Record the failed hosted trial of unittest-parallel on the two required test jobs

[seathatflowsinourveins]

Scope

• What this PR changes, in one or two sentences: records the failed preregistered trial of unittest-parallel 1.8.6 on the two required test jobs (hosted run 37109532421 of draft Trial: unittest-parallel for the two required test jobs (draft, preregistered, never merged) #646, head 1e4bb5ab): a validated observed convergence record (decision reject, 36 observations), a dated decision record with the unit's completeness critique (not adopted; PR-A and PR-A2 are not opened), a durable sanitized receipt, supporting artifacts with the sha256 of the originals that expire on 2026-11-02, byte copies of the 15 files the preregistration froze (the observed record's frozen inputs), and four anti-pattern rows. Docs and evidence only: no workflow, test, ruleset or setting changes.
• Base commit: 4ced2923063db6a6dcafa9f25af5ee05a4153c75
• Lane: lane:foundation
• Owned paths touched: docs/decisions/2026-10-03-suite-parallelism-trial-outcome.md, evidence/receipts/suite-parallelism-trial-20261003.json, evidence/artifacts/suite-parallelism-trial-20261003/ (13 files, among them the observed convergence record experiment.json), evidence/artifacts/suite-parallelism-trial-20261003/frozen/ (15 byte copies, stored as .txt, except the two lock copies, which are named *-pins.frozen so no scanner pattern matches them), docs/harness-defaults.md (five rows at the top of the anti-pattern table), manifests/evidence.json (registration and the convergence_records entry only, the branch's last commit, per the hot-file protocol in docs/lanes.md)

SOTA sources

• craigahobbs/unittest-parallel at release commit bda5d77dc1a2fa2df90f5f5a7de297ea375e345c: src/unittest_parallel/main.py lines 150-160 (spawned pool, pool.map at 160), 164-211 (report only after the map), 240-242 (exit status), 62-63 (--thread); README.md lines 43-71; issue 27; commits after the release; PyPI JSON (1.8.6 latest on 2026-10-03).
• python/cpython at v3.12.3 (f6650f9ad73359051f3e558c2431a109bc016664): Lib/unittest/async_case.py lines 35-38, Lib/unittest/case.py lines 159-161, 176-182 and 612-617 (a class-level skip only marks the class and applies when a test runs), Lib/unittest/loader.py line 94 (one instance per test name), Lib/multiprocessing/pool.py lines 367, 528-546, 774 and 809-831, Modules/_pickle.c lines 3646-3704; at v3.13.16 (cbc944f4bc59639a444dd971c737788ba2283a91): Lib/unittest/async_case.py line 42.
• github/docs at 068546469ae7f079368d12f969991a045121c4a0: data/reusables/actions/workflows/triggering-a-workflow-paths5.md (a pull request's paths filter uses the three-dot diff). GitHub REST: workflow jobs, artifacts.
• Completeness critique, read on 2026-10-03 after the trial: repository files with GET requests at these pinned commits (each tag resolved through the GitHub commits API): pytest-dev/pytest 9.1.1 at cf470ec0bf7eb89cd97dd56df4859eae5db46447; pytest-dev/pytest-xdist v3.8.0 at 1e3e4dc16523c8a8f6c67d95a950166420718c99; mtreinish/stestr 4.2.1 at 2802f1425f45b9ad0857ff4685ab469dcd2cf2e0 (stestr/output.py lines 154-161); testing-cabal/testtools 2.9.1 at 088c98e24961ecf6d94ea5204457f2dcffe2f1c6; nose-devs/nose2 0.16.0 at c93ca65ac6f62475209aa8944fda14c671615549; cgoldberg/concurrencytest 0.1.11 at 266e27c833f9b1ecb47142bdddbac5199bebab72 (concurrencytest.py lines 89-90, 158-160 and 179-187); testing-cabal/subunit 1.4.6 at c85605280cb1d975b3076b9ad7ac38f17e858945; CleanCut/green 4.0.2 at 4de285b05b8e6b161159be2241b219dc5176f0e5; zopefoundation/zope.testrunner 8.3 at 1061ccc4b6b5a824870f2142bc948863b7ba1f70; python/cpython v3.13.16 at cbc944f4bc59639a444dd971c737788ba2283a91 (Lib/unittest/loader.py, Lib/test/libregrtest/); craigahobbs/unittest-parallel at bda5d77d (main.py:123-128, :142-160, README.md:59-70); github/docs at 2bd66de8cea336061c9ea060c9b37385136e6ab3 (workflow-syntax.md: background, wait, wait-all, cancel, parallel; limits.md lines 64-69, concurrent macOS jobs per plan). With GET requests that no commit pins: the GitHub changelog of 2026-06-25; the GNU findutils 4.10.0 announcement; PyPI JSON per release and the repository records; unittest-parallel's issue tracker (issue 22, issue 27 and a search for "pickle") and CPython's (python/cpython#83438, bpo-39257, and searches for IsolatedAsyncioTestCase with pickle, pickling or multiprocessing). Locally: the installed GNU xargs 4.10.0 manual page (the gnu.org manual did not answer).
• In-repository, at the base commit: docs/acceptance-evidence-policy.md:55-63 (preserve failed attempts, declare every sanitization), docs/evidence.md:18 (compatibility_attempt), docs/lanes.md:94-141 (hot-file protocol), blueprints/convergence-practice/contract-reference.md:27-28 (distinct source, input and evaluation artifacts) and :84-92 and scripts/validate_convergence.py:99-123 and :136-146 (usage fields, failure coverage, run status, scope of qualifying runs), tools/sota-convergence/landscape-sweep/build_inputs.py:21-22 and :225-234 (the seeds format); for matrix sharding, .github/main-ruleset.json:35 and :41, docs/github-automation.md:26-28 and tests/test_workflow_hardening.py:1249-1257. At the trial head 1e4bb5ab: the 15 frozen files (byte-copied under frozen/) and the preregistration's decision record, "Alternatives".

Evidence-class table

Claim	Evidence class	Command / receipt
The observed convergence record: 36 observations, of which 12 are timed ubuntu-24.04 runs (S passed 3 of 3; L4, L4F and L4C failed 9 of 9 with exit 1), 4 are ubuntu-24.04 control jobs (passed; each observation's exit code is its control step's own exit, and the commands' own exits and the controls file's step seconds, 0, 0, 1 and 0 s under S, L4, L4F and L4C, are in its scope text) and 20 are macos-15 jobs that never received a runner (skipped, no exit code, no quality); decision reject, no qualification, token usage null (unknown), native_retries 0 for the 16 executed jobs, runner time in each executed observation's scope text (11,461 s over the 16 observed jobs); frozen_inputs cite byte copies of the 15 files the preregistration froze (3 sources, 3 inputs, 9 evaluation, in its order, each with its preregistered sha256) and the preregistration copy at the end of sources; every departure from the plan stated in limitations	native_cli_execution observations generated from the raw run directories, the jobs API (GET, steps included) and result.json; scripts/validate_convergence.py checks declared consistency and artifact hashes, not truth	evidence/artifacts/suite-parallelism-trial-20261003/experiment.json
The 15 byte copies under frozen/ equal their originals at 1e4bb5ab byte for byte and carry the preregistered sha256 values (15 of 15); all pass scripts/validate.py --scan-file; 264,054 bytes	reproducible artifact check (git show at 1e4bb5ab against each copy, and sha256 against the preregistration's frozen_inputs)	evidence/artifacts/suite-parallelism-trial-20261003/frozen/; receipt data.retained_files
Completeness critique: the missed modality (no whole-suite local run under the candidate runner; the hosted behaviour of background steps never examined), the sources missed before the trial (not recorded as read by the preregistration or its decision record: unittest-parallel's issue tracker, whose issue 27 is titled with the trial's exact exception; CPython's tracker, issue 83438; the suite's own async and type()-built test modules; GitHub's changelog), nine candidate classes the trial did not evaluate plus matrix sharding with its blockers, each relayed claim marked VERIFIED or LEAD-ONLY, and seeds for the next ci-supply-chain sweep	source review, by source class: repository files with GET requests at the pinned commits above; repository records, PyPI JSON, the GitHub changelog, the GNU findutils announcement and the two issue trackers with GET requests that no commit pins; the GNU xargs exit statuses from the installed GNU xargs 4.10.0 manual page; the unittest.mock count (106 of 234 test modules, 100 at module level), the suite's modules and the matrix-sharding blockers locally with git grep and reads at 1e4bb5ab and the base. The candidate list is a lead from a read-only single-family Codex lane	decision record, "Completeness critique"
On ubuntu-24.04 (CPython 3.12.3, image 20260927.320.1) S ran 9,823 tests, OK (skipped=967), exit 0, in 3 of 3 runs (1,688, 1,659 and 1,186 s); all 9 unittest-parallel runs (L4 649/438/459 s, L4F 539/692/710 s, L4C 667/659/704 s) exited 1 with TypeError: cannot pickle '_contextvars.Context' object from pool.map (main.py:160); 199 ids never ran at module level and 82 at class level (in S, 90 of the 199 and 8 of the 82 were ok and the rest skipped, all 66 IsolatedAsyncioTestCase tests among the skipped); oracle: ubuntu-24.04 reject, macos-15 no verdict; id mapping ok	native_proven (hosted run 37109532421, attempt 1; receipt retained with hashes and sanitized excerpts)	evidence/receipts/suite-parallelism-trial-20261003.json; evidence/artifacts/suite-parallelism-trial-20261003/
Every control and crash control passed: the controls file reported Ran 8 and FAILED (failures=2, errors=2, skipped=1, expected failures=1, unexpected successes=1), exit 1 under S and 5 under unittest-parallel; the crash control exited 3 under S and 124 (the 300 s bound) under every parallel arm	synthetic: synthetic fixtures executed on the hosted runner (run 37109532421)	receipt data.controls; control-logs.txt
K4 timing FAIL in 3 of 3 L4C runs; test_cli_refuses_live_base_url_from_env_file_without_network ERROR in 3 of 3 L4F runs	native_proven (observed); causes not established, the L4F one unknown	receipt data.non_missing_mismatches
Jobs-API test-step seconds equal each run's own; the 20 macOS jobs never received a runner; 5 of this repository's macOS jobs held runners in 46 of 50 minutes of the queued window	independent observation of platform records (read-only gh api --method GET); nearest template class source_review	receipt data.jobs, data.cancellation
Root causes: IsolatedAsyncioTestCase stores a contextvars.Context in __init__, so its suites cannot be pickled even when every test in them is skipped, as all 66 were on ubuntu-24.04; four type()-built classes in tests/test_adoption_version_probes.py are not importable under their __qualname__	local_integration (toy modules on CPython 3.12.3 and 3.13.16; the three real modules on 3.12.3 only; the coordinator also saw the version-probes PicklingError on 3.13.16, an output not in the retained artifacts) plus source_review (async_case.py, case.py, loader.py, pool.py, _pickle.c at pinned tags)	local-reproductions.txt; receipt data.root_causes
Upstream issue 27 (closed by its reporter, no maintainer comment), README silent on asyncio and pickling, 2 commits after the release touching only AGENTS.md, PyPI latest 1.8.6	source_review (GET reads on 2026-10-03)	receipt data.upstream
Speed ratios (L4 0.277 and 0.547; L4F 0.417 and 0.599; L4C 0.402 and 0.594)	computed from measured seconds; not an adoption claim (no arm eligible, runs incomplete, three runs per arm, and each slowest-over-fastest ratio divides by the single fastest S run, 1,186 s, inside a serial spread of 1,186 to 1,688 s)	receipt data.arms_ubuntu_24_04
Usage: the hosted jobs ran deterministic commands and no model; runner time is the per-job seconds of the 18 ubuntu-24.04 jobs (11,524 s in total); the 20 macOS jobs were never assigned a runner; the model usage of the planning, build and review sessions is unknown and not counted as zero	independent observation of platform records (jobs API) for the job seconds; unknown for model usage	receipt data.usage, data.jobs
The run's cancellation at 09:12:40Z	coordinator decision, not part of the preregistered rule; recorded with the run and job records	decision record, "The cancellation"
Repository checks on this branch	local_integration	commands below

Local commands run

$ python3 scripts/validate.py
passed (69 components, 187 receipts, 9,450 hashed files); exit 0
$ python3 scripts/validate_convergence.py evidence/artifacts/suite-parallelism-trial-20261003/experiment.json --root . --json
valid: true, 36 observations; exit 0
$ python3 scripts/validate_convergence.py --all-recorded --root . --json
valid: true, 27 records (this record among them, 36 observations); exit 0
$ python3 scripts/evidence_manifest.py --check
{"files": 9450, "status": "passed"}; exit 0
$ python3 scripts/component_matrix.py --check
{"rows": 32, "status": "checked"}; exit 0
$ python3 scripts/new_host_grand_list.py --check
{"status": "passed", "layers": 32, "winners": 66}; exit 0
$ python3 -m unittest tests.test_osv_lockfile_coverage.LockfileInventoryTests.test_every_tracked_lockfile_and_manifest_is_listed tests.test_blind_checkout.RepositoryClassificationTests.test_every_blueprint_value_under_a_label_key_is_classified tests.test_workflow_security_coverage.NewWorkflowSecurityCoverageTests.test_all_published_workflows_are_listed_and_covered
Ran 3 tests, OK (zizmor 1.30.1 and actionlint 1.17.0 on PATH); exit 0
$ python3 -m unittest discover -s tests -p test_adoption_docs_consistency.py
Ran 39 tests, OK (skipped=1); exit 0
$ python3 -m unittest tests.test_blind_checkout tests.test_ecosystem_manifest tests.test_jcodemunch_config tests.test_verdict_review_gate tests.test_gitleaks_config tests.test_validate tests.test_claude_repository_evidence tests.test_workflow_hardening tests.test_osv_lockfile_coverage tests.test_workflow_security_coverage tests.test_pre_push_gate tests.test_new_host_grand_list tests.test_catalog_freshness_propose tests.test_convergence_contract tests.test_evidence_manifest tests.test_lane_packets tests.test_claude_recovery tests.test_gpt6_family_tiering_20260926 tests.test_local_inference_latest_20260926
Ran 927 tests, OK (skipped=30); exit 0 (at the head 405014de)
$ python3 scripts/validate.py --scan-file <each of the 31 added or changed files other than manifests/evidence.json>
passed, 31 of 31; exit 0
$ for each frozen copy: git show HEAD:<copy> | sha256sum, against git show 1e4bb5ab:<original> | sha256sum and the preregistered sha256
15 of 15 equal (264,054 bytes); exit 0
$ rg -n -i -f <pattern file kept outside the repository> <every added line of git diff origin/main HEAD, 13,371 lines>   # host path prefixes, the user and host names, the email identifier, the account's private repository names (word-bounded), dollar figures (\$[0-9])
4 matches; exit 0: the shell positional parameters kind="$1" and control_file="$2" at lines 252-253 and 452-453 of frozen/github-workflows-macos-suite-parallel-trial.yml.txt, a byte copy that cannot change without breaking its preregistered hash; they are not dollar figures. The same patterns over the other 13,367 added lines: no match; exit 1
observed-record generator kept outside the repository: re-derives every per-run value from the raw run directories (exit-code.txt, step-seconds.txt, command.txt, meta.json, runtime.json, git-status.txt, log.txt), the jobs API read with GET (steps included) and result.json, independently of the receipt, re-hashes the 15 frozen files of the preregistration at 1e4bb5ab and checks each byte copy under frozen/ against them
194 of 194 checks (exit codes, test-step seconds equal to the jobs API's, job seconds, missing ids per run equal to compare.py's, crash lines, control exits, macOS jobs without a runner, exact speed ratios, 15 of 15 frozen hashes, 15 of 15 byte copies); exit 0
consistency pass kept outside the repository over the decision record, the receipt, the observed record, the artifacts README, the draft thread replies and this description: the observed record against the re-derivation, every number repeated across them, each missing id's serial outcome parsed from the three S logs (module level 90 ok / 109 skipped, class level 8 ok / 74 skipped, identical in all three S runs), the controls files' raw step seconds (0, 0, 1 and 0 s), each frozen copy against git show at 1e4bb5ab and the preregistration's role lists, the receipt's retained-file hashes (28 files, recursively), the new sections and the superseded phrasings absent
655 of 655 checks; exit 0
$ git fetch origin main; git merge-tree --write-tree --name-only origin/main HEAD
main has not moved (4ced2923); clean, exit 0

Local reproductions (unittest-parallel 1.8.6, coverage 7.16.2):
$ python -m unittest_parallel -j 2 --level module -v        # toy module: one plain and one IsolatedAsyncioTestCase class
TypeError: cannot pickle '_contextvars.Context' object; exit 1 (CPython 3.12.3 and 3.13.16)
$ python -m unittest_parallel -j 2 --level class -v         # same toy module
the plain test ok, then the same TypeError; exit 1 (both)
$ python -m unittest_parallel -j 2 --level module -v        # plain-only toy module
Ran 2 tests, OK; exit 0 (both)
$ python -m unittest_parallel -j 2 --level module -s tests -t . -p test_adoption_version_probes.py
_pickle.PicklingError: Can't pickle <class 'tests.test_adoption_version_probes.LinuxReportUnderbash32Tests'>: attribute lookup LinuxReportUnderbash32Tests on tests.test_adoption_version_probes failed; exit 1 (3.12.3)
$ python -m unittest_parallel -j 2 --level module -s tests -t . -p test_adaptive_paper_fees.py   # and test_adaptive_paper_transport.py
TypeError: cannot pickle '_contextvars.Context' object; exit 1 each (3.12.3)
$ python -m unittest_parallel -j 4 --level class -v -s tests -t . -p <each of the three modules>
6 of 22, 54 of 61 and 57 of 116 results, then the PicklingError or the TypeError; exit 1 each (3.12.3)
$ python -m unittest tests.test_order_throughput
$ python -m unittest_parallel -j 2 --level module --disable-process-pooling -s tests -t . -p test_order_throughput.py
Ran 76 tests, OK (skipped=1); exit 0 each (3.12.3 and 3.13.16)

Read-only GETs on 2026-10-03: gh api --method GET for the run, its 38 jobs (again with steps for the observed record), its 18 artifacts, PR #646 and the review comments of this PR, the macOS jobs of the repository's runs around the window, craigahobbs/unittest-parallel (repository, tags, issue 27 and its comments, compare, contents at bda5d77), python/cpython (including Lib/unittest/case.py and loader.py at v3.12.3, and loader.py and Lib/test/libregrtest at v3.13.16), github/docs contents at the pins, and the critique's sources at the pinned commits above (contents, tag-to-commit lookups and repository records); PyPI JSON, the GitHub changelog page and the GNU findutils announcement with curl (the gnu.org xargs manual did not answer; the installed GNU xargs 4.10.0 manual page was read instead). For the delta review, also with gh api --method GET (queries in the URL, no -f or -F): stestr/output.py and concurrencytest.py at their full release commits and the two tag-to-commit lookups, unittest-parallel's issue list, its issue search for "pickle" and issue 27 with its comment, python/cpython issue 83438 with its comments and eight issue searches, one code search of github/docs, the github/docs commit lookup (2bd66de8) and limits.md at 2bd66de8, the state of PRs #646 and #652 and the repository's owner type; and the GitHub changelog page again with curl.

Decision record

docs/decisions/2026-10-03-suite-parallelism-trial-outcome.md: the decision (not adopted), the observed convergence record it cites, what was tried with the preregistered rule and a pointer to #646 at 1e4bb5ab, what happened, the cancellation and why, root causes, what the attempt does not show, departures from the preregistration (among them where the observed record lives and the seven departures of its shape), the completeness critique (missed modality, missed sources, candidate classes including matrix sharding) with its seeds, overturn conditions and next-trial preconditions.

Host evidence

Not applicable: no files under evidence/hosts/ change.

Checklist

• New/changed GitHub Actions are pinned to a full commit SHA with a version comment (no workflow changes in this PR).
• New/changed workflows declare top-level permissions: contents: read (no workflow changes in this PR).
• No secrets are printed, logged or committed; no new required secret was added.
• No new paid hosting, subscription or billing surface was introduced.
• Peer-owned untracked files and worktrees were preserved (not deleted, moved or overwritten).

Notes for the merger: manifests/evidence.json is the shared hot file; if main moves, take main's copy, register again docs/decisions/2026-10-03-suite-parallelism-trial-outcome.md, docs/harness-defaults.md, evidence/receipts/suite-parallelism-trial-20261003.json and the 28 files under evidence/artifacts/suite-parallelism-trial-20261003/ (15 of them under frozen/), append evidence/artifacts/suite-parallelism-trial-20261003/experiment.json to convergence_records again, then run the --write commands, scripts/validate.py and scripts/validate_convergence.py --all-recorded --root . --json (docs/lanes.md). #632 inserts five rows at the same top of the anti-pattern table, so whichever of the two merges second has a trivial table conflict (keep both sets of 2026-10-03 rows). This branch was rebuilt on main at 4ced2923 (after #634 and #636), with this PR's five rows above the 2026-10-02 rows; when main moves again, git merge-tree is likely to report conflicts in those same two files (keep both sets of rows; take main's manifests/evidence.json and register again as above). #646 was closed unmerged at 11:05:52Z on 2026-10-03, ten seconds after this PR opened; do not reopen it or open another pull request that adds its workflow, because either re-runs the whole trial. The record names no private repository and no dollar figure.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-03T11:13:40.715580Z	571520b	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 571520b084

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[seathatflowsinourveins]

Exact-head SOURCE ACCEPT at 571520b084c2710d4619c4ecdd8ef9b0a9e1df62, base/current main read 4ced2923063db6a6dcafa9f25af5ee05a4153c75. Scope: publication and classification of the failed suite-parallelism attempt. This supplies no adoption, macOS execution, SDK qualification or fresh test run.

I checked the actual pinned publication, original hosted compare artifact, frozen preregistration and GitHub run/job records. The decision follows the recorded eligibility rule: nine Linux candidate runs exit 1 and remain ineligible; three serial runs exit 0 and are eligible; Linux is rejected. Twenty macOS jobs have no assigned runner and zero steps, so macOS has no verdict. Candidate timing below the speed thresholds does not qualify an incomplete, crashing suite.

The frozen experiment equals the retained byte copy: 43,208 bytes, SHA256 305fc2fa2ddd2c0a0eba4d85d748807db87302bde59a6b6489de866bee416c08. The original run is attempt 1 at full head 1e4bb5ab7c79faa83e4cd8d04cf8dc6d33c7b0de. Original compare artifact 11270430044 has ZIP SHA256 bc0136254601b3ece5ecfec7c2a0525bcbc60c0f606de1b03c05ff7c729d6f2c; its result is 815,073 bytes/SHA256 f26cb0c1c460029ffd481a844bc524ae5e83c84b73fda52cc3efe3496475eb9a. The six declared unchanged fields, all retained fields of twelve runs, reconstructed missing-id sets and their serial outcome counts match the original. Summary and checkout-status byte copies match too.

Original job records reconcile all twelve arm-step durations and the 11,524 seconds across eighteen Linux jobs. Serial arm-step durations are 1,688/1,659/1,186 seconds. Step success alone is not an arm exit-code proof: this trial intentionally records failing arm exits inside a successful collection step. The original result retains the actual arm exits. The publication explicitly records cancellation outside the preregistered rule and the changed record location. Those departures stay visible. The proposed cross-platform failure is explicitly an inference; it supplies no macOS run.

The source mechanism is supported by unittest-parallel at bda5d77dc1a2fa2df90f5f5a7de297ea375e345c, which sends suites through a spawned pool before reporting, and CPython at f6650f9ad73359051f3e558c2431a109bc016664, which stores a context while constructing an async test instance. The L4F error's cause and K4 contention explanation remain unknown; no suppressed traceback is invented.

All 27 actual Git source identities verify. The registry preserves all 9,420 base paths in relative order, adds fourteen, replaces only the owned harness-defaults binding, and retains all 187 receipt records and 26 convergence records unchanged. All fifteen content bindings and twelve retained-file bindings match their actual bytes. Preserve current-main foreign rows when composing later merges.

The evidence classes and usage correctly distinguish repository execution, synthetic controls, local reproductions, source inspection, independent platform observations and unknown planning/build/review model usage. These are not upstream acceptance. I did not redownload the twenty full raw logs or rerun the trial.

Verification custody: 34 native GH commands exit 0, retained stdout/stderr manifest SHA256 9b3734aad72471ddfc2e9e0c5143d38a962ff0715daa0cc33d23690668e76297; original artifact and source bytes were independently checked by root. Source ACCEPT is limited to this failed-attempt publication and its declared boundaries. Required hosted checks and fresh-main composition remain merge conditions.

[seathatflowsinourveins]

Claude session native-agent-stack-5f: landing at head e87dd3b9be8a8476b160a862aba2d1fae3824947. The command center (wsl-architecture-design) gave its ACK at this exact head; under the user's 2026-10-04 decision it owns merges that touch the hot registry. The read-only cross-family review (GPT-6.1 Sol max, the packaged Codex SDK worker) returned FINDINGS at 13e084c93aab; this head is reached from it through recorded carry edges (equal owned patch-ids, or a cross-family delta read returning ACCEPT at the edge's target), with its 1 P1 item(s) resolved in one repair round as recorded.

Observed main 14048b840425c2569e0df60a6596e94e601da15b. coordination/merge_tree_landing_check.py (sha256 a11649fcf2e5844c…) exit 0:

main 14048b840425c2569e0df60a6596e94e601da15b head e87dd3b9be8a8476b160a862aba2d1fae3824947 base 14048b840425c2569e0df60a6596e94e601da15b merged-tree 3c54d436ffbc01183bce44f7e328d4cd46dcd890 merge-tree-exit 0
ok   1: clean three-way merge
ok   2: merged-vs-main paths 32, outside PR-owned 0 []
ok   3: main drift 0 paths, overlap with PR-owned inputs (registry excluded) []
ok   4: registry foreign rows equal True, order preserved True, PR-owned rows 31, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (9926 rows)
LANDABLE
condition 5: merged files[] sorted and unique (tree 3c54d436ffbc01183bce44f7e328d4cd46dcd890)

Required checks at this head: 7 pass 1 skipping . Unresolved review threads: 0.

[seathatflowsinourveins]

Claude session native-agent-stack-5f: post-merge observation. Landed as 4cfac6c4add2446c315882be72477f7dff65a60f (parent 14048b840425c2569e0df60a6596e94e601da15b), tree 3c54d436ffbc01183bce44f7e328d4cd46dcd890. The landed tree equals the checked merged tree and the parent is the observed main.