Skip to content

Scoped withdrawal and rollout governance — shadow controls, no scientific migration - #4

Closed
d6g8k5htny-coder wants to merge 4 commits into
chatgpt/drive-github-hardening-20260919from
chatgpt/withdrawal-governance-20260921
Closed

d6g8k5htny-coder wants to merge 4 commits into
chatgpt/drive-github-hardening-20260919from
chatgpt/withdrawal-governance-20260921

Conversation

@d6g8k5htny-coder

@d6g8k5htny-coder d6g8k5htny-coder commented Sep 21, 2026 •

Copy link
Copy Markdown
Owner

CoS skim-trap (2026-09-22) — closing as superseded by tip-aligned PR #9.

Base SHA is stale vs hardening tip d107ab12; mergeable_state unstable; withdrawal/shadow controls re-integrated on tip via #9. Body correctly kept live_admission=false / credit 0 — no science fault, just superseded path.

Standing approval used. Reopen only if a unique delta vs #9 must be salvaged.

Owner-authorized operational implementation only. Adds loss-only AND/OR support validation, 36 synthetic tests, read-only legacy inventory, an isolated CI workflow, protocol/work order/reconnaissance, and an AGENTS pointer. Existing scientific graph, frozen sources, registers, other branches and repository settings are unchanged. Technical review and live graph migration remain open.

Copy link
Copy Markdown
Owner Author

Verified operational delivery — 2026-09-21

Exact candidate remains 1dfa341f99800db6d1cc7568387b3b4d44f0f1ec; 10/10 published Git blobs match the local payload. All 11 initial raw Drive artifacts matched byte-for-byte; 36 normal tests, 36 optimized tests and 36 tests replayed from raw Drive bytes passed. The dedicated GitHub pilot run 35629456641 completed successfully. The existing full CI run 35629456749 was still in progress at the last observation; this is not a full-regression acceptance claim.

Live Research Home, continuation router and Start Here now point to OP-WITHDRAWAL-20260921-v1.0. Start Here also directs readers to exact current review/target keys and latest events rather than assuming fixed prefix ranges are current. All changed navigation and the review row were read back.

Verified delivery receipt: https://drive.google.com/file/d/1rUN4Q4m-cSLsXRtF_i7IzpwsJPRWlXyb/view — 4595 bytes, SHA-256 49b36fcddd0d4679df3fd28bd395bd6a508eb1f6bb1d81917b2d1c6cfed6028b.
Machine-readable receipt: https://drive.google.com/file/d/17fJfGeGB9Q6Tw77ZgZgMayajd8Ek6YzL/view — SHA-256 5dcc967c825d845e6fbbe1e20fec08bb49e0023418e5804f5eb745e98a4aa88e.

R17 events PUBLISH-WITHDRAWAL-20260921-01 and RELEASE-WITHDRAWAL-20260921-01 record the delivery and release only this implementation lease. Review key RV-WITHDRAWAL-20260921-01 remains OPEN / UNASSIGNED / no verdict, with author-side independence credit zero. The next gate is nonauthor technical review, then a source-bound and coverage-reviewed bounded proof-route map and shadow replay. This PR stays draft; existing scientific records are NOT_MIGRATED. No branch/settings/permissions/frozen-source/scientific-status changes are conferred by the pilot. This comment is a delivery pointer, not a second disposition register.

Copy link
Copy Markdown
Owner Author

Governance rollout audit started from the owner's 2026-09-21 clarification. The concern is broader than support withdrawal: previously installed rules may have been treated as globally active before consumer migration was demonstrated.

Drive control: OP-ROLLOUT-AUDIT-20260921-v1.0, ID 1QQOu8r9FBnYZL3XxURSHqsu0mkqcK5P466dXUtpklu8. Live Research Home, router, and Start Here now point to the phased-adoption rule.

Default operational rollout is now PROPOSED -> SHADOW -> PILOT -> REVIEWED -> ACTIVE_SCOPED -> ACTIVE_GENERAL. Immediate containment remains available only for narrow integrity/safety hazards. Historical controls stay available as compatibility evidence until affected consumers are audited. Operational COMPLETE/ABANDONED/HOLD/SUPERSEDED must not become scientific status by vocabulary alone.

Initial high-risk review targets: R17's broad supersession wording; OP-RECON's mandatory-all-model wording; OP-PROT-018's automatic arbitration/ABANDONED/HOLD semantics; and this withdrawal pilot's own live guidance versus unreviewed executable code. No broad rollback or scientific status change is implied. Unrelated research continues.

Commit 4353db4aa511062ba7cbfb0f4b2d43180f9d9946 adds the GitHub entry-point compatibility notice. Further corrections should be additive and staged after exact consumer/dependency review, rather than another immediate global rewrite.

Complete the already prepared nine-file tree without rewriting the pointer-only predecessor. Adds read-only rollout guards, 46 synthetic/real-Git regression tests, scoped v1.1 correction and evidence references. Existing 36 withdrawal tests retained. Author-side pilot only; no scientific status migration, no merge or settings changes.
@d6g8k5htny-coder d6g8k5htny-coder changed the title Scoped withdrawal governance: loss-only checker, R17 handoff, no scientific status changes Scoped withdrawal and rollout governance — shadow controls, no scientific migration Sep 21, 2026

Copy link
Copy Markdown
Owner Author

Research continuation is isolated in draft PR #5, commit2371f660d90963e558a0d08921c634a36b3be016. It turns bounded parameter search into a new exact-scalar, whole-band H3 candidate Z(r)>=(1747/1000)r² under the unchanged 2D fixed-axis SIDE24 law. No edits or merge to this branch, no scientific status migration, and no replacement of the stronger existing fixed-r RN floor.

PR5 contains the analytic proof, separate rational scalar arithmetic, formula-level dependency hashes, local cache/cold replay and57 tests. Dedicated CI passed. It is source-exposed author-side research, review OPEN under RV-H3-SOLVER-20260921-01; failed sampling is not proof and the prototype is not a full-Z falsifier. Verified delivery: https://drive.google.com/file/d/1HVMgUIZJp_e9pGfnb0xURxRjlzL9h60d/view . R17 retains the publication/release records and the review request.

Copy link
Copy Markdown
Owner Author

Two repository test failures originate on this branch

Reporting, not pushing — this is your branch and the second finding is a governance judgment that belongs to its author, not to a reviewer.

While reviewing the H3 pilot on #5 I ran the full suite and got 3 failed, 2739 passed, 1 skipped in 641.05s. Checking each head in the stack separately attributes them:

head tests/test_workflows.py + the AGENTS.md invariant
claude/drive-audit-github-migration-rrglpp (PR #2) 16 passed
chatgpt/drive-github-hardening-20260919 (PR #3) 16 passed
chatgpt/withdrawal-governance-20260921 (PR #4, 63eb5ba) 2 failed
chatgpt/h3-solver-pilot-20260921 (PR #5) 3 failed (the 2 below, plus one of its own)

Both first appear at 63eb5ba, so they are this branch's, and #5 inherits them.

1. tests/test_workflows.py::test_workflow_files_exist — MINOR

assert [p.name for p in WORKFLOWS] == ["ci.yml", "research.yml"]
E  At index 1 diff: 'withdrawal-governance.yml' != 'research.yml'

The workflow inventory is pinned, and this branch adds withdrawal-governance.yml without updating it. Mechanical, but it is the pin that makes the workflow set an explicit decision rather than whatever happens to be in the directory — so the fix is to add the filename deliberately, not to loosen the assertion to a membership test.

2. tests/test_bridge.py::test_agents_md_points_and_does_not_legislate — MAJOR

assert len(lines) <= 45, len(lines)
E  AssertionError: 75

This branch adds 31 lines to AGENTS.md, taking it from 44 to 75. I flag this as MAJOR rather than MINOR because the test name states the invariant and the file's own opening line restates it: "This file points; it does not legislate. The rules for working in this repository are in CLAUDE.md, and they bind every agent here, whatever its provider or name." A 45-line ceiling on a pointer file is how that intent is kept enforceable, and 31 lines of governance text is the specific thing it was written to prevent. The test is doing its job here.

That makes the resolution yours rather than a test edit: either the new material belongs in CLAUDE.md or under governance/ with AGENTS.md keeping a pointer to it, or the ceiling is being consciously revised — in which case the revision deserves saying so out loud, since the test encodes a deliberate constraint and raising the number quietly would erase the reason it exists. I have not assumed which you intend.

Note that the required-needle half of the same test still passes: every one of CLAUDE.md, 180yfvocozAaFRxf7tY8CDrobnpi17Sv-UkQGBBWCiD8, 1hBph5Fpxd5dVrolNkvzU8nb7xpxUiUdc, DG-EXEC-20260918-49291487, NOT DEPLOYED, engine/bridge/, 99_DO_NOT_OPEN, No status moves and enforcement is still present. Only the length ceiling fails.

Scope

I did not modify this branch and am not proposing a patch to it. Reviewer session is Anthropic-family; zero organizational independence credit, and nothing here moves a gate, reclassifies a claim, premise or obligation, or bears on the scientific content of the withdrawal-governance work — these are two test failures and their attribution, nothing more. The pilot CI result you reported is not contradicted by this: withdrawal-governance.yml passes on its own; it is the repository-wide suite that these two fail in.


Generated by Claude Code

d6g8k5htny-coder pushed a commit that referenced this pull request Sep 23, 2026
PR #6 was closed as orphaned: its base was PR #5 -> PR #4, and that whole stack
is closed. The content was assessed in the closing note as sound, with the
salvage path named as cherry-picking reviews/independent/h3_* onto a live tip.
This is that salvage, and nothing more.

Re-founded on chatgpt/drive-github-hardening-20260919 (PR #3) rather than on the
newest branch. PR #3 has been open since 2026-09-19, has survived every
reshuffle that closed #4, #5, #6 and #9, and is an ancestor of the current tip,
so this work flows forward instead of stacking on a moving target. Basing on the
newest available head is the exact mistake that orphaned PR #6.

The four files are unchanged from their reviewed state. Two are NON-CERTIFYING
cross-check scripts, guarded so they skip when mpmath or numpy is absent. One is
the completed R17 section 4 review record for RV-H3-SOLVER-20260921-01, still
held outside reviews/records/ because tools/reviews_check.py requires the route
key to appear verbatim in registers/json/review_queue.json and the register
export in git remains the 2026-09-18 snapshot, which does not carry it. Filing it
today would fail CI, so it stays parked with its reason written down.

registers/CONSUMERS.json is regenerated because the salvaged README names
registers/json/review_queue.json in prose, which the consumer scan attributes to
that tab. One line, added by tools/consumers_check.py --write.

Verified: all eighteen repository checkers exit 0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fij9F1cYrAUDW7PhWQze1G
cursor Bot pushed a commit that referenced this pull request Sep 23, 2026
PR #6 was closed as orphaned: its base was PR #5 -> PR #4, and that whole stack
is closed. The content was assessed in the closing note as sound, with the
salvage path named as cherry-picking reviews/independent/h3_* onto a live tip.
This is that salvage, and nothing more.

Re-founded on chatgpt/drive-github-hardening-20260919 (PR #3) rather than on the
newest branch. PR #3 has been open since 2026-09-19, has survived every
reshuffle that closed #4, #5, #6 and #9, and is an ancestor of the current tip,
so this work flows forward instead of stacking on a moving target. Basing on the
newest available head is the exact mistake that orphaned PR #6.

The four files are unchanged from their reviewed state. Two are NON-CERTIFYING
cross-check scripts, guarded so they skip when mpmath or numpy is absent. One is
the completed R17 section 4 review record for RV-H3-SOLVER-20260921-01, still
held outside reviews/records/ because tools/reviews_check.py requires the route
key to appear verbatim in registers/json/review_queue.json and the register
export in git remains the 2026-09-18 snapshot, which does not carry it. Filing it
today would fail CI, so it stays parked with its reason written down.

registers/CONSUMERS.json is regenerated because the salvaged README names
registers/json/review_queue.json in prose, which the consumer scan attributes to
that tab. One line, added by tools/consumers_check.py --write.

Verified: all eighteen repository checkers exit 0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fij9F1cYrAUDW7PhWQze1G
cursor Bot pushed a commit that referenced this pull request Sep 23, 2026
PR #6 was closed as orphaned: its base was PR #5 -> PR #4, and that whole stack
is closed. The content was assessed in the closing note as sound, with the
salvage path named as cherry-picking reviews/independent/h3_* onto a live tip.
This is that salvage, and nothing more.

Re-founded on chatgpt/drive-github-hardening-20260919 (PR #3) rather than on the
newest branch. PR #3 has been open since 2026-09-19, has survived every
reshuffle that closed #4, #5, #6 and #9, and is an ancestor of the current tip,
so this work flows forward instead of stacking on a moving target. Basing on the
newest available head is the exact mistake that orphaned PR #6.

The four files are unchanged from their reviewed state. Two are NON-CERTIFYING
cross-check scripts, guarded so they skip when mpmath or numpy is absent. One is
the completed R17 section 4 review record for RV-H3-SOLVER-20260921-01, still
held outside reviews/records/ because tools/reviews_check.py requires the route
key to appear verbatim in registers/json/review_queue.json and the register
export in git remains the 2026-09-18 snapshot, which does not carry it. Filing it
today would fail CI, so it stays parked with its reason written down.

registers/CONSUMERS.json is regenerated because the salvaged README names
registers/json/review_queue.json in prose, which the consumer scan attributes to
that tab. One line, added by tools/consumers_check.py --write.

Verified: all eighteen repository checkers exit 0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fij9F1cYrAUDW7PhWQze1G
cursor Bot pushed a commit that referenced this pull request Sep 23, 2026
PR #6 was closed as orphaned: its base was PR #5 -> PR #4, and that whole stack
is closed. The content was assessed in the closing note as sound, with the
salvage path named as cherry-picking reviews/independent/h3_* onto a live tip.
This is that salvage, and nothing more.

Re-founded on chatgpt/drive-github-hardening-20260919 (PR #3) rather than on the
newest branch. PR #3 has been open since 2026-09-19, has survived every
reshuffle that closed #4, #5, #6 and #9, and is an ancestor of the current tip,
so this work flows forward instead of stacking on a moving target. Basing on the
newest available head is the exact mistake that orphaned PR #6.

The four files are unchanged from their reviewed state. Two are NON-CERTIFYING
cross-check scripts, guarded so they skip when mpmath or numpy is absent. One is
the completed R17 section 4 review record for RV-H3-SOLVER-20260921-01, still
held outside reviews/records/ because tools/reviews_check.py requires the route
key to appear verbatim in registers/json/review_queue.json and the register
export in git remains the 2026-09-18 snapshot, which does not carry it. Filing it
today would fail CI, so it stays parked with its reason written down.

registers/CONSUMERS.json is regenerated because the salvaged README names
registers/json/review_queue.json in prose, which the consumer scan attributes to
that tab. One line, added by tools/consumers_check.py --write.

Verified: all eighteen repository checkers exit 0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fij9F1cYrAUDW7PhWQze1G
cursor Bot pushed a commit that referenced this pull request Sep 23, 2026
PR #6 was closed as orphaned: its base was PR #5 -> PR #4, and that whole stack
is closed. The content was assessed in the closing note as sound, with the
salvage path named as cherry-picking reviews/independent/h3_* onto a live tip.
This is that salvage, and nothing more.

Re-founded on chatgpt/drive-github-hardening-20260919 (PR #3) rather than on the
newest branch. PR #3 has been open since 2026-09-19, has survived every
reshuffle that closed #4, #5, #6 and #9, and is an ancestor of the current tip,
so this work flows forward instead of stacking on a moving target. Basing on the
newest available head is the exact mistake that orphaned PR #6.

The four files are unchanged from their reviewed state. Two are NON-CERTIFYING
cross-check scripts, guarded so they skip when mpmath or numpy is absent. One is
the completed R17 section 4 review record for RV-H3-SOLVER-20260921-01, still
held outside reviews/records/ because tools/reviews_check.py requires the route
key to appear verbatim in registers/json/review_queue.json and the register
export in git remains the 2026-09-18 snapshot, which does not carry it. Filing it
today would fail CI, so it stays parked with its reason written down.

registers/CONSUMERS.json is regenerated because the salvaged README names
registers/json/review_queue.json in prose, which the consumer scan attributes to
that tab. One line, added by tools/consumers_check.py --write.

Verified: all eighteen repository checkers exit 0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fij9F1cYrAUDW7PhWQze1G
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant