f1cf3add50ddb39ec712c55ee162181b20518137383722464e4a77", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: 16a2ee7c938bf5aa8be7c7295b0d7e257880600497530f97f49d85be8122683e Applicability: One destination-native agent9 evaluation563accd27791,source16a2ee7c938b. Original risk/cost/execution;14 matched source paths and strict CLI qualified. Published feedback only;no generalization claim. Observation: Accepted563accd27791/agent9 ties20.53 with same published financial metrics.14 standalone paths/strict CLI qualify;exact submitted checkpoint compared before V7 restore,ledger passes. Hidden activation/full-path equality unknown;original targets open. Interpretation: Outcome: replan retained: - Exact V7 best20.53,residual drift gains/noise loss/native tie,constant10.65 and all original targets/counterevidence. changed: - Test benchmark-relative variance preference with V7 mu and original absolute budget,gate/cost/execution controls. stopped: - Adoption/repeated evaluation of this tied source;native endpoint equality treated as hidden trace equivalence.", + "tags": [ + "writeback-result" + ], + "title": "Does the frozen separate-residual HAC5 class-change candidate improve destination-native score over exact V7?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "ff71a0106ff21251", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T13:31:03+00:00", + "result_id": "question_e47b818223d553141128adcfb9b00e91eb0f2a640c3bb2d37942a3f114344ba6", + "run_id": "explore-result:479092931a1b0bcd2984201a9542d9e6a629bc133bd4f6deedcaf1d18d795f53", + "schema_version": "loopx_explore_result_event_v0", + "status": "resolved", + "summary": "Exact V7 mu/risk/cost/execution;fixed60/30/10 anchor.9 synthetic cases and5 preyear public2020-24 starts.14 paths with source/disabled/prefix controls. No native transfer claim.", + "title": "Does replacing absolute variance preference with benchmark-relative variance improve full-cost public Sharpe robustness under the original absolute budget?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "49f896a31b1b0ebe", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-benchmark-risk" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_e47b818223d553141128adcfb9b00e91eb0f2a640c3bb2d37942a3f114344ba6", + "recorded_at": "2026-10-15T13:31:03+00:00", + "result_id": "finding_479092931a1b0bcd2984201a9542d9e6a629bc133bd4f6deedcaf1d18d795f53", + "schema_version": "loopx_explore_result_event_v0", + "status": "refuted", + "summary": "Input revision: a0bf2614e571e0b7c6d427aedc7ef1a5415121e565f7748811e453c938337d4f Applicability: Exact V7 mu/risk/cost/execution;fixed60/30/10 anchor.9 synthetic cases and5 preyear public2020-24 starts.14 paths with source/disabled/prefix controls. No native transfer claim. Observation: 14 paths/28 prefixes/28 controls pass;527 solves,1054 checks,27 feasible warnings. Synthetic9/9 improve;public1/5,mean/worst0.656/-0.666,turnover213%,DD5.56%. Opening exposure and ongoing targets both change;rare repairs alone cannot explain turnover. Interpretation: Outcome: replan retained: - Exact V7 best20.53,original targets,benchmark synthetic/return gains and scoped public/native/cost counterevidence. changed: - Use fixed opening/continuation2x2 ablations to locate preference effects before changing coefficients or solver. stopped: - Immediate promotion before staged effects are understood;blaming all losses on repairs/warnings without controls.", + "tags": [ + "writeback-result" + ], + "title": "Does replacing absolute variance preference with benchmark-relative variance improve full-cost public Sharpe robustness under the original absolute budget?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "78ff7038c5892871", + "event_kind": "node", + "goal_id": "portfolio-warm-on-20261015-01", + "node_kind": "question", + "recorded_at": "2026-10-15T13:44:53+00:00", + "result_id": "question_cf25eae8558040c086d11abacd1093662881bf06ab3e98e6ad81b4563183ef9c", + "run_id": "explore-result:a55f92771a6525c3c6d61ea4166fda004c74e8c1679ebab488716fee9da35097", + "schema_version": "loopx_explore_result_event_v0", + "status": "resolved", + "summary": "Fixed14 cases:9 synthetic and5 preyear public2020-24.2 staged arms;28 paths/56 prefixes/source and28 endpoint controls. Original coefficients,mu,absolute budget,costs and events. No transfer claim.", + "title": "Do matched timing interventions distinguish opening effects from ongoing benchmark-preference losses under unchanged V7 risk and execution?" + }, + { + "agent_id": "warm-on-agent", + "boundary": { + "absolute_paths_recorded": false, + "credential_values_recorded": false, + "raw_logs_recorded": false, + "raw_session_transcript_recorded": false, + "raw_task_text_recorded": false, + "raw_trajectory_recorded": false + }, + "event_id": "03502c447ba623b8", + "event_kind": "finding", + "evidence_refs": [ + "evidence:v7-benchmark-risk-timing" + ], + "goal_id": "portfolio-warm-on-20261015-01", + "node_id": "question_cf25eae8558040c086d11abacd1093662881bf06ab3e98e6ad81b4563183ef9c", + "recorded_at": "2026-10-15T13:44:53+00:00", + "result_id": "finding_a55f92771a6525c3c6d61ea4166fda004c74e8c1679ebab488716fee9da35097", + "schema_version": "loopx_explore_result_event_v0", + "status": "confirmed", + "summary": "Input revision: 78573e5951013eae0b334e9c609ad78e1a77b74a0141433da9cbc0072099403c Applicability: Fixed14 cases:9 synthetic and5 preyear public2020-24.2 staged arms;28 paths/56 prefixes/source and28 endpoint controls. Original coefficients,mu,absolute budget,costs and events. No transfer claim. Observation: 28 paths pass:ongoing preference loses public4/5 from V7 opening,turnover219%;opening loses3/5,negative worstSharpe. Conditional interactions/sign changes preclude an opening-only explanation or additive effects. Ongoing allocation improves but selection worsens. Interpretation: Outcome: replan retained: - V7 best20.53,original targets,all timing/full-preference synthetic gains and public/native/cost counterevidence. changed: - Test removal of within-class covariance-bonus heterogeneity;keep class-common preference and original decisions. stopped: - Opening-only schedule promotion;assigning losses solely to opening,warnings or independent additive effects.", + "tags": [ + "writeback-result" + ], + "title": "Do matched timing interventions distinguish opening effects from ongoing benchmark-preference losses under unchanged V7 risk and execution?" + } + ] +} diff --git a/apps/presentation/site/scripts/seo-smoke.mjs b/apps/presentation/site/scripts/seo-smoke.mjs index 6683b17106..1e2cbe7fcf 100644 --- a/apps/presentation/site/scripts/seo-smoke.mjs +++ b/apps/presentation/site/scripts/seo-smoke.mjs @@ -23,14 +23,14 @@ for (const [, href] of skill.matchAll(/\]\((https:\/\/[^)]+)\)/g)) { } } const titles = new Set(); -for (const [path, subject] of [["", "long-running"], ["benchmarks/swe-marathon/", "SWE-Marathon"], ["benchmarks/lhtb/", "LHTB"]]) { +for (const [path, subject] of [["", "long-running"], ["benchmarks/swe-marathon/", "SWE-Marathon"], ["benchmarks/lhtb/", "LHTB"], ["lab/time-machine/", "Time Machine"]]) { const html = await readFile(resolve(directory, path, "index.html"), "utf8"); const title = html.match(/(.*?)<\/title>/s)?.[1]; assert(title?.toLowerCase().includes(subject.toLowerCase()), `${path}: descriptive title`); assert(!titles.has(title), "page titles must be distinct"); titles.add(title); assert.equal((html.match(/<h1(?:\s|>)/g) ?? []).length, 1, `${path}: one static primary heading`); - assert((html.match(/<h2(?:\s|>)/g) ?? []).length >= 2, `${path}: real sections before JavaScript`); + assert((html.match(/<h2(?:\s|>)/g) ?? []).length >= (path === "lab/time-machine/" ? 1 : 2), `${path}: real sections before JavaScript`); assert(html.includes('id="root">') && !html.includes('id="root"></div>'), `${path}: populated React root`); assert.equal((html.match(/rel="canonical"/g) ?? []).length, 1); assert(html.includes(`rel="canonical" href="${origin}${path}"`), `${path}: canonical matches route`); @@ -42,6 +42,18 @@ for (const [path, subject] of [["", "long-running"], ["benchmarks/swe-marathon/" const structured = JSON.parse(html.match(/<script type="application\/ld\+json">(.*?)<\/script>/s)?.[1] ?? "null"); assert.equal(structured?.["@type"], "SoftwareSourceCode"); assert.equal(structured?.codeRepository, "https://github.com/loopx-project/loopx"); + assert(html.includes('lab/time-machine/'), "homepage discovers the interactive research experience"); + } + if (path === "lab/time-machine/") { + assert(html.includes("Local results improved."), "question exists before JavaScript"); + assert(html.includes("c858f51021dc1c1cf08263f9c04b692a5e29b40b"), "immutable archive remains accessible without JavaScript"); + const base = html.match(/<script\b[^>]*\bsrc="([^"]*)site-assets\/[^"]+"/)?.[1]; + assert(html.includes(`href="${base}"`), "return link preserves deployment base"); + const scenes = JSON.parse(await readFile(new URL("../src/time-machine/scenes.json", import.meta.url))); + for (const scene of scenes) { + assert.equal(await readFile(resolve(directory, path, "data", scene.file), "utf8"), + await readFile(new URL(`../public/lab/time-machine/data/${scene.file}`, import.meta.url), "utf8"), "capture copied unchanged"); + } } const description = html.match(/<meta name="description" content="([^"]+)"/)?.[1]; assert(description?.length > 50 && description.length < 240, `${path}: useful snippet`); @@ -68,4 +80,4 @@ if (process.argv.includes("--with-docs")) { const index = await readFile(resolve(directory, "sitemap.xml"), "utf8"); for (const [, url] of index.matchAll(/<loc>(.*?)<\/loc>/g)) await access(resolve(directory, url.slice(origin.length))); } -console.log(`SEO publication: agent skill and owner links, three prerendered routes, ${locations.length} canonical sitemap targets, assets and metadata passed`); +console.log(`SEO publication: agent skill and owner links, four prerendered routes, ${locations.length} canonical sitemap targets, assets and metadata passed`); diff --git a/apps/presentation/site/scripts/time-machine.test.mjs b/apps/presentation/site/scripts/time-machine.test.mjs new file mode 100644 index 0000000000..f3fabe662c --- /dev/null +++ b/apps/presentation/site/scripts/time-machine.test.mjs @@ -0,0 +1,64 @@ +import assert from "node:assert/strict"; +import {createHash} from "node:crypto"; +import {readFile} from "node:fs/promises"; +import test from "node:test"; +import {readPinnedCapture} from "../src/time-machine/records.ts"; + +const scenes = JSON.parse(await readFile(new URL("../src/time-machine/scenes.json", import.meta.url))); +const buffer = bytes => bytes.buffer.slice(bytes.byteOffset, bytes.byteOffset + bytes.byteLength); +const digest = bytes => createHash("sha256").update(bytes).digest("hex"); +const bytesFor = scene => readFile(new URL(`../public/lab/time-machine/data/${scene.file}`, import.meta.url)); + +test("all three pinned captures preserve original records and explicit evidence anchors", async () => { + assert.deepEqual(scenes.map(scene => scene.count), [76, 134, 105]); + for (const scene of scenes) { + const bytes = await bytesFor(scene); + const source = JSON.parse(bytes); + const rows = await readPinnedCapture(buffer(bytes), scene); + assert.deepEqual(rows, source.events, "the renderer does not rewrite source text, status or time"); + assert.equal(source.provenance.trajectory_publication.revision, "c858f51021dc1c1cf08263f9c04b692a5e29b40b"); + assert.equal(source.provenance.claim_scope, "diagnostic_only"); + assert.equal(source.provenance.score_countable, false); + assert.equal(source.source_counts.edge_count, 0); + assert.equal(source.source_counts.parent_records, 0); + const finding = rows.find(row => row.result_id === scene.evidenceId); + assert.equal(finding.event_kind, "finding"); + assert.equal(scene.en.choices.length, 2); + assert.equal(scene.zh.choices.length, 2); + assert.equal(scene.en.metrics.length, scene.zh.metrics.length); + } +}); + +test("different or edited bytes cannot inherit the curated scene's authority", async () => { + const [portfolio, graph] = await Promise.all(scenes.slice(0, 2).map(bytesFor)); + await assert.rejects(readPinnedCapture(buffer(graph), scenes[0]), /digest/); + const edited = Buffer.from(portfolio); + edited[10] ^= 1; + await assert.rejects(readPinnedCapture(buffer(edited), scenes[0]), /digest/); + await assert.rejects(readPinnedCapture(new ArrayBuffer(0), scenes[0]), /digest/); +}); + +test("artifact-shape failures reject even if an editor updates the digest", async () => { + const original = JSON.parse(await bytesFor(scenes[0])); + for (const mutate of [ + data => {data.events[0].schema_version = "invented_schema";}, + data => {data.events[0].recorded_at = "not a date";}, + data => {data.events[0].goal_id = "another-capture";}, + data => {data.events[1].event_id = data.events[0].event_id;}, + data => {data.events.splice(data.events.findIndex(row => row.result_id === scenes[0].evidenceId), 1);}, + data => {data.source_sha256 = "0".repeat(64);}, + ]) { + const data = structuredClone(original); mutate(data); + const bytes = Buffer.from(JSON.stringify(data)); + await assert.rejects(readPinnedCapture(buffer(bytes), {...scenes[0], sha256: digest(bytes)})); + } +}); + +test("a record reader preserves append order rather than inventing historical state", async () => { + const data = JSON.parse(await bytesFor(scenes[0])); + data.events[0].recorded_at = "2030-01-01T00:00:00Z"; + const bytes = Buffer.from(JSON.stringify(data)); + const rows = await readPinnedCapture(buffer(bytes), {...scenes[0], sha256: digest(bytes)}); + assert.deepEqual(rows, data.events); + assert.equal(rows[0].recorded_at, "2030-01-01T00:00:00Z"); +}); diff --git a/apps/presentation/site/src/App.tsx b/apps/presentation/site/src/App.tsx index ac580ebbbe..d8b1f4de9a 100644 --- a/apps/presentation/site/src/App.tsx +++ b/apps/presentation/site/src/App.tsx @@ -147,7 +147,7 @@ const content = { ["Research & evaluation", "LHTB × LoopX", "Compare five execution mechanisms across 46 long-horizon terminal tasks and inspect where recoverable state helps.", "Read the study"], ["Research · Chinese", "DeepSWE behavior discoveries", "Explore how domain hints affect implementation and validation in individual cases.", "Read the behavior analysis"], ["Research · Chinese", "DeepSWE × Sol", "Read the historical 113-task results, continuation mechanisms, and patch-delivery checks.", "Read the research brief"], - ["Cases", "All LoopX showcases", "Browse public cases, interactive walkthroughs, and their evidence boundaries.", "Browse all cases"], + ["Interactive research", "If you were there", "Make a choice for the Agent, then see three real research reversals and inspect their original records.", "Try the Time Machine"], ], developer: "Projection developer tools", }, @@ -266,7 +266,7 @@ const content = { ["研究与评测", "LHTB × LoopX", "对比五种执行机制在 46 个长程终端任务上的结果，观察可恢复状态在什么场景有效。", "阅读研究简报"], ["研究与评测", "DeepSWE 行为分析", "从具体案例观察领域提示如何影响实现选择与验证行为。", "阅读行为分析"], ["研究与评测", "DeepSWE × Sol", "解读 113 任务的历史结果、持续执行机制，以及补丁如何进入最终验收。", "阅读研究简报"], - ["案例", "完整案例目录", "浏览公开案例、交互式讲解及其证据边界。", "浏览全部案例"], + ["交互体验", "如果你在场", "先替 Agent 做一次选择，再看三个真实研究片段里的反转与原始记录。", "试试 LoopX 时光机"], ], developer: "开发者投影工具", }, @@ -1006,7 +1006,7 @@ export function App() { `benchmarks/lhtb/${language === "zh" ? "?lang=zh" : ""}`, "benchmarks/deepswe/behavior-discovery/", "benchmarks/deepswe-sol/", - `docs/showcases/index${language === "en" ? ".en" : ""}.html`, + `lab/time-machine/${language === "zh" ? "?lang=zh" : ""}`, ]; return ( <a className="learn-card resource-card" key={title} href={`${basePath}${paths[index]}`}> diff --git a/apps/presentation/site/src/main.tsx b/apps/presentation/site/src/main.tsx index 0d5329cbc7..45acd83c2c 100644 --- a/apps/presentation/site/src/main.tsx +++ b/apps/presentation/site/src/main.tsx @@ -1,4 +1,4 @@ -import { StrictMode } from "react"; +import { lazy, StrictMode, Suspense } from "react"; import { createRoot } from "react-dom/client"; import { App } from "./App"; import { LhtbBrief } from "./LhtbBrief"; @@ -6,6 +6,7 @@ import { SweMarathonBrief } from "./SweMarathonBrief"; import "./styles.css"; import "./swe-marathon-brief.css"; import "./lhtb-brief.css"; +import "./time-machine/time-machine.css"; const rootElement = document.getElementById("root"); if (!rootElement) throw new Error("LoopX homepage root is missing"); @@ -14,9 +15,12 @@ const pathSegments = window.location.pathname.split("/").filter(Boolean); if (pathSegments.at(-1) === "index.html") pathSegments.pop(); const isSweMarathonBrief = pathSegments.slice(-2).join("/") === "benchmarks/swe-marathon"; const isLhtbBrief = pathSegments.slice(-2).join("/") === "benchmarks/lhtb"; +const isTimeMachine = pathSegments.slice(-2).join("/") === "lab/time-machine"; +const TimeMachine = lazy(() => import("./time-machine/TimeMachine")); createRoot(rootElement).render( <StrictMode> - {isSweMarathonBrief ? <SweMarathonBrief /> : isLhtbBrief ? <LhtbBrief /> : <App />} + {isTimeMachine ? <Suspense fallback={<p role="status">Loading / 正在读取…</p>}><TimeMachine /></Suspense> + : isSweMarathonBrief ? <SweMarathonBrief /> : isLhtbBrief ? <LhtbBrief /> : <App />} </StrictMode>, ); diff --git a/apps/presentation/site/src/page-metadata.ts b/apps/presentation/site/src/page-metadata.ts index 08f9e97ceb..3860143fbc 100644 --- a/apps/presentation/site/src/page-metadata.ts +++ b/apps/presentation/site/src/page-metadata.ts @@ -1,6 +1,17 @@ // Public URLs are independent of the local preview's deployment base. export const siteUrl = "https://loopx-project.github.io/loopx/"; export const pageMetadata = { + timeMachine: { + path: "lab/time-machine/", + en: { + title: "LoopX Time Machine — If you were there", + description: "Make three choices for a research Agent, then compare them with real recorded outcomes. Inspect the original evidence, failed routes and remaining uncertainty.", + }, + zh: { + title: "LoopX 时光机 — 如果你在场", + description: "先替研究 Agent 做选择，再看真实记录里的结果。逐条阅读原始证据、失败路线和仍未回答的问题，体验长程研究如何延续。", + }, + }, home: { path: "", en: { diff --git a/apps/presentation/site/src/render-static.tsx b/apps/presentation/site/src/render-static.tsx index 0b9fd6d8ed..d1339ee622 100644 --- a/apps/presentation/site/src/render-static.tsx +++ b/apps/presentation/site/src/render-static.tsx @@ -2,10 +2,11 @@ import { renderToString } from "react-dom/server"; import { App } from "./App"; import { LhtbBrief } from "./LhtbBrief"; import { SweMarathonBrief } from "./SweMarathonBrief"; +import TimeMachine from "./time-machine/TimeMachine"; import { pageMetadata, siteUrl, type PublicPage } from "./page-metadata"; export { pageMetadata, siteUrl }; export function render(page: PublicPage) { - return renderToString(page === "home" ? <App /> : page === "lhtb" ? <LhtbBrief /> : <SweMarathonBrief />); + return renderToString(page === "timeMachine" ? <TimeMachine /> : page === "home" ? <App /> : page === "lhtb" ? <LhtbBrief /> : <SweMarathonBrief />); } diff --git a/apps/presentation/site/src/time-machine/TimeMachine.tsx b/apps/presentation/site/src/time-machine/TimeMachine.tsx new file mode 100644 index 0000000000..69ec43e2d0 --- /dev/null +++ b/apps/presentation/site/src/time-machine/TimeMachine.tsx @@ -0,0 +1,177 @@ +import {useEffect, useRef, useState} from "react"; +import {ArrowLeft, ArrowRight, ArrowUpRight} from "lucide-react"; +import {usePublicPageNavigation} from "../usePublicPageNavigation"; +import {captureRevision, datasetUrl, readPinnedCapture, type CapturedRecord} from "./records"; +import scenes from "./scenes.json"; + +const strings = { + en: { + back: "LoopX home", name: "Time Machine", + title: "If you were there,\nwould you let it continue?", + deck: "Make a choice for the Agent. Then see what the real record says.", + caseNav: "Choose a research episode", count: "Episodes explored", stage: "Your decision", + locked: "Choose first. Then see what happened.", lockedNote: "No quiz score. Just evidence worth questioning.", + loading: "Reading the original capture…", error: "This capture could not be verified. No result is shown.", retry: "Try again", + next: "Try another episode", chosen: "Your choice", choiceNote: "Both choices reveal the same recorded outcome; no alternative outcome is simulated.", + boundary: "Curated from public solver records. Source labels are the solver’s claims, not independent replication or evidence of a LoopX performance gain.", + local: "Your choices stay in this page’s memory. Reload starts over. No Agent is started.", + reader: "Read the original records", openRecord: "Inspect this finding", readerHelp: "A single-record reader in original append order. It does not restore a research state or show everything known at a past time.", + cursor: "Original record number", record: "Record", previous: "Previous record", following: "Next record", original: "Original text · untranslated", identifiers: "Source identifiers", + sources: "Fixed public source", fixture: "This exact capture", totals: "76 + 134 + 105 = 315 Explore records. Record counts are not experiment counts.", + close: "Back to the question", noJs: "Enable JavaScript to make choices and browse records. The fixed public archive is also available directly.", + lesson: "What stays with the next decision?", + }, + zh: { + back: "LoopX 首页", name: "时光机", + title: "如果你在场，\n会让它继续吗？", deck: "先替 Agent 做一次选择，再看看真实记录里发生了什么。", + caseNav: "选择研究案例", count: "已探索片段", stage: "你的判断", locked: "先做选择，再看结果。", lockedNote: "没有答题分数，只有值得追问的证据。", + loading: "正在读取原始记录…", error: "无法核验这份记录，当前不显示结果。", retry: "重试", + next: "换一个案例", chosen: "你选择了", choiceNote: "两种选择都展示同一段实际经过，不模拟另一种选择的结果。", + boundary: "三道题编排自公开研究记录。来源中的状态是原 solver 的声明，未经这里独立复验，也不证明 LoopX 带来性能提升。", + local: "选择只保留在当前页面内存，刷新后重新开始，不会启动 Agent。", + reader: "自己翻原记录", openRecord: "核对这条发现", readerHelp: "按原追加顺序逐条阅读；不恢复研究状态，也不表示某个时点已知的全部知识。", + cursor: "原始记录序号", record: "记录", previous: "上一条记录", following: "下一条记录", original: "原文 · 未翻译", identifiers: "来源标识", + sources: "固定版本公开来源", fixture: "这份原始记录", totals: "76 + 134 + 105 = 315 条 Explore 记录，记录数量不是实验次数。", + close: "回到问题", noJs: "启用 JavaScript 后可做选择、浏览原记录，也可以直接打开固定版本的公开来源。", lesson: "下一次判断，能留下什么？", + }, +}; +const readCase = () => { + const id = typeof window === "undefined" ? "" : new URLSearchParams(window.location.search).get("case"); + return Math.max(0, scenes.findIndex(scene => scene.id === id)); +}; + +export default function TimeMachine() { + const [language, setLanguage] = usePublicPageNavigation("timeMachine"); + const [current, setCurrent] = useState(readCase); + const [answers, setAnswers] = useState<Record<string, number>>({}); + const [loaded, setLoaded] = useState<{id: string; rows: CapturedRecord[]} | null>(null); + const [failedFor, setFailed] = useState<string | null>(null); + const [attempt, setAttempt] = useState(0); + const [position, setPosition] = useState(0); + const resultTitle = useRef<HTMLHeadingElement>(null); + const questionTitle = useRef<HTMLHeadingElement>(null); + const reader = useRef<HTMLDetailsElement>(null); + const original = useRef<HTMLHeadingElement>(null); + const pendingFocus = useRef(false); + const c = strings[language]; + const scene = scenes[current]; + const records = loaded?.id === scene.id ? loaded.rows : null; + const failed = failedFor === scene.id; + const copy = scene[language]; + const answer = answers[scene.id]; + const chosen = answer !== undefined; + const basePath = import.meta.env.BASE_URL; + const fileUrl = `${basePath}lab/time-machine/data/${scene.file}`; + const record = records?.[position]; + + useEffect(() => { + const restore = () => setCurrent(readCase()); + window.addEventListener("popstate", restore); + return () => window.removeEventListener("popstate", restore); + }, []); + + useEffect(() => { + const abort = new AbortController(); + let active = true; + setLoaded(null); setFailed(null); setPosition(0); + if (reader.current) reader.current.open = false; + void (async () => { + const response = await fetch(fileUrl, {signal: abort.signal}); + if (!response.ok) throw new Error(`Capture HTTP ${response.status}`); + const rows = await readPinnedCapture(await response.arrayBuffer(), scene); + if (active) setLoaded({id: scene.id, rows}); + })().catch(() => {if (active) setFailed(scene.id);}); + return () => {active = false; abort.abort();}; + }, [fileUrl, scene, attempt]); + + useEffect(() => { + if (chosen && records && pendingFocus.current) { + pendingFocus.current = false; + resultTitle.current?.focus({preventScroll: true}); + resultTitle.current?.closest(".tm-result")?.scrollIntoView({behavior: "instant", block: "nearest"}); + } + }, [chosen, records]); + + function select(index: number) { + if (index === current) return; + // Clear synchronously: no previous case may appear under a new question. + setLoaded(null); setFailed(null); setPosition(0); pendingFocus.current = false; + setCurrent(index); + const url = new URL(window.location.href); + url.searchParams.set("case", scenes[index].id); + url.hash = ""; + window.history.pushState(window.history.state, "", url); + window.requestAnimationFrame(() => questionTitle.current?.focus({preventScroll: true})); + } + function choose(index: number) { + pendingFocus.current = true; + setAnswers(previous => ({...previous, [scene.id]: index})); + if (records) setPosition(records.findIndex(row => row.result_id === scene.evidenceId)); + } + function inspect() { + if (!records || !reader.current) return; + setPosition(records.findIndex(row => row.result_id === scene.evidenceId)); + reader.current.open = true; + window.requestAnimationFrame(() => original.current?.focus()); + } + + return <div className="tm-page"> + <a className="skip-link" href="#tm-question">{c.stage}</a> + <header className="tm-header"> + <a href={`${basePath}${language === "zh" ? "?lang=zh" : ""}`} className="brand" aria-label={c.back}> + <span className="product-mark" aria-hidden="true"><span/><span/></span>LoopX <span className="tm-brand-suffix">/ {c.name}</span> + </a> + <div className="tm-language" aria-label="Language"> + <button type="button" aria-pressed={language === "en"} onClick={() => setLanguage("en")}>EN</button> + <button type="button" aria-pressed={language === "zh"} onClick={() => setLanguage("zh")}>中文</button> + </div> + </header> + <main className="tm-shell"> + <div className="tm-intro"><div><p className="tm-eyebrow">IF YOU WERE THERE</p><h1>{c.title}</h1><p>{c.deck}</p></div> + <p className="tm-counter">{c.count} {Object.keys(answers).length} / 3</p> + </div> + <nav className="tm-cases" aria-label={c.caseNav}>{scenes.map((item, index) => + <button key={item.id} type="button" aria-pressed={index === current} onClick={() => select(index)}>{item[language].label}</button>)}</nav> + <section className="tm-stage" aria-label={c.stage}> + <div className="tm-question"><div><p className="tm-tag">{copy.tag}</p> + <h2 ref={questionTitle} id="tm-question" tabIndex={-1}>{copy.question}</h2><p>{copy.intro}</p></div> + <div className="tm-choices">{copy.choices.map((choice, index) => + <button type="button" key={index} disabled={!records} aria-pressed={chosen && answer === index} onClick={() => choose(index)}> + {choice}<ArrowUpRight size={18} aria-hidden="true"/> + </button>)}</div> + </div> + <div className="tm-result" aria-busy={!records && !failed}> + {failed ? <div className="tm-error"><p role="alert">{c.error}</p><button type="button" onClick={() => setAttempt(value => value + 1)}>{c.retry}</button></div> : !records ? + <p role="status">{c.loading}</p> : !chosen ? <div className="tm-locked"><span aria-hidden="true">?↗</span><h3>{c.locked}</h3><p>{c.lockedNote}</p></div> : + <div className="tm-revealed"><small>{copy.status}</small><h3 ref={resultTitle} tabIndex={-1}>{copy.title}</h3><p>{copy.body}</p> + <dl className="tm-metrics">{copy.metrics.map(([label, value]) => <div key={label}><dt>{label}</dt><dd>{value}</dd></div>)}</dl> + <p className="tm-your-choice">{c.chosen}: {copy.choices[answer]}. {c.choiceNote}</p> + <div className="tm-result-actions"><button type="button" onClick={inspect}>{c.openRecord}</button> + <button type="button" onClick={() => select((current + 1) % scenes.length)}>{c.next}<ArrowRight size={16} aria-hidden="true"/></button></div> + </div>} + </div> + </section> + <div className="tm-boundary"><p>{c.boundary}</p><p>{chosen && records ? copy.lesson : c.local}</p></div> + <details className="tm-reader" ref={reader}><summary>{c.reader}<span>{scene.count} {c.record}</span></summary> + <p>{c.readerHelp}</p> + {!records ? <p role="status">{failed ? c.error : c.loading}</p> : <> + <div className="tm-reader-controls"><label htmlFor="tm-cursor">{c.cursor} <strong>{position + 1} / {records.length}</strong></label> + <input id="tm-cursor" type="range" min="1" max={records.length} value={position + 1} onChange={event => setPosition(Number(event.target.value) - 1)}/> + <div><button type="button" disabled={position === 0} onClick={() => setPosition(value => value - 1)}><ArrowLeft size={16} aria-hidden="true"/>{c.previous}</button> + <button type="button" disabled={position === records.length - 1} onClick={() => setPosition(value => value + 1)}>{c.following}<ArrowRight size={16} aria-hidden="true"/></button></div> + </div> + {record && <article className="tm-original"><small>{c.original}</small><p className="tm-record-meta"><time>{record.recorded_at}</time> · {record.status} · {record.event_kind}</p> + <h3 ref={original} tabIndex={-1}>{record.title || c.record}</h3><p className="tm-record-body">{record.summary}</p> + <details><summary>{c.identifiers}</summary><code>{record.result_id}<br/>{record.event_id}<br/>{captureRevision}</code></details> + </article>} + <a className="tm-source" href={fileUrl} target="_blank" rel="noopener">{c.fixture}<ArrowUpRight size={14} aria-hidden="true"/></a> + </>} + <button className="tm-return" type="button" onClick={() => {if (reader.current) reader.current.open = false; questionTitle.current?.focus();}}><ArrowLeft size={14} aria-hidden="true"/>{c.close}</button> + </details> + {Object.keys(answers).length === scenes.length && <section className="tm-takeaway"><h2>{c.lesson}</h2><p>{language === "zh" ? "失败的路线、有效的范围、尚未回答的问题。长程研究需要把这些一起带进下一轮。" : "Failed routes, useful scope, and unanswered questions. Long-running research needs to carry all three into the next round."}</p> + <a href={`${basePath}docs/book/${language === "en" ? "en/" : ""}`}>{language === "zh" ? "了解 LoopX 如何延续工作" : "See how LoopX continues work"}<ArrowUpRight size={16} aria-hidden="true"/></a></section>} + <footer className="tm-footer"><span>{c.totals}</span><a href={datasetUrl} target="_blank" rel="noopener" referrerPolicy="strict-origin-when-cross-origin">{c.sources}<ArrowUpRight size={14} aria-hidden="true"/></a></footer> + <noscript><p>{c.noJs}</p></noscript> + </main> + </div>; +} diff --git a/apps/presentation/site/src/time-machine/records.ts b/apps/presentation/site/src/time-machine/records.ts new file mode 100644 index 0000000000..cbfb8a6f19 --- /dev/null +++ b/apps/presentation/site/src/time-machine/records.ts @@ -0,0 +1,44 @@ +/** Verbatim public capture reader, not an Explore state reducer or validator. + * Explore owns v0 semantics. Pinned bytes prevent a different capture or edited + * record from silently acquiring these editorial scenes. Status stays source text. + */ +export interface CapturedRecord { + schema_version: string; + goal_id: string; + event_id: string; + event_kind: string; + result_id: string; + recorded_at: string; + status: string; + title?: string; + summary: string; +} + +interface CapturePin {count: number; sha256: string; sourceHash: string; evidenceId: string} +export async function readPinnedCapture(bytes: ArrayBuffer, pin: CapturePin): Promise<CapturedRecord[]> { + const digest = Array.from(new Uint8Array(await crypto.subtle.digest("SHA-256", bytes)), + byte => byte.toString(16).padStart(2, "0")).join(""); + if (digest !== pin.sha256) throw new Error("Capture digest mismatch"); + const capture = JSON.parse(new TextDecoder().decode(bytes)); + if (capture.source_sha256 !== pin.sourceHash || !Array.isArray(capture.events) || + capture.events.length !== pin.count) throw new Error("Wrong capture"); + const seen = new Set<string>(); + for (const event of capture.events) { + if (!event || event.schema_version !== "loopx_explore_result_event_v0" || + event.goal_id !== capture.goal_id || + ["event_id", "event_kind", "result_id", "recorded_at", "status", "summary"].some( + key => typeof event[key] !== "string") || + !Number.isFinite(Date.parse(event.recorded_at)) || seen.has(event.event_id) || + (event.title !== undefined && typeof event.title !== "string")) throw new Error("Invalid capture record"); + seen.add(event.event_id); + } + if (!capture.events.some((event: CapturedRecord) => event.result_id === pin.evidenceId)) { + throw new Error("Scene evidence is missing"); + } + // Original append order is the record reader's order, even for equal or + // out-of-order timestamps. No latest-state, historical knowledge or causal claim. + return capture.events; +} + +export const captureRevision = "44c7af3db271ab1991428d3ff1991ca922434eff"; +export const datasetUrl = "https://huggingface.co/datasets/huangrt01/loopx-edgebench-research-checkpoints-20261015/tree/c858f51021dc1c1cf08263f9c04b692a5e29b40b"; diff --git a/apps/presentation/site/src/time-machine/scenes.json b/apps/presentation/site/src/time-machine/scenes.json new file mode 100644 index 0000000000..4e34ba639a --- /dev/null +++ b/apps/presentation/site/src/time-machine/scenes.json @@ -0,0 +1,170 @@ +[ + { + "id": "portfolio", + "file": "portfolio.json", + "count": 76, + "sha256": "8966c1e8d37f38f410886d25bcddb939a47d3dd0f7e9aed7eee95925b4330fea", + "sourceHash": "b876dd7660e0aab19db717f60e86865494a74b8fa27b42a5a67ee295ef524397", + "evidenceId": "finding_5b6717a721136e001cd0cd18d38ac167431bbe265445e669bac92c72d990aef2", + "en": { + "label": "Portfolio", + "tag": "01 / A better-looking plan", + "question": "Local results improved.\nReplace the main strategy?", + "intro": "83 local paths and three standalone suites pass. Several local Sharpe metrics improve. Would you adopt this version?", + "choices": [ + "Adopt this version", + "Check the native evaluation" + ], + "title": "The local good news did not survive the native evaluation.", + "body": "The solver recorded 13.84 for the new version, against 20.53 for retained V7. It restored V7 and kept the failed experiment as evidence for the next question.", + "metrics": [ + [ + "New version · recorded", + "13.84" + ], + [ + "Retained V7 · recorded", + "20.53" + ] + ], + "lesson": "Progress must return to the original objective.", + "status": "Source label: refuted" + }, + "zh": { + "label": "投资组合", + "tag": "01 / 看起来更好的方案", + "question": "本地表现更好了。\n要替换主方案吗？", + "intro": "83 条本地路径与三组独立测试通过，多个本地 Sharpe 指标也更好。轮到你决定是否采用。", + "choices": [ + "采用这个版本", + "先看真实评估" + ], + "title": "局部的好消息，没有变成最终收益。", + "body": "原记录报告：新版本在原生评估中得到 13.84，保留版本是 20.53。Agent 恢复了旧版本，并保留这次失败带来的线索。", + "metrics": [ + [ + "新版本 · 原记录", + "13.84" + ], + [ + "保留版本 · 原记录", + "20.53" + ] + ], + "lesson": "进展要回到原目标上验证。", + "status": "来源记录：否证" + } + }, + { + "id": "graph", + "file": "graph.json", + "count": 134, + "sha256": "254a7644c725c2a8e6fde59e617aea739e92c22fa381673945f35af2f5a33f29", + "sourceHash": "b105de360ed7d015ae5d7bf96b545879e12507b1fc97eaeb5fb6fa2c0bbc00ab", + "evidenceId": "finding_b5860e437b2905fbc0b9db2bbdc3ba567783817eee67e0ecc433bf4e65c00b1a", + "en": { + "label": "Graph", + "tag": "02 / A tempting improvement", + "question": "Accuracy jumps.\nIs that success?", + "intro": "On two generated-graph comparisons: 41.2% → 85.0% and 61.9% → 88.2%. Is that enough to adopt the new approach?", + "choices": [ + "Adopt the approach", + "Check its scope first" + ], + "title": "It answered one question. Others stayed open.", + "body": "The same record reports 32% more time on the independent-edge control, with accuracy unchanged there. Improvement in the native score remains unproven. A useful finding can retain its scope and uncertainty.", + "metrics": [ + [ + "Generated graph · one comparison", + "41.2 → 85.0%" + ], + [ + "Independent-edge control · time", + "+32%" + ] + ], + "lesson": "A valuable finding can still carry an unknown.", + "status": "Source label: confirmed; native gain unproven" + }, + "zh": { + "label": "图分类", + "tag": "02 / 一个诱人的提升", + "question": "准确率大幅提高。\n就算成功了吗？", + "intro": "生成图上的两个对照：41.2% → 85.0%，61.9% → 88.2%。这足够让你为新方案按下采用吗？", + "choices": [ + "这次值得采用", + "再检查适用范围" + ], + "title": "它回答了一个问题，还没回答全部问题。", + "body": "原记录同时报告：独立边控制组的时间成本增加 32%；原生评分是否改善尚未证明。这条发现可以保留，但适用范围要跟着它一起留下。", + "metrics": [ + [ + "生成图 · 对照一", + "41.2 → 85.0%" + ], + [ + "控制组 · 时间成本", + "+32%" + ] + ], + "lesson": "一个有价值的发现，可以同时带着未知。", + "status": "来源记录：确认；原生收益未证" + } + }, + { + "id": "ad", + "file": "ad.json", + "count": 105, + "sha256": "e0b4f6f5c4adbfd81ba0bb81c188794b68c6d01fd25852904404e9de7f35086e", + "sourceHash": "de04b679e9f382b650e891e5c7dd7146bc690500435c511f107c9d24da4f4efa", + "evidenceId": "finding_c652328314cf9ac2757f85dfcbdb9907bb719cf1a1e36185926640694663a9f2", + "en": { + "label": "Ad placement", + "tag": "03 / After the first win", + "question": "The screening run won.\nRoll out the trick?", + "intro": "Give the highest-loss ads optimization priority. It sounds reasonable, and six screening seeds looked better too.", + "choices": [ + "Roll out the ordering", + "Try fresh samples first" + ], + "title": "Fresh samples changed the first impression.", + "body": "High-loss-first ordering won screening but lost on the pooled score of 32 fresh seeds. The solver stopped adopting that ordering, kept its current best and moved to a new hypothesis.", + "metrics": [ + [ + "Screening · recorded", + "6 seeds" + ], + [ + "Fresh samples · recorded", + "32 seeds" + ] + ], + "lesson": "Creativity includes letting go of a beautiful idea.", + "status": "Source label: tentative" + }, + "zh": { + "label": "广告优化", + "tag": "03 / 第一次赢了以后", + "question": "小规模筛选赢了。\n要推广这个技巧吗？", + "intro": "让损失最大的广告先获得优化机会。这个直觉很合理，六个筛选种子上的表现也更好。", + "choices": [ + "把它推广出去", + "换一批样本再看" + ], + "title": "新的一批样本，改写了第一印象。", + "body": "原记录报告：优先处理高损失广告的排序，在筛选中获胜，却在后续 32 个新种子的合并评分中退化。Agent 停止采用这个排序，保留当前方案，转向新的假设。", + "metrics": [ + [ + "筛选样本 · 原记录", + "6 个种子" + ], + [ + "新样本 · 原记录", + "32 个种子" + ] + ], + "lesson": "创造力也包括：及时放下一个漂亮的想法。", + "status": "来源记录：暂定" + } + } +] diff --git a/apps/presentation/site/src/time-machine/time-machine.css b/apps/presentation/site/src/time-machine/time-machine.css new file mode 100644 index 0000000000..d6bed02d9b --- /dev/null +++ b/apps/presentation/site/src/time-machine/time-machine.css @@ -0,0 +1,89 @@ +.tm-page { color: var(--ink); line-height: 1.7; } +.tm-header { max-width: 1200px; margin: auto; padding: 0 28px; height: 80px; display: flex; align-items: center; justify-content: space-between; border-bottom: 1px solid var(--border); gap: 12px; } +.tm-brand-suffix { font-size: 14px; font-weight: 400; color: var(--body); letter-spacing: 0; } +.tm-language { display: flex; gap: 4px; } +.tm-language button { min-height: 44px; min-width: 44px; padding: 8px; border: 0; background: none; color: var(--body); cursor: pointer; } +.tm-language button[aria-pressed=true] { color: var(--ink); text-decoration: underline; text-underline-offset: 6px; } +.tm-page :is(button, a, input, summary):focus-visible { outline: 2px solid var(--blue); outline-offset: 4px; } +.tm-page :is(h2, h3):focus { outline: none; } +.tm-shell { max-width: 1120px; margin: auto; padding: 48px 28px 32px; } +.tm-intro { display: flex; align-items: end; justify-content: space-between; gap: 32px; margin-bottom: 32px; } +.tm-eyebrow { font: 12px/1.5 var(--font-mono); letter-spacing: 1.4px; color: var(--blue); margin: 0; } +.tm-intro h1 { font-size: clamp(32px, 3.5vw, 44px); line-height: 1.25; letter-spacing: -.045em; font-weight: 600; margin: 12px 0 16px; white-space: pre-line; } +.tm-intro p:not(.tm-eyebrow) { font-size: 14px; color: var(--body); margin: 0; } +.tm-counter { font: 12px/1.5 var(--font-mono); white-space: nowrap; padding-bottom: 8px; } +.tm-cases { display: flex; gap: 24px; border-bottom: 1px solid var(--border); } +.tm-cases button { min-height: 52px; border: 0; border-bottom: 2px solid transparent; background: none; padding: 10px 0; color: var(--body); font-size: 14px; cursor: pointer; } +.tm-cases button[aria-pressed=true] { color: var(--ink); font-weight: 600; border-bottom-color: var(--ink); } +.tm-stage { display: grid; grid-template-columns: 1fr 1fr; background: var(--surface); border: 1px solid var(--border); border-top: 0; min-height: 400px; border-radius: 0 0 16px 16px; overflow: clip; } +.tm-question { padding: 32px 40px; display: flex; flex-direction: column; align-items: start; justify-content: space-between; gap: 24px; } +.tm-tag { font: 12px/1.5 var(--font-mono); color: var(--body); margin: 0; } +.tm-question h2 { font-size: 28px; line-height: 1.45; font-weight: 600; letter-spacing: -.025em; white-space: pre-line; margin: 10px 0 16px; scroll-margin-top: 24px; } +.tm-question p:not(.tm-tag) { font-size: 14px; color: var(--body); margin: 0; } +.tm-choices { display: grid; gap: 10px; width: 100%; } +.tm-choices button { min-height: 48px; text-align: left; border-radius: 6px; background: var(--ink); color: var(--surface); border: 1px solid var(--ink); padding: 10px 16px; display: flex; justify-content: space-between; align-items: center; gap: 12px; font-size: 14px; cursor: pointer; } +.tm-choices button+button { background: var(--surface); color: var(--ink); border-color: var(--border); } +.tm-choices button[aria-pressed=true] { outline: 2px solid var(--blue); outline-offset: 2px; } +.tm-page button:disabled { cursor: default; opacity: .5; } +.tm-result { padding: 32px 36px; background: var(--soft); border-left: 1px solid var(--border); display: flex; flex-direction: column; justify-content: center; font-size: 14px; } +.tm-locked { text-align: center; } +.tm-locked>span { display: block; font: 80px/1.2 var(--font-mono); letter-spacing: -7px; color: var(--faint); } +.tm-locked h3 { font-size: 16px; font-weight: 500; margin: 24px 0 8px; } +.tm-locked p { font-size: 12px; color: var(--body); } +.tm-revealed { animation: tm-appear 200ms ease-out; } +.tm-revealed small { font-size: 12px; color: var(--body); } +.tm-revealed h3 { font-size: 22px; line-height: 1.4; letter-spacing: -.025em; margin: 12px 0; } +.tm-revealed p { color: var(--body); margin: 0 0 16px; } +.tm-metrics { display: grid; grid-template-columns: 1fr 1fr; gap: 16px; margin: 20px 0; border-top: 1px solid var(--border); padding-top: 16px; } +.tm-metrics dt { font-size: 12px; color: var(--body); } +.tm-metrics dd { font: 22px/1.5 var(--font-mono); letter-spacing: -.03em; margin: 4px 0 0; } +.tm-your-choice { font-size: 12px; } +.tm-result-actions { display: flex; flex-wrap: wrap; justify-content: space-between; gap: 0 16px; } +.tm-page :is(.tm-result-actions, .tm-error, .tm-reader-controls)>button, .tm-result-actions button, .tm-reader-controls button, .tm-return { border: 0; background: none; color: var(--blue); min-height: 44px; padding: 8px 0; text-align: left; font-size: 13px; display: inline-flex; gap: 8px; align-items: center; cursor: pointer; } +.tm-boundary { display: flex; justify-content: space-between; gap: 32px; padding: 20px 0; color: var(--body); font-size: 12px; border-bottom: 1px solid var(--border); } +.tm-boundary p { margin: 0; } +.tm-boundary p:first-child { max-width: 650px; flex: 2; } +.tm-boundary p:last-child { flex: 1; } +.tm-reader { padding: 12px 0; } +.tm-reader>summary { min-height: 48px; cursor: pointer; font-size: 15px; font-weight: 500; padding: 12px 0; } +.tm-reader>summary>span { float: right; font: 12px/2 var(--font-mono); color: var(--body); } +.tm-reader>p { max-width: 720px; font-size: 13px; color: var(--body); } +.tm-reader-controls label { display: flex; justify-content: space-between; color: var(--body); font-size: 12px; } +.tm-reader-controls input { display: block; width: 100%; min-height: 44px; accent-color: var(--ink); margin: 0; } +.tm-reader-controls>div { display: flex; justify-content: space-between; } +.tm-original { background: var(--surface); border: 1px solid var(--border); border-radius: 12px; padding: 24px; overflow-wrap: anywhere; } +.tm-original small, .tm-record-meta { font: 12px/1.8 var(--font-mono); color: var(--body); } +.tm-original h3 { font-size: 15px; font-weight: 500; margin: 12px 0; scroll-margin-top: 24px; } +.tm-record-body { font: 13px/1.8 var(--font-mono); color: var(--body); white-space: pre-wrap; margin: 12px 0; } +.tm-original details { color: var(--body); font-size: 12px; } +.tm-original summary { min-height: 44px; padding: 10px 0; cursor: pointer; } +.tm-original code { display: block; font: 12px/1.8 var(--font-mono); } +.tm-source { display: inline-flex; gap: 6px; align-items: center; min-height: 44px; color: var(--blue); font-size: 13px; margin-right: 24px; } +.tm-takeaway { border-top: 1px solid var(--border); padding: 28px 0; max-width: 700px; } +.tm-takeaway h2 { font-size: 24px; letter-spacing: -.03em; } +.tm-takeaway p { font-size: 14px; color: var(--body); } +.tm-takeaway a { color: var(--blue); display: inline-flex; gap: 8px; align-items: center; min-height: 44px; font-size: 14px; } +.tm-footer { display: flex; gap: 20px; justify-content: space-between; align-items: center; padding: 20px 0 0; border-top: 1px solid var(--border); color: var(--body); font-size: 12px; } +.tm-footer a { display: inline-flex; align-items: center; gap: 4px; min-height: 44px; text-decoration: underline; text-underline-offset: 3px; } +@keyframes tm-appear { from { opacity: 0; transform: translateY(4px); } to { opacity: 1; transform: none; } } +@media (prefers-reduced-motion: reduce) { .tm-page *, html:has(.tm-page) { animation: none; scroll-behavior: auto; } } +@media (max-width: 700px) { + .tm-header { height: 64px; padding: 0 20px; } + .tm-header .brand { font-size: 16px; gap: 6px; } + .tm-brand-suffix { font-size: 12px; } + .tm-shell { padding: 28px 20px 24px; } + .tm-intro { display: block; margin-bottom: 20px; } + .tm-intro h1 { font-size: 32px; } + .tm-counter { padding-top: 12px; } + .tm-cases { gap: 24px; } + .tm-stage { grid-template-columns: 1fr; } + .tm-question, .tm-result { padding: 24px; } + .tm-question h2 { font-size: 24px; } + .tm-result { border-left: 0; border-top: 1px solid var(--border); min-height: 220px; } + .tm-locked>span { font-size: 56px; } + .tm-locked h3 { margin: 16px 0 8px; } + .tm-metrics dd { font-size: 20px; } + .tm-boundary { flex-direction: column; gap: 12px; } + .tm-footer { flex-direction: column; align-items: start; gap: 4px; } + .tm-original { padding: 20px; } +}