Repository navigation
Museum conditional route: all hypotheses preserved, no status promotion - #166
Conversation
One cumulative-transfer pilot transcribes all six inputs and one conditional conclusion from existing pinned proof bytes. Adds read-only shared CLI, 24 focused controls and a hosted integration replay. No new status register, no existing scientific verdict or source pin changed. Engineering review required.
|
Fresh hosted/readback evidence at 2072c09:
The hosted runner issued a non-failing compatibility warning for the reused checkout action's Node20 target being run under Node24. No CI protection was weakened and no dependency update was bundled into this view. Engineering review remains requested; no real-browser/layout/GPU test or new scientific review is claimed. This PR is still a draft, not a live Pages deployment. |
…ntity Three new negative controls reject hash-correct proof bytes relabeled with a different valid-looking commit and require complete exported identity. All 27 focused tests pass; seven weakened modules detected. Preserved old tests, source proof, config, manifest and scientific statuses. Browser smoke test blocked by administrator before assertions; no protection changes or browser success claimed.
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
Browser success-path coverage and duplicated startup verification remain unresolved; the fixture arithmetic wording also needs clarification.
Review effort: Lite
Findings: 2
Open (2)
What changed in this PR
Adds an engineering-only, source-verified conditional proof route with offline validation, tests, documentation, and CI coverage. Scientific status remains unchanged.
Changes:
- Adds browser and CLI projection with immutable source identity checks.
- Integrates the conditional route into the museum UI.
- Adds fixtures, regression tests, documentation, and workflow verification.
| File | Description |
|---|---|
tools/museum_conditionals.mjs |
Read-only conditional projection CLI |
tests/test_museum_identity.mjs |
Source identity regression tests |
tests/test_museum_conditionals.mjs |
Projection and rendering tests |
tests/test_museum_conditionals_cli.mjs |
CLI and HTML integration tests |
tests/fixtures/cumulative_transfer_source.txt |
Pinned proof fixture |
docs/site/museum.html |
Conditional route UI integration |
docs/site/conditionals.mjs |
Browser verification and projection logic |
docs/site/CONDITIONAL_ROUTES.md |
Route scope and replay documentation |
.github/workflows/museum-conditionals.yml |
Hosted verification workflow |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
|
Source-bound response to Copilot review at a66366a: both inline integration findings are confirmed and remain unresolved. The shared-source identity correction is implemented and independently regression-tested, but it does not add a successful startConditionals integration path or remove duplicate central startup verification. The appropriate repair is to share the existing verified museum startup context rather than invent a second status/cache authority, then test successful config/manifest/index/status/proof flow plus failure clearing and request reuse. This needs review against museum.mjs and its existing integration harness. Neither review thread is marked resolved, and the PR stays draft/unmerged. The byte-frozen source fixture must not be silently edited to address prose suggestions; mathematical/source corrections need a separate source-bound successor. Fresh exact-head hosted results inspected: all four PR workflows succeeded (36288398289,36288398274,36288398270,36288398271). Expanded frontend job108533526522:69 PASS,0 FAIL,2 inherited SKIP. Committed config/manifest CLI replay and cmp passed. These tests do not establish the missing successful startup coverage. Chromium localhost smoke test remains BLOCKED_BY_ADMINISTRATOR before any assertion; no browser protection changed. Scientific effect NONE. |
… its success path
- museum.mjs: extract verifyStartup and export a per-fetcher memoized
verifiedMuseum({fetcher}) returning {manifest, cachedFetch}; startMuseum
consumes it. Rejected startups are not retained.
- conditionals.mjs: consume verifiedMuseum instead of repeating the
config/manifest/index/status fetches; drop the separate core.mjs import.
- tests/fixtures/museum_fixture.mjs: shared synthetic projection helper
(fixture now accepts a commit and per-claim proof substitutions);
test_museum_frontend.mjs imports it instead of its private copy.
- tests/test_museum_conditionals_startup.mjs: mocked successful startup,
shared startup/byte cache across startMuseum + startConditionals,
non-retained rejection, altered bytes, non-audited descriptor, stale
manifest, missing container.
- CONDITIONAL_ROUTES.md: document the shared startup and the new test.
Ran: node --test tests/test_museum*.mjs tests/test_public_shop_frontend.mjs
(78 tests, 76 pass, 2 skipped fixture-gated); CLI replay --source/--check;
python unittest discover (120 OK).
Co-authored-by: Dylan Roy <d6g8k5htny@privaterelay.appleid.com>
…ional-route-20260926 Co-authored-by: Dylan Roy <d6g8k5htny@privaterelay.appleid.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
f9ae228 (the #166 amendment) committed tests/__pycache__ and tools/__pycache__ .pyc files to main by accident. Remove them from the index and add a .gitignore so they cannot return. Working-tree copies are untouched; nothing else changes. Co-authored-by: Dylan Roy <d6g8k5htny@privaterelay.appleid.com>

Current head and scope
Engineering-only conditional proof view; scientific effect NONE. Current head
a66366a6ab056f961140f60347141bf35dae63de, successor to2072c096df9ab4a47867bb48c10b0f04cbff8ef1, based on maindbb9dcf64d23ecb45267ce2779f69f17b2fa6265. Owner continuation coordinated on main#86 comment5851860129.The existing cumulative-transfer correction is projected as six complete inputs (setup/quantifiers, H1-H4, definition) feeding one ALL route to the conditional conclusion. Positive versus zero coefficient cases and the warning against inferring a density asymptotic are retained. Application remains NOT_EVALUATED. This is not a new theorem, a general theorem prover, or a second status database. Math-#89's scalar/Hessian candidate is NOT added to accepted cards.
Author-found identity defect repaired
The original offline implementation checked commit syntax, source hashes and coherent URLs, but could accept the correct proof bytes attributed to a different valid-looking commit if the local config/manifest were changed coherently. Three test-first regressions exposed the gap: all 24 existing focused tests passed while the three new checks failed.
The shared selector now requires the exact audited Math commit
d6628da09384728992dcbe6e921cc28ba85aebb0. The exported projection retains the full repository/path/commit/blob/bytes/SHA256/URL descriptor instead of only a digest. Changing source revision requires an explicitly reviewed projection update.tests/test_museum_identity.mjsexercises the module and actual offline CLI; the two original test files are unchanged. This correction is not a claim that the pre-existing browser fetched bytes from a fabricated commit: the demonstrated defect was local offline attribution.Implementation and preservation
docs/site/conditionals.mjsshares the museum's bounded fetch, manifest and proof/review checks; verifies source SHA256, Git blob, bytes and audited identity before transcription; failure clears the conditional view.tools/museum_conditionals.mjsis a read-only offline CLI using the same projection. --check rejects omissions, ALL-to-ANY changes, extra dependencies and invented status fields by exact regeneration comparison.No STATUS, PROOF_INDEX verdict, LANDING_CLAIMS, claim graph, old proof body, museum.json, config.json, pinned arithmetic or query- file is changed.
Exact source and implementation identities
The cumulative proof fixture is 3272 bytes, Git blob
044ac5fdaf403a38e33983e31f0ad69f8e76d6d5, SHA25683f653393dc6245980f6848e2bfc65ac9ad4c7d8304b028b88764356fd3825b6, from the audited Math commit above.At this new head:
50316ae818f4bc8375e784df6b453525b9c99e70351ccea8d8601fff0177bf41.db894ec40b2c3d437aaa748110181ae8c9dc37636108deaccac87070b47c3728.e8d1698415c3c83d4cdf8d83111a0649db3942ef929c7356b96884dc426842ad.All three Git blob identities were read back and match the tested local files. The successor modifies only the module/documentation and adds the separate identity test; no source fixture is rewritten.
Verification and limits
Current local Node v22.16.0: 27 tests PASS, 0 failed, including the real CLI identity regression. Seven syntactically valid weakened modules are detected, including removing the audited-commit guard. Tests and mutations run on isolated temporary files; no other repository paths were edited.
Historical head2072c09: 24 focused tests passed; initial test-first baselines were 16 projection failures plus 8 CLI/HTML failures. Its four hosted workflows succeeded. The expanded frontend job had 66 passes and 2 explicitly recorded inherited skips, and the actual committed config/manifest CLI replay passed. Those results do not assert success for this successor before its hosted runs complete.
A real Chromium smoke-test attempt was BLOCKED during localhost navigation with
net::ERR_BLOCKED_BY_ADMINISTRATOR, before any module assertion. No browser protection was changed. No real-browser layout/GPU success is claimed. Local tests use bounded metadata fixtures around exact proof bytes; the hosted workflow also exercises committed config/manifest data. No full unrelated local repository regression is claimed; whole-repository cloning is blocked by github.com DNS failure in this container.Review required
Nonauthor engineering review: exact six-input transcription, quantifiers, normalization/fixed lower comparison, positive/zero cases, immutable identity attribution in both browser and CLI, fail-closed behavior, no new status authority, and existing museum compatibility. The browser integration gap remains explicit. This PR remains draft and is not deployed.