Skip to content

Make proactive harness improvement explicit in the shared core - #881

Merged
seathatflowsinourveins merged 2 commits into
mainfrom
codex/ns2604-harness-rules-g6a-20261008
Oct 8, 2026
Merged

seathatflowsinourveins merged 2 commits into
mainfrom
codex/ns2604-harness-rules-g6a-20261008

Conversation

@seathatflowsinourveins

@seathatflowsinourveins seathatflowsinourveins commented Oct 8, 2026 •

Copy link
Copy Markdown
Collaborator

Scope

Make automatic harness improvement explicit in bullet four of the shared engineering core, using the existing six instruction carriers. Organize the current responsibilities into five groups with a trimmed enforcement map and a source-backed amendment to the earlier decision.

  • Base commit: 0d5672b08beb87a06daee792afeb66481d0c71e6.
  • Candidate head: c7a2759f0a50e3df139732fb85a7cfec42acd75e; prior reviewed core head: a301fa29befc4f98bf50d46696ff5116fb78991f.
  • Lane: lane:shared; renewed acknowledgements required from rnd-r1, overlap-token, memory-h2h and skills-lifecycle before merge.
  • Owned paths: six instruction carriers; CORE/pin/budget constants in two test modules; three decision/enforcement documents; five receipts and their index under evidence/artifacts/harness-rules-20261008/; matching evidence-manifest rows. Exactly 18 paths.

G6a contains the core change. Settings ownership follows in G6b, and wider maintained-record rewrites follow in G6c. Both remain separate from this draft.

The correction after the first read removes settings-state and F9/template-merge conclusions from the enforcement map. Against the prior reviewed head, only the map and its one evidence-registration row change: two files, +5/-5. General native-hook enforcement and bounded role/session coverage remain. Native registration, integrity validation and independent delta review pass at the correction; the four shared-boundary ACKs carry forward under the reviewed two-file condition.

SOTA sources

Evidence-class table

Claim Evidence class Command / receipt
Exact approved core appears once in each of six carriers; first three bullets unchanged source_review Core SHA-256 85eb7d41af97d086f76e4b96e55ddaf76240240fe55425a054f1954c85ff59c9; independent reduced-scope audit at candidate head
CORE/pin/budget updates retain all six method bodies and 17 assertions local_integration / source_review 32 targeted tests pass; AST comparison against base; D-test-impact.json retains the separate original preparation controls
Controlled text projections add 114 Claude estimated memory tokens and 66 Codex gateway-reported input tokens native_proven D-native-two-scope-before.json and D-native-two-scope-proposed.json; installed native counters with fixed inputs and fresh one-turn Codex threads
Both research routes ran; substantive claims checked against primary sources native_proven / source_review B-run-receipts.json and B-source-receipts.md; retrieval failures and mutable-documentation limits retained
Reduced-candidate integrity and scope validation pass local_integration python3 scripts/validate.py: 70 components, 10,877 hashed files, four profiles, 234 receipts; rc 0
Local full discovery at prior core head a301fa2 UNVERIFIED The one SUITE-GO run timed out at 1500 seconds, rc -15, without a final summary; no completed local counts or pass inferred
Full hosted validation at prior core head a301fa2 local_integration All eight shards and aggregate PASS in run37825316710: 11,699 run, 10,657 pass, 0 fail, 0 error, 1,042 skip; native qualified counts complete
New-head validation at c7a2759 UNVERIFIED Native python3 scripts/validate.py and the independent map/registration delta audit pass. Fresh hosted shard/context results remain pending; the previous run is separate evidence

Counter results are text projections. They do not establish billing savings, effective service tier, semantic adherence or final live loading. Original preparation controls remain separate from reduced-head acceptance.

Local commands run

$ python3 scripts/validate.py
rc 0; 70 components; 10877 hashed files; 4 profiles; 234 receipts.

$ git diff --check 0d5672b08beb87a06daee792afeb66481d0c71e6
rc 0 at the final candidate before commit.

$ nice -n 10 ionice -c2 -n7 python3 -B -m unittest discover -s tests -t .
One run at a301fa29befc4f98bf50d46696ff5116fb78991f, 18:29:34Z–18:54:35Z.
Timed out at the 1500-second watchdog, rc -15, without a final summary.
Run/pass/fail/error/skip counts unavailable; no local full pass inferred.

The completed base comparison covered tests.test_skills_manifest, tests.test_new_wsl_client_config and tests.test_codex_worker_lane: 384 run, 0 failures, 0 errors, 12 skipped. Full base discovery also timed out at 1500 seconds, so there is no completed full-base count. Module counts, incomplete local discovery and completed hosted-workload counts remain distinct. No additional local full run is started for the map-only correction.

At prior core head a301fa2, all eight validate shards and the aggregate pass. Native artifact validate-shard-summary-37825316710-1/aggregate.json has SHA-256 839f4bfb3a7c1746ef59521d3d5a3ddc02d538fb9f07747c285ae5c1931eb10d; no unknown or unqualified count reports. That CI run tests synthetic merge 2a38a67ecd16eab21bbc54171ddce60a30d37f7d, whose verified parents are main a484989a43f70ff2575b6fb054cf00c9834e7679 and prior head a301fa29befc4f98bf50d46696ff5116fb78991f. These are the prior run's per-shard results; new-head results are pending.

Shard Result Run Fail Error Skip
0 SUCCESS 1817 0 0 75
1 SUCCESS 1304 0 0 202
2 SUCCESS 1836 0 0 217
3 SUCCESS 1682 0 0 99
4 SUCCESS 1046 0 0 37
5 SUCCESS 1169 0 0 80
6 SUCCESS 1661 0 0 168
7 SUCCESS 1184 0 0 164

At prior head a301fa2, native gh pr checks 881 --required returned rc0: validate, secret-scan, dependency-review, sota-sources, verdict-review-gate, osv-scanner and token-report all SUCCESS. Required results at c7a2759 remain pending and must pass before landing.

Decision record

docs/decisions/2026-10-08-harness-rules-consolidation.md records behavior, sources, alternatives, measured limits and overturn conditions. docs/harness-rule-enforcement.md retains twelve rule families, gaps G01–G10 and seven observed required contexts. The earlier decision's amendment changes only its six-line region; every byte outside that region remains as at base.

The shared core grows by 169 UTF-8 bytes and 22 words. Declared startup ceilings are 4,965 bytes for Claude and 5,071 bytes for Codex.

Host evidence

No evidence/hosts/ files change; the host-receipt checklist does not apply. The explicit 5f landing cue precedes CC-run F9 and fresh-session load checks in both clients. Live user instructions and client configurations have not been applied.

Checklist

  • New/changed GitHub Actions are pinned to full commit SHAs with version comments — no workflow changes.
  • New/changed workflows use scoped permissions — no workflow changes.
  • 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 files and worktrees were preserved.
  • Required shared-surface acknowledgements are recorded against the reduced patch; the map-only two-file correction preserves their reviewed boundaries.
  • All eight validate shards pass at the current PR head.
  • CC re-read, co-op GPT review and explicit 5f landing cue are recorded.

@seathatflowsinourveins seathatflowsinourveins added the lane:shared Touches files owned by both lanes; needs both lanes' acknowledgement label Oct 8, 2026
@seathatflowsinourveins
seathatflowsinourveins marked this pull request as ready for review October 8, 2026 20:02
@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

@seathatflowsinourveins
seathatflowsinourveins force-pushed the codex/ns2604-harness-rules-g6a-20261008 branch from c7a2759 to 2dbdce7 Compare October 8, 2026 20:32
@seathatflowsinourveins
seathatflowsinourveins merged commit e48de2e into main Oct 8, 2026
36 checks passed
@seathatflowsinourveins
seathatflowsinourveins deleted the codex/ns2604-harness-rules-g6a-20261008 branch October 8, 2026 20:41
seathatflowsinourveins added a commit that referenced this pull request Oct 9, 2026
### Scope

The dated readiness manifest still reports G6 as HOLD after #881 landed and the F9 receipt accepted G6a. Split that gate into G6a (MET), G6b (ongoing settings ownership), and G6c (ongoing records, #885/#886), and bind G5's six closure requirements without granting G5 acceptance. Update the matching R&D and follow-up state displays from the same accepted receipts.

- Base commit: `aba02ec3456d383bcc2fc72883db098f9be7918a`
- Lane: `lane:foundation`
- Owned paths touched: `tools/north-star/sources.json`, `catalogs/north-star/readiness.json`
- The existing builder regenerates the complete dated view, including current main-input provenance digests. Its 20 layer scopes and 36 SDK items remain. The evolving Monitoring status source matches the owner selector twice, so that single field becomes UNVERIFIED; its bounded snapshot is distinct from the separately preserved Monitoring follow-up.

## SOTA sources

- Existing reference implementation: [native-agent-stack at aba02ec, tools/north-star/build_readiness.py](https://github.com/seathatflowsinourveins/native-agent-stack/blob/aba02ec3456d383bcc2fc72883db098f9be7918a/tools/north-star/build_readiness.py), `Receipts.claim` and the supported `--write`/`--check` interface. This change uses the shipped receipt adapter and its digest-bound selectors. The gate-refresh procedure is documented in [tools/north-star/README.md, Refresh on a gate change](https://github.com/seathatflowsinourveins/native-agent-stack/blob/aba02ec3456d383bcc2fc72883db098f9be7918a/tools/north-star/README.md#refresh-on-a-gate-change).
- The original adapter's retained upstream survey is `research/fullspeed-20261008/ns-readiness-manifest/upstream-research.md` under the durable state root; the implementation follows [Python 3.13 re](https://docs.python.org/3.13/library/re.html) for exact text selection and [SLSA v1.2 verification](https://slsa.dev/spec/v1.2/verifying-artifacts) for the distinction between input identity and acceptance. This is a source-binding refresh of that accepted implementation.
- Gate evidence: `coordination/command-center/windows/f9-g6a-881-20261008/RECEIPT.md`, SHA256 `24906eaf9cc3f041ed6bd500dd41095e4282a551ad1403d80a76a709ce7cf6a1`; line 3 identifies [#881](#881 merge commit and timestamp, line 5 records F9 application, lines 14–15 record both fresh-session checks, and line 17 adjudicates G6a and the remaining G6 scopes. Installed `gh` 2.102.0 independently returned the same #881 merge metadata and OPEN draft record PRs [#885](#885) and [#886](#886).
- G5 requirements: `coordination/command-center/ITEM-ns2604-coop-20261008T210228Z.md`, SHA256 `d3868b6a0a81826407ade426c6f8a2c6511401177d3801eb24b3ae0e9023aa43`, lines 16–21. The six criteria cover population witnesses; row evidence with conservative PENDING dispositions; the mined-list census; validation of the full 368-star/45-field/2-document asset; both-family reads of 113 action rows plus seeded 59-row samples per stratum with zero defects; and a cued release of the hash-pinned asset. These fields record requirements rather than passed checks.

### Evidence-class table

| Claim | Evidence class | Command / receipt |
| --- | --- | --- |
| G6a is MET, #881 merged at `2026-10-08T20:41:37Z` as `e48de2e6fa1dd3b740af4793ea5824ea35476dba`, and F9/fresh-session checks are retained | source_review | F9 receipt at the SHA256 above; bounded `gh pr view 881 --json number,title,state,isDraft,mergedAt,mergeCommit,url` |
| G6b/G6c continue, with #885/#886 covering records | source_review | F9 receipt line 17; bounded GitHub PR metadata |
| G5 has six required closure conditions and remains the open START gate | source_review | G5 task record at the SHA256 above, lines 12 and 16–21 |
| The manifest reproduces the declared sources and joins each new gate field to exactly one retained selector | local_integration | Builder `--write`, `--check`, and focused gate/receipt integrity assertions, rc 0 |
| Adapter and page-publication regressions pass | synthetic | `tests.test_north_star_readiness`, 32 fixture tests, rc 0 |
| Repository publication integrity and scope pass | local_integration | `python3 scripts/validate.py`, rc 0; 70 components, 10,900 hashed files, 4 profiles, 239 receipts |

### Local commands run

```sh
nice -n 10 ionice -c2 -n7 python3 -I tools/north-star/build_readiness.py --write
# rc 0; 8 gates, 20 layers, 36 SDK items
nice -n 10 ionice -c2 -n7 python3 -I tools/north-star/build_readiness.py --check
# rc 0; manifest SHA256 0e2dba792fff6d4adbc13b47995474591f1024e8253bd7e5caf3a5c4db619aae
timeout 600 nice -n 10 ionice -c2 -n7 python3 -m unittest tests.test_north_star_readiness
# rc 0; 32 tests, OK
nice -n 10 ionice -c2 -n7 python3 scripts/validate.py
# rc 0; integrity and scope passed
git diff --check
# rc 0
```

Durable logs under `research/fullspeed-20261008/ns-readiness-manifest/`:

- `g6-split-targeted-tests.log`, SHA256 `fcee3223b60efc73e55954ea5d703f2b75952d3a6b8d9965e7127379f70b03cc`.
- `g6-split-publication-validate.log`, SHA256 `9106406d71891d803b9ed1569f54c5df6195c29846e57e0f1c284e993ce92652`.

### Decision record

This is the accepted gate-receipt refresh requested by task `task-ns2604-coop-20261008T212726Z`. The two named immutable source records above establish its scope; source selectors retain full digests and exact line locators. A replacement adjudication receipt would supersede these values through the same supported refresh procedure.

### Host evidence

This PR changes the readiness view and its selectors. Host-evidence receipt requirements are not applicable to its owned paths.

### Checklist

- [x] New/changed GitHub Actions are pinned to a full commit SHA with a version comment (no floating tags). Not applicable to this data refresh.
- [x] New/changed workflows declare top-level `permissions: {}` and grant each job only what it needs. Not applicable to this data refresh.
- [x] No secrets are printed, logged or committed; no new required secret was added without a documented owner.
- [x] No new paid hosting, subscription or billing surface was introduced.
- [x] Peer-owned untracked files and worktrees were preserved (not deleted, moved or overwritten).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

lane:shared Touches files owned by both lanes; needs both lanes' acknowledgement

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant