Skip to content

CI and tests: warm caches, fail fast, keep only what catches real bugs #194

Open
0 of 1 issue completed

Description

@andersalm

Test audit, 1 October (develop c9205a74). Rule used: keep a test only if it can catch a bug that a user or another component would see. Delete tests of source text, wording, wiring, mock echoes, incidental defaults, tautologies and duplicates.

Facts

  • Rust: 4,697 tests; about half of the Rust lines are test code. Most are fine: about 2.5% in elastos-server and about 5% elsewhere fail the rule (serde round-trips, constructors without asserts, help-text pins, copies of components.json, include_str! of production .rs).
  • scripts/: 413 files, 192k lines. 265 files (107k lines) are named nowhere in the repository or are one-off proofs from past tasks. 145 of 201 JS and Python test files run nowhere; 35 of those fail today.
  • CI runs three text checkers (.github/workflows/ci.yml:52-54: check-wci-alignment.sh, home-entropy-check.mjs, browser-entropy-check.mjs, about 29k lines). They assert on source text and copy, for example scripts/browser-entropy-check.mjs:1237 requires a launcher to contain def write_all(fd, data):. They break on refactors and let real bugs pass.
  • Good tests run nowhere: 95 capsule unit tests (0.15 s) and the 53 signed-install tests in scripts/install-bootstrap-test.py.
  • CI takes about 47 minutes. The time is in the install journeys (35-45 min) and test-elastos (19 min: 11 min compiling, plus two tests of about 3 minutes each, carrier.rs:10593 and collaboration_presence.rs:1314). The protected-content crates (feature after 0.8.0) cost about 6 minutes per run.

Candidate lists (file, test, category, reason): https://gist.github.com/andersalm/b00b1a33a8fbc35778216d42763ca412. They are candidates, not decisions. Tests and scripts that pin Home as a web page (about 140 tests and 40 scripts) stay until #180 replaces them, and a script is deleted only after the check below.

Part 0 merged (#196, #198). Part 1 merged (#224, 4 Oct). Part 2: draft PR #225.

Done means

Done means, Part 0: warm caches and stop slow steps early

Done means, Part 1: run the good tests, replace the text checkers

Done means, Part 2: delete what protects nothing

  • Coverage gate: before a test or script is deleted, the PR lists the behavior it covered and the test that still covers it. Authority, wire contracts, provenance and lifecycle stay covered. Proof: the list in each deleting PR.
  • AGENTS.md verification commands name only existing tests and scripts, updated to their replacements. Proof: a reference check in the PR.
  • Before a script is deleted, its use on the seed, in releases and on the Jetson is checked (Anders confirms the seed). Proof: the PR lists each deleted script with "no caller found" or its replacement.
  • scripts/ has 150 files or fewer and 85k lines or fewer; every remaining script has a caller (ci.yml, justfile, install or release code, or AGENTS.md). No Markdown file names a deleted script. Proof: a reference check in the PR.
  • Rust: the listed delete candidates are removed (about 140 in elastos-server and other crates) and the duplicate groups become table tests. Proof: cargo test -- --list totals before and after.
  • No test reads production source to assert on it. Proof: git grep include_str! finds no production .rs, .js or .html used by a test.
  • The two 3-minute tests run in under 20 s each, and no real-time sleep of 1 s or more remains in Rust tests. Proof: test-elastos finishes in under 17 minutes.
  • Protected-content tests run only when their files change, plus a nightly run.
  • custody-harness-smoke saves and restores a develop cache like the other jobs. Today it never runs on develop pushes, so it builds cold (30 minutes) whenever a PR touches Carrier or custody files and then sets the PR wall time.
  • Total CI wall time is 38 minutes or less (from 47). Proof: three consecutive green runs on develop.
  • The Home-as-web-page tests and scripts are listed in Hacked-Home regression suite in CI #180 and are not deleted before Home: split into a Runtime-owned host and a sandboxed Home GUI capsule #176 and Runtime places scoped, short-lived app capabilities into each frame, with real stop #177 land.

CI/build coherence review

The source/log/API audit reviewed published train 8679214a0e2a6e090807e7ef956ff01a7e844cd1 (tree fbe1f9b4b488a3a4c52322428299d6c14456fd3c) and prepared cache candidate e9cea26a63f78f50b6ab2efc1dbbab3795de5243 (tree be2f7fc5dcc5ccd0d04c20c0ae505ea21e21d88b). The findings below describe that pre-implementation state. The current local implementation and checks are recorded in the next section. GitHub develop was verified at bc7832389519bd130ac0b078aea74332fa831316. OMP has since published its clean train f7224ac701743a70bc14978fdb2ac74abb6d0f87 (tree 2bf4f1e740e2b2fea38144af785f0872d7e5fd5d), zero ahead/behind its remote. Its four added commits change host-lock cleanup and protected-content fixtures; its CI/build/release sources match the reviewed train.

Verdict: the approved direction remains sound, but the current pipeline has incomplete integration and avoidable work. Each distinct build input set needs one construction; tests and release assembly should consume verified outputs. Different targets, test harnesses, profiles and real version stamps remain distinct input sets. Isolated test Homes and reusable build inputs serve separate purposes.

Verified cost and coherence findings

  1. The merge queue is inactive. Live GitHub reports null queues for develop/main, no repository/inherited rulesets and zero merge-group CI history. Develop protects nine checks with strict=false and zero required approvals; human review currently depends on the operator process. The merge-group YAML trigger only prepares the workflow. PR, merge-group and develop triggers all select the full matrix. GitHub's merge limits do not combine merge-group builds. Any reduction in develop runs must change the release evidence selector and retain a trusted cache/artifact producer in the same change.
  2. Publication still requires a full develop push. scripts/release-publish.sh:54-73 admits only a successful exact-SHA develop-push CI run and all protected jobs. It rejects PR/queue evidence. This makes a second integration run a current contract requirement, rather than an inherent CI requirement.
  3. Pre-merge packaging and normal merge disagree. release-package.yml:61-68 admits a new open PR head H, but a normal merge creates a different develop commit M. The package receipt remains bound to H, while publication requires a develop-push run at H. The PR run at H and develop run at M cannot pass that gate, even if the trees match. No equivalent-tree acceptance or receipt remap exists. Build from the final merged SHA until a reviewed qualified-build admission contract preserves original producer provenance and verifies complete input equivalence.
  4. Journey and shipping artifacts differ. Train/e9 PR/queue/develop journeys set LTO=false/codegen-units=16; release preparation retains manifest profiles and builds musl targets on Linux. Source-home uses native host Linux targets. The receipt records source, target, tool versions, lockfiles and file hashes, but omits effective profile/flags/SDK identity. Release acceptance needs the exact shipping artifacts. The signed-origin path already verifies, signs and imports prepared bytes without rebuilding. Monthly GitHub unstamped source releases are a separate documented role, so their different stamp is intentional; enforce artifact identity and keep their acceptance claims explicit.
  5. Release preparation recompiles dependencies. N and N+1 use different fresh target/build directories, delete them between versions and configure no compiler cache. N+1 already reuses N support bytes correctly. Share verified dependency construction while preserving version outputs; prove cache reuse across different directory paths before claiming this saving. Reuse the qualified ARM engine archive with its existing verification gate rather than building its unchanged recipe again.
  6. Some work has no consumer. Mac/Linux x64 package on develop/queue events, then skip uploads. Keep ARM packaging for its ELF compatibility check; limit other packaging to a real consumer. The Mac tail can also switch back to the original Runtime stamp after the two hop builds. Linux Carrier compiles a debug Runtime after a release Runtime exists; a verified prebuilt release input can preserve the same journey while avoiding that second profile. Its repeated release-provider invocations generally reuse fingerprints.
  7. A source suite is executed twice. release-platform-input-test.py:1642-1652 launches the installer-bootstrap suite, then ci.yml invokes the same suite directly. Keep one authoritative invocation and all cases. Source-gate also starts sccache despite being compile-free. Capsules already share build intermediates through .cargo/config.toml; their separate final target paths serve process tests. Clippy/test/list are distinct purposes, and Cargo invocation counts are not cold-build counts.
  8. Compiler caching is useful but incomplete. The actual PR275 Mac job had zero shared hits and 4401 read-only write errors. Prepared e9 proves writable-local/read-only-remote topology, not remote persistence or Runtime timing. Pinned sccache cannot cache linker crate types and its Rust cache requires link emission; metadata-only check/clippy work remains outside that main cache path. Current GitHub cache limit is 10GB with 9,283,342,129 bytes stored, mostly legacy target caches and no visible sccache entries. Migration/retention needs an explicit storage budget. Only accepted trusted producers should fill shared reusable outputs; read-only PR consumers remain appropriate.
  9. Product acceptance remains broader than green CI. Mac covers installed account/System/update/refusal/Undo with virtual passkeys. Linux covers source-home/provider/Carrier installation, but lacks that full signed GUI update/Undo journey. Hosted ARM ELF/glibc checks establish compatibility rather than a Jetson hardware run. CI model tests use a fake engine, Chat layout uses fixture APIs, and product Browser launch/app matrix are disabled by default. Real installed AI reply, community Chat and suitable-host Browser WebRTC media/input/reconnect/close evidence remain required on the offered targets.

Measured PR275 run: Mac 44m15s = source-home 26m21s + two Runtime hop builds 9m27s + installed GUI 4m59s + other 3m28s. Source gate is 5m29s, including 4m release-source fixture work and 57s pre-push-gate fixtures. These are measurements of published ddd8ea5, before local e9, not a controlled warm/cold comparison.

Measured release run: Mac N/N+1 construction 55m50s of a 60m07s job. Its workflow head is 57eb779, while logged build inputs bind source 67c268c51743b206c4235b6fc413d42e3eb2fc97, tree b2d3636ce89c22c4da0621dfbf96650a4d928cd9, install alpha.7/update alpha.8. This historical run does not prove the new train or prepared cache candidate.

Coherent completion under the approved plan

Keep the current candidate fixed and keep installed/authority checks. Stage 1 must align integration policy, required checks and the release evidence selector, complete source-home local/shared reuse and prove a real warm reader. Apply the narrow duplicate-invocation, discarded-package and platform-dependency corrections in a necessary batch; the x64 proof can run independently of the ARM engine consumed only by ARM. Until the gate changes, release inputs come from the final merged SHA.

Stages 2 and 3 must connect actual qualified producers and consumers: key complete build inputs (source closure, locks, compiler/target/sysroot/SDK, features, effective profile/flags and stamp), verify digest and producer attestation, reuse or rebuild on a miss, run installed acceptance on those exact prepared bytes, then sign and publish the same hashes. Preserve the existing support-origin receipts. The separate build-key prototype has no demonstrated product consumer yet; additional producer proofs alone do not establish reuse. Keep the Runtime-only version layer distinct from unchanged support.

Stage 4 remains after the approved week of clean cold comparisons: dependency-aware test selection with full-suite fallback. Completion needs measured cold/warm job and runner times, cache hit/miss/write/eviction data, exact shipping-artifact receipts, and accepted Mac/Jetson user journeys. The existing 38-minute/three-green-run requirement stays open until measured.

Parallel CI/release coherence implementation

Task base and PR target remain develop, under the approved build plan. Codex owns the source-home cache, redundant build work, release compiler reuse and queue/evidence integration. OMP owns alpha.8, provider version separation and the qualified artifact-key/store work. OMP read the coordination comment and confirmed that this batch stays local until alpha.8 lands. Its release preparation and CI will use the same final merged source commit. Each worker retains its own checkout and installed Homes.

  • Implement a reviewable local batch with fresh outputs, installed journeys, production release profiles and exact-source authority preserved.
  • Prove local dependency reuse, source/stamp/path invalidation, release-loop cleanup and publication refusals.
  • Integrate the current final alpha.8/develop base with a normal merge, pass the exact-candidate pre-push gate and review, then publish the batch for review. Proof: ci: reduce repeated builds and align release evidence #275 (merged 8 Oct)
  • Prove hosted trusted cache writing and warm reading, queue behavior and exact shipping-artifact acceptance before claiming the repeated-build problem solved. The cache still stores almost nothing; open in CI speed: make the compiler cache store entries and shorten the macOS path #310.

Candidate and implemented behavior

PR #275 merged on 8 Oct (941ef3b).

  • Compile-free source-gate runs the installer suite once through the existing platform-input entry point. It retains every installer case and uses ordinary Rust tooling without starting sccache.
  • Mac and Linux x64 package only when the event has an upload consumer. ARM keeps its package and ELF/glibc check on every event. x64 source-home starts after source-gate; ARM alone waits for the engine producer. Shared Linux steps use one YAML anchor, with all nine protected check names preserved.
  • Linux Carrier checks the current locked release binary. Cargo still verifies source freshness; this avoids a second debug-profile Runtime build in a release journey.
  • Release preparation accepts full source/tree SHAs on develop. An open PR head alone is refused. Preparation can run beside exact-source CI; publication still requires green evidence and accepted release inputs.
  • Mac/Linux release preparation uses the pinned compiler-cache action when present in the source. Earlier accepted sources retain their normal cold build. Each runner has a bounded 2G writable private tier and reads shared GHA. Production profile settings stay with the release build. N/N1 use fixed Cargo output paths, require both paths absent and free of symlinks before each build, and remove them afterward. Staging receipts stay separate. Cache statistics bind source, tree, platform and both versions.
  • ARM release preparation can restore the existing exact recipe-keyed engine archive. Both restored and cold archives pass the approved SHA verification. Original producer sidecars are retained. A cache miss uses the existing build; release dispatches save no shared engine cache.
  • Publication accepts completed successful ci.yml evidence at the exact package SHA from either a develop push or a develop merge group. Every protected job must succeed; unreadable or empty required-check configuration is refused. The develop-push fallback remains. Source/tree equivalence and receipt remapping are outside this change.

Checks and proof

ci-release-policy-test.py: 45/45 passed before the final fixed-path correction. Both affected methods then passed in 3.739s on the final source, including ten real release-loop shell cases with inert builders: stable N/N1 paths, output cleanup, preserved cache data, and refusal of existing directories or dangling symlinks. Final policy SHA-256: af591b1c16880f6eefd6ef82fae20718862af9607f1435734dd574848df37b4e.

release-source-upstream-test.py: 18/18; release-publish.test.mjs: 47/47; prepare-release-platform-test.py: 19/19. The full platform-input run passed 110 cases and hit the installer wrapper's 180s timeout while several fixture suites ran together. The single installer admission wrapper then passed serially in 109.281s; its timeout and assertions stay unchanged. This records the initial failure and the bounded rerun. These fixtures use inert payloads, isolated keys and local transport; they do not establish installed product acceptance.

Product-data lint and its 84 tests, Runtime/chain-provider formatting, shell/Python/Node syntax, diff checks and pinned Actionlint 1.7.12 pass. Independent review of publication admission, the probe and integration found no blocking issue. The published candidate's earlier pre-push result belongs to ddd8ea51; the final local candidate still needs its own pre-push gate when OMP releases the Mac build slot.

The checksummed sccache 0.18.0 / Rust 1.91.0 executable probe passes eight stages with both Cargo output trees cleared before each stage. Cold A: 0 library hits/2 misses; warm A: 2/0; independent B: 2/0; source change: 1/1; release-stamp change: 1/1; repeat changed: 2/0; changed output paths with unchanged source/stamp: 0/2; repeat at those paths: 2/0. Library hashes and linked output pass. Private disk totals: 10 hits, 6 misses, 6 writes. Read-only loopback GHA: 0 hits, 6 misses, 0 writes; 13 lookups and zero uploads. Compiler/cache errors are zero. Owned servers stop and temporary fixtures/tools are removed. The small result is privately retained and bound to the exact source digest, commit and tree. This proves the mechanism in a small fixture, rather than whole Runtime release timing or remote persistence. The path miss agrees with the pinned Rust hasher, which includes Cargo environment values.

Queue and remaining acceptance

Native develop queue ruleset 24624626 is prepared with enforcement disabled: normal MERGE, ALLGREEN, one build/merge entry and a 180-minute check timeout. Existing branch protection and all nine checks are unchanged. Integrate the trigger and evidence guard first, then use one reviewed canary to check that the tested queue SHA is the accepted develop source. Keep the full CI matrix and develop-push fallback through that check. The queue is not active evidence yet.

After alpha.8 lands, Codex integrates the final base, runs the exact-candidate gate, publishes this batch and completes independent/current-base review. Request required human reviewer irzhywau when the PR leaves draft. Measure hosted Mac/Linux cold/warm job and runner times, cache hits/misses/writes/eviction and real production N/N1 builds; resolve the shared cache storage budget. OMP's artifact store must connect qualified producers and consumers with complete inputs and original provenance. Dependency-aware test selection retains its approved cold-comparison gate. Exact shipping artifacts, real local AI reply, community Chat and Browser media/input/reconnect/close remain subject to the Mac/Jetson journey gates. The 38-minute/three-green-run requirement stays open until measured. No Runtime build, additional CI run, signing, live mutation or deployment was started for this local batch.

Current efficiency check (7 October)

OMP's alpha.8 train is integrated as 4933398378173ee84a8676fa24c7cfcb49a3a8c9 (PR276); its develop CI and all three release platform jobs passed. The release-note/process/smoke-assertion correction then merged as 91ba79c467cf9cae764d8ff3b76d7c08043e6081 (PR278). It changes no Runtime/provider production code. The exact-source release policy still requires new candidate CI and package provenance, so that correction caused another full CI and package run.

Completed run durations from creation to last update: develop CI on 49333983, 71m40s; PR278 CI, 68m31s; release package on 91ba79c4, 55m58s. These are end-to-end elapsed durations including scheduling and finalization, rather than CPU times or controlled cache benchmarks. At the 14:23 UTC snapshot, develop CI 37626076405 was still in progress. No later result is implied by this record.

A concrete unused build remains in the green develop Mac job: packaging ran 06:08:54–06:16:23 UTC (7m29s), then the upload step was skipped. The job itself lasted 65m52s. The prepared package-consumer guard directly removes this work on that event; a measured new run is still needed to establish total savings.

The artifact-key/store branch has review progress, but remains outside develop with production reuse off. Its real different-CPU construction proof and pipeline consumer remain open. Synthetic receipt tests supply guard coverage rather than that host proof. The next efficiency work is current-base integration and measured rollout of the prepared batch, then qualified artifact consumption and metadata-only assembly, followed by dependency-aware test selection under the approved full-suite comparison gate. Preserve exact source and producer provenance throughout.

Earlier correction PR #271 is merged. Its CI run, head 9eced7928d0124a788abcac27aa6dff40ad715af, passed every applicable job: source-gate start to final job completion 54m32s, Mac source-home 49m02s, Linux x64 34m31s, Linux ARM64 35m00s. These are baseline timings for that candidate, not acceptance of the current product or warm GHA cache performance.

Browser plan reconciliation — 2026-10-08

Browser tasks consume qualified exact-input artifacts and original producer provenance under this issue's reuse rules. Use source/behavior checks first, freeze the affected closure, then one necessary affected-output build under the shared Mac lease. Reuse matching image/helpers and read completed CI receipts. This efficiency rollout is its own work; complete #194 closure is not a prerequisite for #217 interface work or the next #131 journey. Exact-candidate hardware/media proof remains #293.

Linux target-QA package proposal — Anders, 9 October 2026

OMP owns the decision and CI implementation. Commit 0c7f023e3ced410acfccc8b932a91d04ab703674 is a proposal under this issue. Root will separate this concern from the sole Chat PR #309.

The proposed two workflow conditions admit pull_request and merge_group events for the existing Linux release-package and upload steps. The existing policy test changes with those two conditions. This makes Linux x86/ARM target-QA packages available from the same PR CI matrix, while retaining original producer commit/tree, package hashes and effective build-profile evidence. The prepared policy suite passed 69 cases with independent review.

This proposal changes the earlier PR-packaging skip policy recorded above. Its checkboxes remain the accepted baseline while OMP decides the change. Linux target-QA receipts identify the CI test profile (release, LTO disabled, 16 codegen units); shipping-artifact acceptance retains its own exact-profile gate. Root prepares source separation only. OMP approval and the resulting reviewed CI change remain pending; this update starts neither a workflow nor a CI implementation PR.

Complete test register — acceptance record

The complete source-review register is published for review in draft PR #487 at baea082fc6c8c145fcf6a04afe7c383b24fc0da3, tree db483d42de5f684d8df85973c8a8778c5188a865. The normal fast-forward to origin/feat/test-audit-register completed from 6834807e0c11685a44bd44bc95e1ac5c5a6daf70. The task worktree is clean and local/remote divergence is zero ahead, zero behind. The PR stays draft and this issue stays open in QA / Review.

Completed source review

The register and reader cover 1,926 tracked files and 9,361 attributed body reviews, with zero provisional body records, grouped into 1,753 findings. The reviewed source is 460b1888d344f52cebd1c25866be88ff4085f035, tree 17aeaf23c640e2dec8f653fd16f3771f95a55b9c. That source was 141 commits ahead and zero behind baseline 3beb816438fd567145a8d725e22465bd83277cf0; this publication contains only the register and its reader.

Proposed dispositions: 5,800 retain; 3,459 improve; 41 replace; 52 consolidate; 9 removal candidates. Each record retains its ID, source/body hashes, reason, action and review scope. These are test-quality proposals. Removal candidates require their recorded coverage or ownership decision before deletion.

The census contains 3,351 server Rust, 2,412 Other Rust, 2,622 JavaScript, 849 Python, 116 shell, 3 other and 8 embedded records. All 2,412 Other Rust records and the 199 server records that previously had baseline-only scope have completed named local-contract reconciliation. Other records retain their stated scope. The 71 later Browser records remain separately pinned to 05da461c20db3b224ec87f22d7ed41b79ff6acfb and excluded from the main count. Runtime-generated cases, compiler/feature/platform discovery and installed journeys have separate evidence requirements. Source body reviews give no additional product-test execution credit.

Verification and publication proof

Independent exporter, reader and exact-output reviews passed. The reader verified the census against pinned Git objects, including the separate follow-up records. Product-data checks passed for 42 capsules, 2 profiles and 28 contracts, including 99 tests; both Rust formatting checks and git diff --check passed. The mandatory publication hook passed on the exact published commit, including 1,739 JavaScript tests, zero failures/cancellations/skips, and the complete source/installer fixture suites.

Final artifact manifest SHA-256: 2558227ac3b87f320cdd6d945c9aafca065ee382d029cd89e30e21c67342efdc. Independent output-review receipt: 5a728b615603c99c6131cd5f6ed10d9acd721911eaeed356e179c83610031457. Publication receipt: 5b4f50a7612f047c804aa7728c4b96ec8651303bbbae6e2be8259b400f0ca735; it binds all 18 package files, source and artifact identities, reviews, checks and remote verification. Publication-hook log: a1f83bc16a2d7e39cd2f5e8c3014ebd322221d7d46063a929184457724fd5a17.

Hosted CI run 39887126741 completed with cargo-audit failed on tracked dependency advisories, release-upstream-paths passed and the remaining jobs cancelled. #311 owns dependency-audit remediation. The register publication leaves dependency lockfiles unchanged; the passing local gate does not replace the cancelled hosted checks.

Remaining acceptance and ownership

#194 owns review and adoption of the individual recommendations; #319 owns runner accounting and pre-push/CI coherence. The four-part release audit in #93 owns current candidate decisions, installed journeys and the final release gates. Its open product gaps include Library concurrency #480 and update recovery #484. Later development changes require their own source and installed evidence.

The current public export passed privacy review. Private records preserve detailed evidence. Earlier published draft history remains preserved; its assessment-metadata follow-up has private containment receipts and a separate history decision. This publication grants no release acceptance or risk waiver.

Historical checkpoints and later review evidence

The entries below preserve earlier checkpoint and preparation states. The completed publication and current acceptance state are recorded above.

Publication gate progress on artifact commit baea082fc: its JavaScript section passed 1,739 tests, with zero failures, cancellations or skips (555.052 seconds). The hook is now running the source/installer fixture section; the push has not completed. This run belongs to the artifact branch and remains separate from the 9,361-record audit source census.

Release use: the register recommendations concern test quality within their named source scope. Confirmed product gaps and current candidate decisions remain in the release audit, including Library concurrency #480 and update recovery #484. Source-review evidence for other repairs carries only its recorded scope; complete installed journeys and the nine final release gates remain with #93. Runner-accounting and pre-push/CI coherence remain with #319. This register creates no release acceptance or risk waiver.

Final source-review artifact 056 passed independent output review (5a728b615603c99c6131cd5f6ed10d9acd721911eaeed356e179c83610031457). Prepared normal commit baea082fc6c8c145fcf6a04afe7c383b24fc0da3, tree db483d42de5f684d8df85973c8a8778c5188a865, contains the approved export and reviewed reader/docs. All 9,361 source records preserve their decisions; the sole summary-precedence correction changed one finding and five linked row contexts. Exact pinned-source verification and the worktree reader passed. The task branch is clean, one commit ahead and zero behind origin/feat/test-audit-register; publication will fast-forward draft PR #487 after its mandatory hook gate. The historical draft commits remain preserved.

Projection 055 final review (920478cba048663c57829b1911fbe1c1355adbb1c63a7125e37182cd43ebf624) classified 142 cross-group cases and required one conservative summary correction. Frozen successor 056 applies that single correction; manifest 2558227ac3b87f320cdd6d945c9aafca065ee382d029cd89e30e21c67342efdc retains the same source census and review decisions. Its code delta passed independent inspection; exact output-delta verification is the remaining publication check. Earlier published wording is retained in draft history; no new product-security mechanism was found by this overlap review.

Projection 055 is held from publication. The final bounded comparison found an earlier public finding that repeats a privately classified assessment. The original frozen evidence remains intact; a minimal successor will apply the approved public-summary precedence after the remaining bounded comparison returns. Source identities, dispositions and body-review counts remain unchanged. The draft PR still serves checkpoint 054 at 6834807e.

The final source-review proposals are 5,800 retain, 3,459 improve, 41 replace, 52 consolidate and 9 removal candidates. Each record preserves its reason, action and review scope. Removal candidates concern test-only or unconnected helpers and require the recorded production-coverage or ownership decision before deletion. Scoped output review passed (88456d42e46d70d0d9d5800eef0d40cee1bc518d9d2db614ddbe01e9ede6417c). The final reader verified 1,926 files, 9,361 records and 1,753 finding groups against pinned Git objects, plus 71 separately pinned follow-up records. Broad output review is completing cross-group privacy classification before publication.

Frozen final projection 055 contains 9,361 attributed body reviews, zero provisional body rows and 1,753 finding groups. Manifest cb23e3d3df5a99746255b391f2f1d8e9311b49246756957be57ced74e3e65452 binds input freeze b411bd44742a20e761f756950be658d67f38fe506d5c54e9671d9eda4597163d. Corrected exporter review passed (a7eed266aa0c2caf046572a476acb325753951d2a4a2e09a9e9e4c1121662c25); reader review passed nine isolated integrity checks (c78d2fc117434e1ea4b91f6c7413965251addd9a390e4a4136b398917080a0f1). Product-data checks passed for 42 capsules, 2 profiles and 28 method contracts, including 99 tests; both Rust formatting checks passed. The final data receives independent output review before publication.

Privacy follow-up: the final projection removes extra internal assessment labels found in earlier draft metadata. Detailed classification remains private. Updating the current register and changing already-published history are separate actions; the latter retains its explicit-approval gate.

Final named-contract reconciliation is complete for all 2,412 Other Rust records and all 199 server records that previously had baseline-only scope. Other Rust receipt 05facd0e2702c5545d1b5300c3f016643fb2f1bf19c24a129ece3f6b33977368 binds 287 source files; server receipt 09afdedd5467756153afb7707acf0e51a56a451044bf8f964fd71064a6311625 binds 466 helper spans across 100 pinned files. These are source-review results, with zero additional product-test executions. Independent exporter review found lost scope fields and incomplete private-field handling; the final export is being corrected and will receive a separate exact-output review before publication. Draft PR #487 remains at 6834807e during that check.

Independent review confirmed all 75 supervisor identities, body hashes and proposed dispositions. Source-review receipt: 93cea8aa09e173685d71cb89a1332399c3ae169b012a36842e121567b8b657c5. It required narrower public wording for two finding groups; immutable correction 7ee73f94ebf269d9146b95b385e2aadded3920fab1adff68bccd165410580878 applies those exact recommendations while preserving the original private evidence. Final public-output review remains pending. A preparation reader check verified 1,926 source files, 9,361 records and the separate 71 follow-up records against their pinned Git objects. That check validates artifact identity and links; it adds zero product-test execution credit. The public PR remains at 6834807e until the final review and publication gate pass.

All 9,361 registered source records now have attributed body reviews, with zero provisional bodies in the reconciled working evidence. Receipt eb65eda42b0beff0dbdc189de284b670bc1e4b2fafd2d7401537ddc2acada082 combines unchanged published partition counts with Other Rust union 88d25910dad6c887dd278cbf02ba16ade534d2d94e59ced98986d7f22218f229. This is a source-reading milestone. It grants zero execution credit and leaves named helper/contract assessment, dynamic or target-specific registration and installed journeys distinct. The three reviewer lanes continue those checks and independent review. The next public projection will include the later evidence after reconciliation; draft PR #487 still points to checkpoint 6834807e.

Root079–084 now cover all 75 reserved VZ supervisor declarations with complete body and named local contract reviews: 28 retain, 44 improve, two replace and one consolidate. The exact-ID reconciliation is d9320a5fdc41ae92847c2654fc97d97f86dab82132e30e549caa5e7683615fa9; it verifies 57 newly attributed body reviews and the unchanged source/body hashes. The final two batches add 28 assessments, with completion digests 36b59f42bc9105a1f5f1b9f2b4a0f2186db78e0be834b009893a81a51a1a5472 and e2cb24b5136dbf52b38e5ed47ab5a519d057d2d17c75def169230b5824172e30. Two durability tests should call the production formatter instead of asserting their own literals. Other actions strengthen exact artifact preservation, independent refusal controls and subprocess cleanup. Private boundary evidence stays in controlled records. All 75 assessments are source-only; independent review and the next full-register reconciliation are pending. PR #487 remains at published checkpoint 6834807e; broader named-contract, runner and installed-journey acceptance remains open. Earlier progress paragraphs below are historical checkpoints.

Root081–082 add 27 supervisor transport/display/profile reviews, bringing this reservation to 47/75 complete body-plus-contract reviews. The two batches contain eight retain and 19 improve decisions, with 21 newly read bodies. Actions include a decisive port-release waiter test, isolated signal-handler fixtures, exact display arguments and controlled profile-disk inputs. Private producer/consumer and ownership evidence has explicit projection flags. Completion receipt SHA-256 values: 4ebed77c7b6e4e5255d8731edda23c18ba01f0326f2ead963a78cc98c0d2da6d and 1a7bf94dfb70645ef9ca1b6c42d90e23e20dbef2cd1ecaaa0cb2ffb6195bb775. No Rust/native/network/VM test execution occurred. The remaining 28 supervisor records stay assigned here. PR #487 remains the previously verified public checkpoint; these later batches await a new reconciled projection.

Root080 adds 14 supervisor cleanup/startup reviews: six retain, eight improve, all 14 newly read bodies. This brings the reserved supervisor set to 20/75 complete body-plus-contract reviews. Useful checks already cover pending task ownership, observed completion, disk-lock contention and preserved profile bytes. Recorded actions strengthen fixture deadlines, exact cancellation causes and unsuccessful worker completion. Private resource-ownership details remain in controlled evidence. Receipt SHA-256 bb8e941e2df45ad4337e0b5f60878add630ed7ef4083a551fdeaf0a62dd70050. These are source reviews with zero native/VM/process execution. The published PR and dependency-gate status below are unchanged.

Latest source-review checkpoint: root079 adds six VZ supervisor declarations with complete bodies and named local contracts: one retain and five improve; four were previously provisional body records. It requires complete settlement fields, exact launch-to-path mapping and behavioral admission checks. Private launch-input findings remain in controlled evidence with safe public summaries. Completion receipt SHA-256 3ad4b293bcc165172e81bb11aa5025a450d7332f6b1fba93b35e00f971718520; no native, VM or installed execution occurred. The remaining 69 supervisor declarations stay assigned here. The last verified Other Rust union through RO193/CI012/root078 is 2,005 of 2,412 complete body-plus-contract reviews; later batches remain separate until the next union check. Draft PR #487 remains at 6834807e0c11685a44bd44bc95e1ac5c5a6daf70 with the dependency-audit gate open under #311.

Hosted CI on current head 6834807e completed: run 39876065384 failed cargo-audit, passed release-upstream-paths and cancelled the remaining jobs. The failure log SHA-256 is 9a9f250deb64d6b147c3ba4f6c4b2f4df35b247d753093cc6391a677d2b6ded6. Dependency remediation remains owned by #311; this artifact changes no lockfile. The successful local artifact checks and failed hosted dependency gate are separate results.

Root batches 076–078 complete all 60 reserved VZ FFI declarations at body plus named-local-contract depth: 31 retain, 26 improve, two replace and one consolidate; 38 are newly attributed body reviews. Exact source/body hashes are verified. Completion digests are c0443b13202384526f16e1cfbcabd42477d36cdbe9e0e673c500d2b85c1736af, b894a60f9353f1268480d9cc0ed23f8b043c3a26c1b71d713fb6df3e2323929e and 82c902cf6b53249819606ed2b9c22a2d3f263eebc0f7101520933f7cc7940ce0. Useful native-construction, configuration, byte-bound and lifecycle primitive tests are retained. Changes are needed where assertions fail to prove attachment identity, worker completion, queue independence or the stated error path. These are source assessments; target execution and installed acceptance remain separate. The 75 supervisor declarations remain outside this completed family.

Independent spot-review verified all 44 identities and body hashes in root batch 075. Its sensitivity metadata correction is frozen separately and approved by review receipt 1ec0c8586d5fd54d28b89ea3dfc2d2492fd3f3e18a5c18c41a7af81e127e0c29; future public projection must apply the correction and pass exact-output review. This changes no review count or test decision. Batches 075–078 remain outside published checkpoint 054.

Checkpoint 054 is published for review in draft PR #487 at 6834807e0c11685a44bd44bc95e1ac5c5a6daf70, tree 44ae98aa6f13a15d344b884a8f1146dca3af6267. The fast-forward push updated origin/feat/test-audit-register; post-push verification found a clean worktree, matching PR/remote heads and zero local/remote divergence. Publication receipt SHA-256: ba26f0b93fa2f976869ee5a9bcd3e3756d70b73d52da9843186d32765b8c5dbe. The exact-candidate pre-push gate passed all 1,739 JavaScript behavior tests, the complete source gate and both formatting gates; its log digest is 29672191bebbd3e1ce290a8e7da58c06f2ba0deb6da0fa6915411d3eb088da22. Hosted checks for this head remain separate. The issue and PR stay open while the exhaustive review continues.

The current register contains 9,361 records with 9,128 body reviews and 233 provisional bodies. Review depth has its own census; a body review alone does not complete the first-principles audit. Subsequent batches below remain outside this publication unless their frozen cutoff is named. Older checkpoint descriptions are retained as historical evidence.

Root batch 075 completes 44 VZ provider, VM, network and process-test assessments: 24 retain, 14 improve, five consolidate and one replace. Completion digest: 962b52b26354a432756341a13da5cd7824b4e626bc00a8faee10ae9f08581227. It adds 44 scoped local-contract reviews and 21 body reviews. The reservation summary had counted 20 pending statuses; one additional row had a CI-selection finding but still used a provisional body-review method. Body attribution now controls this count. Existing VM prerequisite/readiness findings are extended, while useful public refusal and repeat-initialization contracts remain. This source-only batch is outside checkpoint 054; independent spot-check is in progress.

Root batch 074 completes 31 VZ configuration/error assessments: 20 retain, 10 improve and one replace; 22 are new body reviews. Completion digest: 1f9cd0f5699a9183acfdfb24d8141e79ccec4bb77d5631a88ba9edfc5631d909. Every body hash and reference range is checked against the pinned source. The review identifies ineffective default-console assertions and stale consumer claims while preserving useful data-contract tests. It supplies source evidence; macOS VM and installed acceptance remain separate. This batch is outside checkpoint 054.

Checkpoint 054 has independent publication approval. Its manifest is fcde7448aafeb34caeb24f5516fa6a49c0c41a77824e7f287317a0923d1ea011; the independent review receipt is 66a4bd6e51ca4ad5eae983ccb47d0a85dbee66e3bebb4afbd091936de2432749. It has 9,361 records: 9,128 attributed body reviews and 233 provisional bodies, with 1,517 finding groups. The exact-source reader verifies all 1,926 files and the separate 71 follow-up records. The deeper census remains separate: 199 server records still need named-contract review; other Rust has 1,383 of 2,412 declarations attested at that depth in this frozen checkpoint. Later source reviews remain outside those counts.

Root batch 073 completes 36 component integration test assessments: 21 retain, 13 improve, one replace and one consolidate; 33 are new body reviews. Completion digest: dadf62cb1ad29a05bd243bf0cc6f56f6895930520bb720e5df106758a00b52d1. The tests call trusted Rust components directly, so actual HTTP and installed journeys have separate acceptance. The helper-coordinate correction digest is 932d4e64b13f97f7814dbef1e37e9d981690bdf07dd665f23b0e8bf28d6059a8; all 36 original body hashes and pinned source ranges are verified. This batch is outside the publication checkpoint.

Hosted CI for published d2a5312 failed cargo-audit; release-upstream-paths passed and the other jobs were cancelled. The log digest is c0634ca70136f389ebae5843aad40c8d15a82e35a1fa52df9b1b621b4a19b2c3; dependency remediation remains owned by #311. The artifact has no dependency edits. Local disk free space is below the repository 10% threshold, so source review continues without new builds or worktrees. The documentation-only pre-push path selects behavior/source checks and formatting without a Rust build.

The user requested a first-principles review of every test and approved this versioned evidence register. This issue owns acceptance and work status. The register holds per-record source identity, review scope, evidence and required action. The exhaustive review remains in progress.

  • Previous checkpoint 052 candidate: draft PR #487, head d2a5312a8ab327751a9b4790b74749f52d9486f5, tree 042c90af9f40fa5dca67dae3aef1cd04547ec50b. It is published on origin/feat/test-audit-register, with zero local/remote divergence verified after the push. The branch contains only the register and its reader. The audited source is separately pinned at 460b1888d344f52cebd1c25866be88ff4085f035, tree 17aeaf23c640e2dec8f653fd16f3771f95a55b9c; at the census it was 141 ahead and zero behind baseline develop 3beb816438fd567145a8d725e22465bd83277cf0.
  • Exercised checks: the reader verified artifact hashes, references, counts and all 1,926 source-file hashes against the pinned Git tree. git diff --check, product-data validation, its 99 tests, Runtime formatting and chain-provider formatting passed. The mandatory pre-push gate passed on d2a5312a, including 1,739 JavaScript behavior tests, the complete source gate and formatting. The older home-entropy-check.mjs command is absent on this artifact base; its current gate is check-product-data.mjs. These artifact-branch checks remain separate from audited-source execution evidence in the register. Rust tests and installed journeys were outside this documentation checkpoint. Hosted CI on the previous head ceca1d3b, run 39854146042 failed its cargo-audit job and cancelled the remaining jobs. The log reports dependency advisories RUSTSEC-2026-0379 through RUSTSEC-2026-0383. The artifact branch changes no Cargo lockfile; fetched develop at the recorded baseline has the same lockfile inputs. Dependency remediation belongs to Install, update, release and CI remediation (9 October review) #311 and remains an external merge/release gate for this draft.
  • Review: independent public-projection review passed for manifest SHA-256 7c2e500ccc3a51ec155391018bb0dc807a5b98e5cad8eddca2520bf1341069bd and exact tree 042c90af9f40fa5dca67dae3aef1cd04547ec50b. The review checked all 2,905 input hashes and public summaries across overlapping finding groups. Its private review receipt SHA-256 is 2d0e9d565f6b4c9192a7a4d3334243e6f5322cce73f30fe4acc6e772aac5e5ff. Private assessment text in the earlier projection was replaced with safe acceptance summaries. The withdrawn unpublished draft is excluded from the published ancestry; the user approved its local replacement.
  • Installed release evidence: source and fixture checks supply no new installed release, hardware, media or cross-device acceptance. Those journeys retain their exact-candidate gates.
  • Remaining gate: finish every provisional body and its relevant helpers; reconcile static extraction with runner discovery and target/feature selection; classify production guards separately from tests; reconcile later source changes; independently review and publish the remaining evidence. Each substantive implementation or deletion needs its owning issue and reviewed change.
  • Next owner: this chat is the sole test-research owner. It continues from stable IDs while preserving immutable private batches for the next reviewed checkpoint. PR docs: preserve the attributed test audit register #487 stays draft and CI and tests: warm caches, fail fast, keep only what catches real bugs #194 stays open.

The previous checkpoint 052 register inventories all 1,926 tracked files and 9,360 typed records. It contains 8,237 attributed body reads, 1,123 provisional body records and 1,186 finding groups. These units include Rust declarations, Node registrations, helpers, standalone scenarios and embedded checks; they are separate from executed-test counts. A body read records the complete declaration and states which helpers and production paths were inspected.

A separate follow-up at source 05da461c20db3b224ec87f22d7ed41b79ff6acfb, tree cfef92b9db42be4d5fdabdcbf550579f37e9e519, records 71 expanded registrations from 54 declarations: 26 retain and 45 improve. These records stay outside the main pinned census. Later source movement requires another explicit reconciliation.

All changed test bodies in the original candidate received direct source review. The published checkpoint includes complete Python and shell body coverage; individual scope limits still apply. The recovered GBA CDP input-fixture record accounts for the increase from 9,359 to 9,360 records. A source-text or serialization test requires a behavior-specific decision before removal.

Audited-source execution retains its original outcomes: the repeated central Node run reported 1,810 cases, with 1,808 passes and two readiness failures. Nine expected release cases were absent from that event accounting and passed in a separate targeted run. Five unselected Node smoke files reported 98 cases, with 96 passes and two failures; a top-level guard prevented another 16 from registering. Targeted Python evidence distinguishes fixture/oracle failures from missing pinned inputs. Aggregate runs supply no missing per-case receipt.

The current action themes are CI selection and result accounting, single-condition rejection fixtures, decisive assertions, bounded process/server cleanup and recovery proof. Confirmed isolated validator probes are retained with exact limits: hosted preflight audio reporting, Mac handoff exit status and the VM preflight child-result adapter need stronger verdict checks. Installed Browser acceptance remains separate. Useful behavior, wire-format, serialization and source-policy coverage stays until replacement evidence exists.

Later completed reviews remain private for the next projection, including all 100 setup, 24 Browser-image, 61 updater, 33 Home command, 73 controller and 18 protected-content operator declarations. The controller assessment has 58 retain and 15 improve decisions; the operator assessment has 7 retain and 11 improve decisions. Existing tests are linked where they already prove the requested behavior. Root has also assessed all 118 protected-content Runtime declarations, including storage, media prerequisites, provider startup, typed adapters, listings, availability, release assembly, mint recovery, wallet approval and viewer lifecycle; 117 are new body reviews and one refines an earlier compatibility assessment. The process tests retain useful cross-provider evidence; their recovery scans need to inspect the journal that the tests actually write. The Runtime assessment has 43 retain, 68 improve and 7 consolidate decisions. It adds exact session settlement, independent authority-field checks, complete recovered facts, effective history-limit fixtures and deterministic concurrency coverage to the next test changes. Stable-ID and body-hash checks account for all 118 declarations. These source reviews cover test quality and scoped implementation contracts. Installed behavior remains a separate acceptance gate. The JavaScript reconciliation now accounts for all 2,622 records across 493 classified files: 1,930 expanded Node registrations and 692 helper, standalone, scenario or criterion records. Each known record has an attributed body and scoped-contract decision, including the 12 previously reviewed Library rows. Counts distinguish 1,898 unique bodies from expanded records. There are 129 records with incomplete static inner-parameter values; runner selection and prior execution omissions remain open. This newer census includes one additional standalone record and is outside the frozen publication checkpoint. Server and Rust-provider workers continue disjoint records in this chat. Stable IDs, file/body hashes and immutable batch manifests preserve completed work; shared mutable audit outputs are excluded. The next checkpoint is published as d2a5312a8ab327751a9b4790b74749f52d9486f5, tree 042c90af9f40fa5dca67dae3aef1cd04547ec50b, with a clean worktree and zero local/remote divergence. The exact-candidate pre-push gate passed all 1,739 JavaScript behavior tests, its complete source gate and both formatting gates. The earlier fixture residue was preserved privately; the successful retry passed the final clean-tree gate. This update adds attributed reviews and an explicit review-depth census. The deeper Rust-provider census separately records 1,048 of 2,412 declarations with complete body and named local-contract review, compared with 1,993 attributed body reads in that partition. These counts have different meanings. All 78 Content declarations now have individual body and scoped helper reviews: 33 retain and 45 improve. The review covers publication, import, materialization, fetch, repair, recovery, status, alerts, quota, federation and market admission. Exact stable IDs and body hashes account for every declaration. Remaining improvements include request-specific signed decision controls, complete forwarded facts and bounded fixture cleanup. Independent source review confirmed the signed-decision test gap and narrowed its reachability: the pinned ordinary server and CLI constructors leave those external integrations disabled. The finding remains a configured-code and regression-contract follow-up, with installed exposure unassessed. These 78 records are outside the frozen publication checkpoint. All 61 room-service declarations now have individual body and named local-contract decisions: 41 retain and 20 improve, covering access, governance, invitations, adoption, session continuity, attachments, recovery, presentation, unread state, namespaces and persistence; 60 are new body reviews and one refines an earlier read. Stable IDs and body hashes account for every declaration. Existing reset-preservation coverage was linked and a duplicate recommendation was withdrawn. Remaining test improvements include exact session effects, controlled write failures and independently varied namespace components. The completion receipt SHA-256 is 8125256ce38e4ede3b772ea05d454e6d1482a0e368fc34e6a2890f4c9a2247ff; these source reviews remain outside the frozen publication checkpoint. Successful adoption tests need an explicit atomicity scope and failure/recovery controls where required. Direct-reference inspection places the reviewed invitation/adoption helpers in test-only call paths at this pin, so current-release significance needs separate product-path evidence. All 35 publisher declarations now have individual body and named local-contract decisions: 21 retain, 12 improve and 2 consolidate. Their immutable completion receipt SHA-256 is 31ee4edf256ba1fa815a670fc926b45cd22d3e113007b6470f38d5dbb6d5bdf8. The review preserves useful import, platform and receipt tests, and identifies duplicate test-only import paths, disconnected state assertions and missing positive controls. Replacement coverage must exercise the shipping implementation before duplicate helpers are removed. All 24 release-publication declarations are also assessed: 17 retain and 7 improve. The review read the complete local admission, held-file, snapshot and test code plus named dependencies. Its completion receipt SHA-256 is cae22935f5a5e04308a308d3ec1cc0a33e9080634ca45e3e04c0fcc7fedaf518. Existing exact-byte, cache-read and independent-signature controls stay; additional tests should isolate size, platform and directory-identity failures and observe partial-copy recovery. These reviews remain outside the frozen publication checkpoint. All 17 notification declarations also have complete local-body and named-helper reviews: 13 retain and 4 improve. The completion receipt is 31506708c1cc9a4b5a6055763c5e68065597dee3228aaefd2233e101c8d9024c; a separate receipt corrects one helper coordinate without changing any decision. These tests preserve useful account, unread-state and lifecycle behavior. Improvements add independent scope controls, observed concurrent operations and intended write effects. They remain outside the frozen publication checkpoint. All 21 operator-control declarations now have full local-body and named-helper decisions: 8 retain and 13 improve, with 19 new body reviews and 2 deeper reviews of existing records. Completion receipt 8b29e7d0698903a29ecc866bb0571559f573b022cd73a3abd46a933404102824 binds every stable ID and body hash. The review keeps useful cancellation and signed-update controls and adds request/effect counters, exact persisted peer state, independent identity checks and controlled network fixtures. These source reviews remain outside the frozen publication checkpoint. The export recheck also caught a storage label being treated as a private finding; the correction preserves substantive public assessments while retaining private evidence controls. The published checkpoint is d2a5312a.

All 25 supervisor declarations now have complete local-body and named-helper reviews: 11 retain, 12 improve and 2 replace, with 23 new body reads and 2 deeper assessments. Completion receipt 6b1600c93fd77b7cde194e1d87e996c05ea7c0307b1d8a9d3adcd4a66d5a1aa3 binds every stable ID and body hash. The review keeps useful acquisition, wire and retry checks. Improvements cover independent integrity decisions, exact failure stages, cache recovery, several-resource ownership and fixture cleanup. Literal-only checks should move to the production validator. These source reviews remain outside the frozen publication checkpoint. The complete push gate passed and the checkpoint is published.

The next frozen Rust-provider depth receipt covers 1,383 of 2,412 declarations with complete body and named local-contract review, including 67 independently delegated declarations. It separately counts 2,179 body reads: 796 still need named-contract review and 233 still need body review. Receipt 945299afd7fa4692c7f36c8d2e4477e2e2f9d26d2c1e26f1ba5df11a9dc2952c binds the exact worker cutoffs. Later reviews continue outside that checkpoint. The next export is being reconciled with the completed JavaScript census and existing broader-context reviews.

The next private checkpoint now accounts for 9,361 records, including one recovered standalone self-test: 9,128 body reads and 233 provisional body records. All 3,351 server declarations have attributed body reads, with 199 earlier records still needing explicit local-contract depth reconciliation. The server census verifies 239 files and 3,347 function bodies plus 4 normalized documentation fences. Projection manifest 2064a8a1892432131b7391dd2cdf5a27be3dc174857f415c3ee244109ec2a0d7 is under independent privacy and integrity review. The published PR remains at d2a5312 until that review and publication gates pass.

All 28 Wallet-contract declarations now have complete body and named local-contract decisions: 17 retain, 10 improve and 1 consolidate. This adds 26 body reads and deepens 2 earlier reads; receipt 83594c73e6d03c8809fc27a59fe566772aaf8a7e797cf96aa591aa025a5c5a28 binds the full set. The review keeps useful wire constants, exact hash vectors, closed shapes and result-variant controls. It calls for independent binding checks, exact accepted limits, parser-stage assertions and a smaller duplicate catalog test. This later source batch remains outside projection053 and has no new execution evidence.

Activity

  1. added this to the 0.7.3 milestone on Oct 1, 2026
  2. changed the title [-]Tests and scripts: keep only what catches real bugs, and run the good tests[/-] [+]CI and tests: warm caches, fail fast, keep only what catches real bugs[/+] on Oct 1, 2026
  3. modified the milestones: 0.7.3, 0.7.2 on Oct 1, 2026
  4. andersalm commented on Oct 1, 2026

    @andersalm
    ContributorAuthor

    Part 0: PR #196 merged into develop at 23:40 UTC (356839f0). Its own PR run, still cold (no cache anywhere yet): all checks green; source-home-linux x86 30 min (median before 43), macOS 35 min (before 40), ARM 36 min (before 35), test-elastos 16 min, wall time about 38 min (before 47). The develop push run 36942035859 now saves the shared caches; the next PR runs show whether they restore warm.

  5. andersalm commented on Oct 2, 2026

    @andersalm
    ContributorAuthor

    Part 0 result. #196 and #198 are merged; develop now holds one cache set of about 9.1 GB (under the 10 GB cap), and PR runs restore it without saving.

    Job Median before (24 Sep-1 Oct) PR run 36955539367 (warm)
    source-home-linux x86 43 min 22 min
    source-home-linux ARM 35 min 18 min
    source-home-macos 40 min 17 min
    test-elastos 19 min 10 min
    lint 5 min 2 min
    test-capsules 15 min 13 min

    Wall time 32 min, set by custody-harness-smoke (30 min, cold; new bullet above). Without it the slowest job is 22 min. Two more PR runs are still to be recorded here.

  6. andersalm commented on Oct 2, 2026

    @andersalm
    ContributorAuthor

    Flaky test seen on develop after #201 merged (2 October, run 37068171419): the model-broker fixture reset in test-elastos failed on attempt 1 and passed on the unchanged commit on attempt 2. Candidate for Part 2 (no real-time dependence or a fixed reset).

  7. andersalm commented on Oct 3, 2026

    @andersalm
    ContributorAuthor

    Decision (3 October), prompted by Sazabi deleting 800k lines of unit tests: no wholesale deletion here. Fresh numbers on develop fe033fe: Rust is 586k lines, about 48% test code, 4,827 tests; 90 script test files (30k lines). Develop CI wall time is 33 min, set by source-home-macos (30 min); test-elastos (11) and test-capsules (14) run in parallel, so removing unit tests saves runner minutes and agent time, not wall time. This issue's 1 October audit already found the real lock-in: the ~29k lines of source-text checkers, about 140 Rust delete candidates and the slow real-time sleeps. Parts 1 and 2 start now on a branch, owned by the update session while it waits for the operator's M1/M2 run; merge after #89 M2 passes, with the usual review. Refused-case tests stay.

  8. andersalm commented on Oct 3, 2026

    @andersalm
    ContributorAuthor

    Anders reports #89 M2 passed. PR224 Part1 merged normally into develop on 4 October at 13:10 UTC by Anders. Irzhy’s existing request remains for review after merge under the current rule. Part2 follows Part1’s integrated history.
    The merge preserves the tested candidate and current-base parents; root independently verified the fetched merge commit, exact tree and source ancestry. Its private receipt and lifecycle inventory retain the identities, green checks and clean source. Branches and histories remain preserved.
    Full local gate PASS914.517s and actual-ref hook/push PASS897.569s cover59 fixtures,92 product-data cases,17 workspace checks,22-package Clippy,3 fresh provider builds and24 native targets3816 PASS/27ignored plus1 custody child; owned command groups ended.
    Current CI37192908134 GREEN, posted Opus PASS, local mechanism/integration Opus PASS and same independent high development PASS cover the merged candidate. Anders directed immediate merge with Irzhy follow-up after merge; that review request stays in place.
    Part2 PR225 prepares clean5db8143f/treec6824704, origin0/61; publichead stays1e683d5f. Normal merge92 preserves reviewed Part2 history and develop3ce;82incoming paths equal develop,59Part2 paths equal their reviewed parent. High/local integration Opus PASS;32workflow additions retain23accepted Mac freshness lines plus9Part2 timing/VZ lines, with general Cargo caches retained. CI-policy19cases/high/localCI Opus PASS. The actual-ref gates refused publication at adapter Default, then three unchanged model argument-count signatures; exact source/cause/log and40/21 then111/40PID/group cleanup are retained privately. Appendbfc derives the same RuntimeNetOnly default; append5db adds only three function-local lint allowances. Scoped all-target Clippy/format/diff PASS, two adapter config cases and all35model adapter cases PASS; same high source rounds PASS with parameters/bodies/refusals/CI unchanged. All14/4 and25/5repair PID/groups ended; Bounded model-arity local Opus PASS22.440s on exact5db; source/output hashes match and both review PIDs/one group ended. Fresh fetch and normal merge confirm unchanged develop; The new actual-ref gate refused publication after167.266s when free space fell below the15%reserve. Root matched exact5db/source/log and all76PIDs/38groups absent; leasefree. At16:06UTC the volume has16.94% free and the stable heavy-build lease is free. The disk reserve gate can now proceed; Release has priority for its prepared publication gate. Update requests the next agreed lease for Part2’s fresh-base/full actual-ref gate after Release’s priority work. All source/history/Homes/protected artifacts stay preserved. Current posted review and hosted CI follow publication. First merged-develop timing is GREEN39m31s; three consecutive green develop runs at or below38m and remaining Done criteria still need proof. Existing122deletion decisions/22tables/139net test-attribute reduction retain their scope; source review establishes neither installed acceptance nor overall speed.

  9. andersalm commented on Oct 3, 2026

    @andersalm
    ContributorAuthor

    Current Part 1 Opus PASS and Part 2 Opus PASS leave these nonblocking notes:

    • Part 1's pinned Playwright versions use npm/CDN inputs without a lockfile or integrity pin. Release tags now depend on that test job's availability; shipped artifacts use their existing release inputs.
    • The native namespace/loopback guard proves a privileged disposable fixture. Installed unprivileged Browser acceptance keeps its separate target gate.
    • App CSP uses the tested shared builder; add a focused ordinary-app connect-src assertion when extending its behavior coverage.
    • Part 2's 20s test filters can silently select zero tests after a rename, and package feature resolution can add compilation time. Current proof records actual executed tests; a later gate should check the selected count.
    • The deleted content-provider success test was a mock echo; executable provider-error and artifact hash/provenance checks remain. A future success test should exercise actual decode/dispatch.
    • Presence capacity now seeds MAX-1 signed stored records and sends the boundary record through project(). The overflow, byte-limit and expiry refusals remain; this reduces repeated setup work.
    • Resolved locally at 1a033ae: product children clear hook-local Git variables; caller-bound Git commands retain their repository context. New fixtures prove the separation.
    • Resolved locally at 1a033ae: quoted helper matches stay within a line and apostrophes in comments retain actual consumer selection.
    • Resolved locally at 1a033ae: the parent owns the heavy-build lock, product children drop caller test-binary variables, and fresh qualified provider inputs remain explicit. Detached children still need their own terminal cleanup proof.
    • Current Part 1 hook fetches the approved origin/develop base; a worktree with an alternate remote name needs its own supported base configuration before using this hook.
    • Current review notes that the deleted System wallet-API and Inbox/System/Wallet credential-presence guards have no direct replacement test. Future behavior coverage should test Runtime-owned signing and credential authority when these app flows change.
    • Current UP-03: Restart and recover safely #109 ec1e938 workflow-only local Opus PASS records unused restored target-build beside fresh per-job intermediates. Cold compilation adds disk and time; the existing reserve checks and job deadlines require actual repaired-run proof. Compiled cache reuse can return after a correct dependency-only or scoped-clean boundary is admitted.
    • Its default-directory cache comments describe the restored path, while new hosted job overrides select fresh intermediates. Keep that distinction clear when updating the cache mechanism; local shared intermediates remain separate. The current Part2 arity repair uses three function-local lint allowances at established transport/worker boundaries, preserving all parameters and behavior. Both plain and reason-bearing lint attributes occur in the repository; review suggestions for optional reason text or future parameter bundles stay outside this gate correction and parked model work.
  10. andersalm commented on Oct 4, 2026

    @andersalm
    ContributorAuthor

    Anders’s latest instruction lifts the #89 M2 hold on PR224 and PR225. The recorded Update ownership stays in place; Part 1 precedes Part 2.
    Each branch must merge current develop and pass local format, workspace all-target check, touched-crate clippy and changed-module unit tests before its next push. Current-base green CI and current required reviews gate each merge.
    The same area will carry an executable pre-push check script and git hook. Unclear failed Mac install/update/Home steps require local proof first; large diffs get Opus review before the long Mac round, and owned failures cancel remaining CI jobs.
    Check-ins now run every fifteen minutes while #89 is open. Turns that only wait for CI hand back and inspect the result at the next check-in.

  11. 76 remaining items

  12. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    #194: Root completed 26 more source assessments for setup profiles, provider contracts, platform selection and engine receipts at 460b1888 (batches 018/019; hashes 21cbbc9e88 and 51a8762b85). Existing icon validation is linked to avoid duplicate tests; new actions cover platform-specific fixtures, valid checksum controls and reliable cleanup.
    Draft #487 remains published at 784d1403; reviewed tree 33b0cae2 stays prepared locally pending the requested history-replacement approval. Later research remains in separate immutable private batches.
    The test audit remains open. Root continues setup and update tests while this chat's reviewers cover the remaining server, Rust package and JavaScript declarations.

  13. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    #194: Root assessed 34 more setup tests at 460b1888, including metadata admission, update planning, receipt repair and release validation (batches020/021, hashes 3cf6c453a7 and ff6dde8e7f). The register preserves strong tests and records precise gaps in failure isolation, checksum controls and claims about acquisition or completed updates.
    These are source assessments. Rust execution and installed-product acceptance remain separate proof requirements; detailed trust-boundary findings stay private.
    Draft #487 remains at published commit 784d1403; safe tree 33b0cae2 is prepared locally. This chat continues the remaining per-test research under #194.

  14. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    #194: All 100 test declarations in pinned setup.rs now have complete body and named direct-helper assessments: 59 retain, 41 improve. The completion receipt (82596570a4) binds every ID, source/body hash and contributing batch; 84 rows strengthen previously provisional coverage and 16 strengthen earlier assessments.
    The review separates installed setup, catalogue copying, engine extraction and network configuration from actual engine/media execution and Community joining. Rust tests were reviewed from source; installed acceptance remains open.
    Root now reviews the 24 Browser image tests, then the reserved update and Home command tests. Draft #487 stays at 784d1403; the prepared register tree remains exactly 33b0cae2 while history-replacement approval is pending.

  15. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    The source review now covers all 24 Browser image test declarations at 460b1888; 18 are retained and 6 have specific improvements for failure controls, deadlines, cancellation, or test isolation.
    The private batch is bound to SHA-256 e28a6ef4219129700a5791a73f07e18dd95949ac30a9892f2ac82863d095921f; this is source evidence, while installed Browser acceptance remains separate.
    Draft PR #487 still holds the published register; the reviewed correction stays local pending the existing history-replacement approval. Update and recovery tests are next.

  16. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    The next 21 update tests now have individual source assessments: 15 retain and 6 improve, bound to batch SHA-256 9e3b87e2aff9242964a73fccab2c00a18e042da0fe09f1fc53cd1952a6226bf6.
    Strong checks cover exact request order, signed channel controls and preserved files; the actions add independent failure controls, CID boundaries and recovery-command cases. Detailed boundary evidence stays private.
    Research continues here through installation, rollback and restart tests. Draft PR #487 remains open; the corrected public register remains local pending the clarified approval.

  17. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    The user approved replacing the unpublished audit draft. Corrected commit ceca1d3b1fa78eecf0617ff302902cee89aeae84 has the independently reviewed tree 33b0cae29eb2bbd57da7b43e8af7021cdc0f3280; the withdrawn draft is absent from its history.
    The reader verifies 1,926 files, 9,360 records, 494 finding groups and 71 separate follow-up records. Product-data validation, its 99 tests and both format gates pass; the retired entropy command is absent on this artifact base, whose documented replacement gates passed.
    Publication target: origin/feat/test-audit-register in draft PR #487, one commit ahead and zero behind. The mandatory push gate runs next; the exhaustive review continues here.

  18. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    All 61 update.rs test declarations now have individual source assessments: 35 retain and 26 improve. The completion receipt is bound to SHA-256 b494ecb738888b062e31f4ff8ad038fd3670137ff2167572c6ef9b8aed5f1573.
    The review records exact failure controls, preservation scope, process cleanup, host coverage and the distinction between configuration installation and a working Home journey. This batch contains source review only.
    These newer assessments remain in private immutable batches for the next reviewed projection. Publication checks for corrected audit commit ceca1d3 are running under draft PR #487; controller and Home command reviews continue next.

  19. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    Nine Home command policy tests are now assessed: five retain and four improve, in batch SHA-256 bd2b666844f2da4b9a20dd34cefb79a715655f0d30d4e50b1d472c84bb5d5bf7.
    Two visibility tests inspect static flags rather than the produced Home action list; their actions now require real visible/hidden output checks. Other actions strengthen exact role, toggle and session projection evidence.
    Research continues in this chat. The normal push of corrected audit commit ceca1d3 is still running its mandatory local release-publishing fixtures; the draft PR remains the publication target.

  20. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    Home command review now covers all 33 declarations: 14 retain and 19 improve. The actions include exact notice assertions, separate action prerequisites, shared-environment isolation and bounded fixture cleanup. These source reviews remain private until the next checked register update; installed Home acceptance stays separate.

  21. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    The controller review now covers 17 of its 73 test declarations. Five need changes, including a valid oversized-JSON fixture and a child-process check for inherited environment; the signed-choice and byte-preservation tests retain useful controls. The individual actions are preserved for the next register checkpoint.

  22. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    Controller review now covers 33 of 73 declarations: 26 retain and 7 improve. The added review preserves the useful signed-recovery, lock-ownership and low-space controls; it records bounded FIFO tests and actual receipt publication as follow-up coverage. Each result remains tied to its source body and helper scope.

  23. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    The corrected register is published in draft PR #487 and its acceptance record is updated. It contains 5,191 body reviews and 4,169 provisional records; independent projection review and the full local push gate passed. Review continues here on the remaining records and later source changes.

  24. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    Controller review now covers 50 of 73 declarations: 39 retain and 11 improve. Added actions require separate checks for both recovery locks, the identity of a retained child and finite waits in cancellation/deadline tests. The request-binding, retry and state-preservation controls remain useful.

  25. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    Controller source review is complete: all 73 declarations have individual assessments, with 58 retain and 15 improve decisions. The review preserves recovery coverage and identifies exact failure assertions, controlled inputs and fixture cleanup to improve.
    The completion receipt is 27f13a75b44e12983d9d49aa4d5bb431af742ee94036233bc43b370d9ab49a4f. This is source review; Rust execution and installed journeys remain separate.
    Draft PR #487 remains at ceca1d3 (5,191 published body reviews); this later batch is private pending the next checked projection. Next: remaining test records and current-source reconciliation.

  26. andersalm commented on Oct 22, 2026

    @andersalm
    ContributorAuthor

    SHR-048/049 disposition (22 Oct, OMP train owner) (#194 comment 6348425125): the receipts from the installed protected-content driver (scripts/protected-content-installed-e2e-proof.sh) are not used as acceptance evidence for 0.8.0-alpha.25 or 0.8.0.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions