Repository navigation
Causal review output records incorrectly label clean reference validation as a no-op #1469
Description
During batch42 validation, Fucoidan produced 14 reference-CLI findings, exposing a wrong interpretation propagated through recent causal review records and PR self-review text.
The installed linkml-reference-validator 0.1.7 CLI prints Total checks: {len(all_results)}, where all_results contains validation issues. Thus a clean run's zero is not a count of executed checks and does not establish a no-op. Existing tests/test_reference_validator_actually_validates.py already documents this and tests real rejection, real acceptance and EvidenceItem field discovery. Closed #257/#466 concern the original diagnosis; this issue concerns its recurrence in newly generated review outputs, not reopening the validator itself.
Acceptance: add a prominent, source-backed correction to the review-output entry point, retain historical raw outputs without rewriting run history, inventory candidate historical misleading statements using an ignored/hidden-inclusive search, correct batch42 reporting and add a regression. Re-run the existing canaries, preserve actual failed output, and publish a correction notice on affected review PRs. Do not label every candidate regex match erroneous: some correctly negate the no-op interpretation.
Fucoidan's 14 findings reproduce unchanged on the pre-edit batch41 head and concern taxonomy snippets present in the separate supplement sidecar, not the main article cache. Retain this as a reference-CLI failure/known gap rather than a clean pass; do not copy supplemental text into the article cache to silence it. The original spreadsheet has not been independently inspected. Existing #476 SupportingReference field-discovery and #1362 selected-.txt-excerpt trust gaps remain open.
Dedup: 783 all-state issue bodies plus the four just-created batch42 findings searched, with exact counter/validator and supplement terms; relevant #257/#466/#476/#783/#785/#827 bodies read. Ignored and hidden local report/test files included. Comments and PR reviews not exhaustively searched. Source-based self-adversarial review, not independent approval.
Activity
realmarcin commentedon Oct 6, 2026
Correction addressed in #1470 at 22b5b12. Exact-head source-based self-adversarial COMMENT: #1470 (review). 440 passed in 203.92s (0:03:23); post-commit docs-current passed. Fucoidan retains 14 unchanged reference-CLI findings, not a clean pass. Close only after protected main merge.
realmarcin commentedon Oct 6, 2026
Correction artifacts and a regression are published in #1470. Appended visible correction notices to all 11 affected batch31-41 PRs; historical outputs and biological records were not rewritten. Issue remains open until the protected main merge.
realmarcin commentedon Oct 15, 2026
Confirmed a recurrence in the batch132 structured review:
checks[reference-cli]was classified failed solely because the CLI printedTotal checks: 0. A fresh real-CLI probe of Phormidium, with source-network lookup forbidden, traces 21 actual EvidenceItem excerpts: clean control exits 0, an invented graph excerpt exits 1, restoration exits 0. The two Discussion excerpts remain outside native traversal (#476), which is a separate real limitation. A focused branch will clarify the explained-command label, add CLI-level canaries, and save append-only corrections to the batch130/132 review wording without rewriting historical output or scientific claims. This reuses #1469; it does not reopen #466 or close supplementary routing #1996.realmarcin commentedon Oct 15, 2026
The bounded batch130/132 recurrence is repaired in PR #2011 at 041d6a0. Saved before/after YAML+Markdown reviews preserve one explicit lineage; 48 focused tests pass, lint/typing pass, and 26 structured bundles validate against the pre-PR base. All 54 declared input-hash occurrences verify in a separately fetched object store using retained provenance tags. Correction notices were appended to PRs #2006 and #2010. The explained command now labels issue counts accurately; native Discussion and supplement-only gaps remain open. Protected CI/merge queue is pending. This scopes the new recurrence only and does not claim every historical publication follow-up in this issue is complete.
realmarcin commentedon Oct 15, 2026
Historical acceptance audit: the original batch42 correction receipt, prominent index notice and regression reached main via merged PR #1975 (42959be), even though standalone #1470 was closed without merging. I read the actual correction notices on all 11 scoped batch31-41 PRs: #1393, #1401, #1409, #1416, #1423, #1430, #1437, #1444, #1451, #1458 and #1464. They explicitly withdraw the no-op interpretation and retain #476 and the separate Fucoidan failure boundary. The original receipt and its historical file hashes remain covered by the passing batch42 regression in the 48-test run. The newer batch130/132 recurrence and their PR notices are addressed by #2011. This completes the historical-publication audit deferred in the earlier update; close this reporting issue after #2011 passes the protected merge queue, without closing #476 or #1996.
realmarcin commentedon Oct 15, 2026
Completed after protected merge of #2011 as 737a95e. All seven PR checks and all seven exact merge-queue checks passed; both full-suite runs report 4936 passed, 86 skipped and 8 deselected. The reviewed and merged trees match exactly. The original batch42 repair reached main via #1975, all 11 historical correction notices were verified, and #2006/#2010 received explicit notices for the new recurrence. Append-only before/after reviews, real-CLI mutation/restoration controls, clarified explained output and retained historical failures complete this reporting issue. This does not close Discussion traversal #476, supplementary routing #1996, or outstanding scientific graph research.