Chemistry evidence-watch maintenance review — 8 November 2026

The failed watch on PR #94 passed its self-test, then reported changed=1 added=0 removed=0 and Exit 1. Only canonical Chemistry changed: its unchanged canonical-evidence-json-v1 observation hash includes requires.

This acknowledgement follows the already proved G17 batch at 3a1b0d5d7a00e531e6870e1a1bc584d323b434bf. Comparing the full Chemistry JSON with merged main 8edadbf9aad3497a9738066e7a95f27ef98ee599 confirms exactly these six prerequisite deletions at four clusters. Every other JSON value, including all contains, is unchanged.

Source cluster	Removed prerequisite
bb707fda-504c-4699-a78c-d0a6c320658f	8d4ef102-e6a6-4d2e-bb6b-e707d3f2e566
871eabde-6e19-59d3-bc7c-60977a9837db	fa203f3a-0e81-5bcb-a4c8-e4374b7f04a3
871eabde-6e19-59d3-bc7c-60977a9837db	1286f2fe-89b7-4454-8e11-85b6abd6e278
08b44b8f-e407-5a1f-82dc-e70e598022cf	1c1420c2-a8e2-520f-8015-6df637a973bd
08b44b8f-e407-5a1f-82dc-e70e598022cf	95dc0ee5-a0af-5682-af32-d66e36fbeb50
10ce2814-8796-5633-9bed-f6990d039b91	326d45bf-9f77-57d5-a054-93e76b034dd5

Merged main and tested QA head 35f01974cd reproduce the stored native hash af344ee30812…; the current source produces 5da87f60f3a8….

Only the observation date and that record's hash and file timestamp advance. The other 159 complete records and all remaining baseline metadata stay exact; the inventory remains 160 with no added or removed paths. The complete prior baseline remains in reachable merged Git history. The technical receipt binds its Git blob, full SHA256, 49,289 bytes and 160 records, the full old/current Chemistry hashes, failed CI logs, existing proof results and local 29-input freeze.

Existing Node proof, both source-Atlas checks and lint passed. The native run passed two tests without errors or skips: 1,240 real configurations, 198 live plus two canonical sources, six missing-role probes and 370,306 comparisons. The regular CQR result retains 21 roots and all ten protected maturity floors (five M7, five M6). Local checks confirmed the actual result files and all 29 frozen inputs; those suites were not rerun. Registry d7e25fe65e56… retains 204 operations, 200 original sources, all 21 historical baseline digests and historical registry prefixes unchanged; the receipt records its full hash.

The manifest's four reopening conditions remain unchanged: visible scope, source-backed projection status, runtime mapping coverage and composition-view eligibility. No source, mapping, view, learner scope, task, image, unit or mastery changes occur here. This adds zero scientific reviews or completions. Only the observation snapshot advances after the proved repair; hashing, checker, manifest and workflow remain unchanged.

All 32 temporary local rejection probes passed: wrong lineage, undeclared data/path/record changes, incomplete or failed receipt/result data, false freeze flags, native hash/coverage/JUnit failures and CQR drift were rejected. The unchanged watch also rejected title, nonvisual source-link and prerequisite drift on temporary copies. Their executed local script, full receipt and result remain in ignored scratch, with hashes in the compact receipt; this documents completed local evidence and introduces no permanent verification framework. python3 scripts/canonical_chemistry_evidence_watch.py self-test and bash scripts/run_canonical_chemistry_evidence_watch.sh pass locally with Exit 0, changed=0 added=0 removed=0 and CHEMISTRY_EVIDENCE_WATCH=OK; both views are regenerated by that normal runner. Future unacknowledged changes still fail. Fresh remote CI remains a separate check.