Repository navigation
Use checked accounting in the inner-wedge RN verifier - #48
d6g8k5htny-coder merged 2 commits into
Conversation
Reuse the source-preserving adapter at the second live RN publication path. Add eight tests of the actual consumer, including exact stored-receipt reproduction, post-accept record corruption, false reported totals, adapter bypass and normal/optimized execution. Baseline refused none of the adversarial paths (7 failed, 1 passed); candidate focused scope:126 passed. Full field reconstruction and repository CI remain separate. No frozen candidate, ledger, source pin or scientific status changes. Authored-by: OpenAI / ChatGPT
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
Delivered evidence is now in Drive: https://drive.google.com/file/d/1GoI3TIoaQ1eSvEbB8XzWy-1-rakmKuRq/view —23,787 bytes, SHA256 I downloaded the ZIP again, verified all11 manifest-listed payloads, applied its patch to the original consumer in a disposable selected checkout, recovered both submitted Git blobs exactly, and reran the same126 focused tests:126 passed in18.90s. This is a delivery readback replay, not another126 distinct tests or a full Gaussian-field reconstruction. Original18.36s JUnit and the deliberate baseline7fail/1pass log remain in the immutable package. The candidate's own full Actions run35941226607 is still separate. At the last inspection it had passed the preceding register/source/RN/H3 checks and was running the actual inner-wedge replay, with full pytest pending. No full-pass or merge claim yet. No workflow or permission changed. Separate mathematical work from this continuation is tracked in issue49: a precise pinned-Gaussian genericity application reusing Claude's already existing Morse-only Borel-event theorem. It does not block this engineering PR or duplicate PR21 migration. — OpenAI / ChatGPT |
…edge-accounting-20260924
ac33581
into
chatgpt/drive-github-hardening-20260919
…ces-inventory The base gained the vault path map (#44) and the checked inner-wedge accounting (#48). No textual conflict: both sides appended to tools/run_checks.py's REQUIRED_COMMANDS and to the CI step list, and git took both. The certificate inventory is unchanged -- 8 certificates, 34 files and 6 archive members, all 40 digests still matching -- because neither merged PR touched a pinned byte. There was a semantic one. `AGENTS.md` is capped at 45 lines by `test_agents_md_points_and_does_not_legislate`, both sides added a pointer, and the merged file was 46. The base's pointer to `quarantine/PATHS.md` is the one that stays: mine is dropped, because the rule it pointed at is in `CLAUDE.md`, which AGENTS.md's first paragraph already tells every agent to read. Raising that cap is a decision about someone else's file and not one to take in a merge. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Cz7WZybv8znP64SpPj6sWY
Continuation of #45: protect the second actual consumer
tools/rn_inner_wedge_check.py:assemble_coverstill consumed rawLedger.total(). It now uses the existing source-preservingchecked_totalboundary. Two production lines change (one import and one call); eight new tests exercise that actual function rather than a duplicate toy implementation.The new controls demonstrate that the old consumer could emit a certified zero after a post-accept record became an unrecognized disposition or an OUTSIDE record with residual1,000,000; it could also accept negative recorded area or a falsified returned total. These are deliberate synthetic corruptions of its ledger, not evidence that the historical Gaussian-field certificate was wrong. The new consumer refuses them. A sentinel requires the actual checked boundary to be called. Normal and optimized Python both reject the corrupted record.
Exact preservation and execution
The positive control reads the original164570-byte candidate (SHA256b685e524cb257ac1c31f1f22f88c5bf4510bcdf7bc575d04b333a2f6613438ba), reconstructs only the consumed assembly premises, and reproduces the complete original cover receipt exactly. This is NOT a fresh reconstruction of the Gaussian/Taylor premises; the complete CLI replay remains in repository CI.
Local Python3.13.5, pytest9.0.2, optional plugin autoload disabled: 126 passed in18.36s (100 cover checks,18 retained wedge-composition checks,8 new consumer checks). The same new suite on the unmodified consumer gave 7 failed,1 passed. Both uploaded Git blobs match the tested local bytes. Snapshot inputs were SHA256/Git-blob verified; the consumer/test preimages and cover module identities match the live74c082e revision.
Base74c082e799d8d338da9adb24a968c1231842851a; candidate1842f1c68ad669b0413aa330c557294cf087eac9; tree30ff9426b7bf03c964ee9c4262a17fa74738fbcd. The base's own completed push run35937914517 has3169 passed,2 skipped; that is evidence for the base, not this candidate. Await this candidate's own full workflow before treating integration as verified.
No historical ledger, driver, geometry module, candidate, source identity, requirement, workflow, permission, canonical register or mathematical status is changed. No expansion of the wedge or closure of the annulus/event obligations is claimed. Claude retains #21's R3 migration and H3 review filing. Coordination: #45 comment5805514836.
Authored by OpenAI / ChatGPT. Author-side engineering evidence, not independent review.