From 03b28a178d4cdbfa0880f394c5b657341f5a1871 Mon Sep 17 00:00:00 2001 From: Timothy Wayne Gregg Date: Sat, 3 Oct 2026 08:44:18 -0400 Subject: [PATCH] Record owner-approved hierarchy software and live tracking split --- docs/HIERARCHY_LIVE_PREPARATION.md | 7 ++++++- docs/HIERARCHY_RELEASE_VERIFICATION.md | 10 ++++++---- 2 files changed, 12 insertions(+), 5 deletions(-) diff --git a/docs/HIERARCHY_LIVE_PREPARATION.md b/docs/HIERARCHY_LIVE_PREPARATION.md index 3b45899..afc704b 100644 --- a/docs/HIERARCHY_LIVE_PREPARATION.md +++ b/docs/HIERARCHY_LIVE_PREPARATION.md @@ -64,7 +64,7 @@ rechecked quotes, the enforced cap, and the consents, then bind it with matches the protocol, the current software parameters, and the same teacher endpoint. See [the study guide](HIERARCHY_STUDY.md) for the exact commands. -## Prepared proposal for issue #20 +## Prepared live proposal tracked in issue #74 The 2026-09-30 local preparation retained 14,960 rows: 10,451 train, 917 validation, 902 calibration, and 2,690 test. It excluded 484 rows without the @@ -103,3 +103,8 @@ approved. Only train examples and train traces would be eligible for teacher sharing; validation, calibration, and test remain excluded. Selection approval does not authorize the held-out phase. The teacher's moving alias must be rechecked and recorded before selection; native Jev remains pinned. + +The 2026-10-03 owner decision separates delivered software in #20/#21 from the +original live requirements retained in #74 and epic #6. The proposal above remains +unapproved for execution. Closing the software trackers grants no spending, data +sharing, semantic review or held-out test permission. diff --git a/docs/HIERARCHY_RELEASE_VERIFICATION.md b/docs/HIERARCHY_RELEASE_VERIFICATION.md index b54b129..475ccdf 100644 --- a/docs/HIERARCHY_RELEASE_VERIFICATION.md +++ b/docs/HIERARCHY_RELEASE_VERIFICATION.md @@ -1,6 +1,6 @@ # Hierarchy v1 release verification -This is the software evidence map for epic [#6](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/6), current through 2026-09-30. It does not approve deployment or claim a measured Jev gain. Check the linked issues and PRs for current hosted status before release. +This is the software evidence map for epic [#6](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/6), tracking refreshed on 2026-10-03. It does not approve deployment or claim a measured Jev gain. Check the linked issues and PRs for current hosted status before release. | Scope | Issue | Merged implementation | |---|---|---| @@ -17,8 +17,8 @@ This is the software evidence map for epic [#6](https://github.com/CompleteTech- | GEPA wording search | [#17](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/17) | [#32](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/32) | | Public CLI, API, and package | [#18](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/18) | [#33](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/33) | | Synthetic examples and traces | [#19](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/19) | [#34](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/34) | -| Offline preregistered study runner and holdout gates | [#20](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/20), **open** | [#35](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/35), [#37](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/37), [#38](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/38), [#41](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/41); a completed independent review of a real live manifest remains open | -| Integration and final release gate | [#21](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/21), **open** | [#36](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/36), merged; final all-child acceptance remains open | +| Offline preregistered study runner and holdout gates | [#20](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/20), **software-complete by owner decision** | [#35](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/35), [#37](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/37), [#38](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/38), [#41](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/41); a completed independent review of a real live manifest remains open in [#74](https://github.com/Jev-Engineering/TypeWright/issues/74) | +| Integration and final release gate | [#21](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/issues/21), **software-complete by owner decision** | [#36](https://github.com/CompleteTech-LLC-AI-Research/systemone-compiler/pull/36), merged; final empirical all-child acceptance remains open in [#74](https://github.com/Jev-Engineering/TypeWright/issues/74) and epic [#6](https://github.com/Jev-Engineering/TypeWright/issues/6) | ## Verification boundaries @@ -26,4 +26,6 @@ This is the software evidence map for epic [#6](https://github.com/CompleteTech- The CI matrix runs pytest, Ruff, wheel build, installed-wheel flat and hierarchy no-key demos, and runtime import isolation on Python 3.11–3.13. Its optional job installs the pinned SDK/DSPy/GEPA extras and runs interface tests with fake providers. No CI job uses a live key. The exact results belong to the PR's check runs; a green configuration file alone proves nothing. -Prior flat-study test-input and normalized-text exclusion lists were generated in the original local checkout under ignored `runs/hierarchy-issues-20260930/`; they are not published because hashes of low-entropy text can be sensitive. A new checkout needs independently verified exclusion lists. Live registration checks both lists against every split; a human must still review semantic and near-duplicate independence. `typewright-study manifest` now writes the complete pre-spend live manifest from a registered live protocol, `typewright-study review` binds a human review of the required attestations to that exact digest, and live selection refuses to start unless the reviewed manifest matches the protocol and current software parameters. The software can therefore produce and enforce a complete manifest, but issue #20 still needs the human inputs: a new independently human-reviewed four-split dataset, a completed review with verified price quote and externally enforced billing cap, paid and teacher-sharing approvals, and after live selection the frozen graph digests and semantic review. Paid calls and teacher-example sharing each require separate explicit authorization. Issue #21 and epic #6 remain open until their acceptance criteria are verified. The existing flat benchmark is not the new hierarchy holdout. +Prior flat-study test-input and normalized-text exclusion lists were generated in the original local checkout under ignored `runs/hierarchy-issues-20260930/`; they are not published because hashes of low-entropy text can be sensitive. A new checkout needs independently verified exclusion lists. Live registration checks both lists against every split; a human must still review semantic and near-duplicate independence. `typewright-study manifest` now writes the complete pre-spend live manifest from a registered live protocol, `typewright-study review` binds a human review of the required attestations to that exact digest, and live selection refuses to start unless the reviewed manifest matches the protocol and current software parameters. The software can therefore produce and enforce a complete manifest, but the live scope retained in issue #74 still needs the human inputs: a new independently human-reviewed four-split dataset, a completed review with verified price quote and externally enforced billing cap, paid and teacher-sharing approvals, and after live selection the frozen graph digests and semantic review. Paid calls and teacher-example sharing each require separate explicit authorization. The owner explicitly approved splitting software and live tracking on 2026-10-03. Issues #20/#21 are closed for their delivered software scope; their original reviewed-live-manifest and empirical all-child requirements remain assigned to #74 and epic #6. Both live trackers stay open until their acceptance criteria are verified. This decision supersedes the earlier conflicting close/reopen comments and does not authorize any live execution. The existing flat benchmark is not the new hierarchy holdout. + +The owner decision and retained requirements are recorded on [#20](https://github.com/Jev-Engineering/TypeWright/issues/20#issuecomment-5968647207), [#21](https://github.com/Jev-Engineering/TypeWright/issues/21#issuecomment-5968647383), [#74](https://github.com/Jev-Engineering/TypeWright/issues/74#issuecomment-5968647572) and [#6](https://github.com/Jev-Engineering/TypeWright/issues/6#issuecomment-5968647770). Fresh software evidence on main `e259dac3028ba110dabe584423d3d3a77de91e6a`: five hosted checks succeeded (core Python 3.11–3.13, optional DSPy 3.3.1/3.4.0). This observes software checks only.