Filed by the os-dev on #15599, session session_012zGPuVVX3deAx9LdjK8jCk, from PR #15770. Filed bare — domain:*, type and priority are triage's. ⛔ Not fixed here: that card's ruling is header-only on one gate.
⚠️ Attribution is written as prose above rather than as the usual footer: GitHub's body sanitizer dropped the rule line and the footer block from this card on create (measured — the rest of the body stored byte-for-byte). Not re-posted, per AGENTS.md.
What was measured
On origin/main 66e68adc6:
scripts/**/*.mjs comment lines carrying a cross-file path:NNN citation: 27 citations across 14 files.- Nothing in this repo resolves any of them.
check-adr-symbol-anchors.mjs registers exactly one corpus — defineCorpus({ id: 'adr', docRoots: [ADR_DIR], docPattern: /\.md$/ }) with ADR_DIR = 'docs/adr' — and git grep -n defineCorpus scripts/ finds no second registration. So scripts/** is outside every symbol-anchor corpus, in both directions: a line citation there is neither resolved nor refused.
Why that matters, demonstrated rather than argued
#15599 is the worked example, and its rot rate is the finding:
|
|
| the citation as written |
packages/client/src/index.ts:310 |
| the #15094 census found the declaration at |
:392 |
| triage read, the same day |
:390 |
One cross-file line citation in a gate header moved twice inside a single day, unnoticed, because nothing checks it. A second one three lines below it (content/docs/api/client-sdk.mdx:659) had drifted off its line too, and was found only because a human read the paragraph.
The repo already ruled on the general shape. scripts/symbol-anchors.mjs#ANCHOR_GRAMMAR's last line reads, verbatim:
⛔ `DIR/FILE.ts:4901` A LINE NUMBER IS NOT AN ANCHOR FORM.
(angle-bracket placeholders spelled as words here — GitHub's body sanitizer eats them.)
That grammar is authoritative for docs/adr/** and enforced there; gate headers in scripts/** are simply outside its reach, and gate headers are exactly where the next fixer starts reading.
What a fix might owe
symbol-anchors.mjs's whole design is that a second corpus is a defineCorpus call and no new mechanism — #13788 already added one that way for the platform checklist. A scripts/** corpus over .mjs comment blocks looks like the same call, plus a migration of the 27 citations to symbol or file-level anchors.
⚠️ Unpriced here, and the reason this is filed rather than done: the extractor was written for prose in .md, and whether it reads a // or /* */ comment block cleanly is not established by this card. That is the measurement a fix owes first.
Refs
#15599 (the two repaired instances), #13788 (the second-corpus precedent), #13556 (the ADR migration and its 72.1% one-way lower bound).
Generated by Claude Code
Filed by the
os-devon #15599, sessionsession_012zGPuVVX3deAx9LdjK8jCk, from PR #15770. Filed bare —domain:*, type and priority are triage's. ⛔ Not fixed here: that card's ruling is header-only on one gate.What was measured
On
origin/main66e68adc6:scripts/**/*.mjscomment lines carrying a cross-filepath:NNNcitation: 27 citations across 14 files.check-adr-symbol-anchors.mjsregisters exactly one corpus —defineCorpus({ id: 'adr', docRoots: [ADR_DIR], docPattern: /\.md$/ })withADR_DIR = 'docs/adr'— andgit grep -n defineCorpus scripts/finds no second registration. Soscripts/**is outside every symbol-anchor corpus, in both directions: a line citation there is neither resolved nor refused.Why that matters, demonstrated rather than argued
#15599 is the worked example, and its rot rate is the finding:
packages/client/src/index.ts:310:392:390One cross-file line citation in a gate header moved twice inside a single day, unnoticed, because nothing checks it. A second one three lines below it (
content/docs/api/client-sdk.mdx:659) had drifted off its line too, and was found only because a human read the paragraph.The repo already ruled on the general shape.
scripts/symbol-anchors.mjs#ANCHOR_GRAMMAR's last line reads, verbatim:(angle-bracket placeholders spelled as words here — GitHub's body sanitizer eats them.)
That grammar is authoritative for
docs/adr/**and enforced there; gate headers inscripts/**are simply outside its reach, and gate headers are exactly where the next fixer starts reading.What a fix might owe
symbol-anchors.mjs's whole design is that a second corpus is adefineCorpuscall and no new mechanism — #13788 already added one that way for the platform checklist. Ascripts/**corpus over.mjscomment blocks looks like the same call, plus a migration of the 27 citations to symbol or file-level anchors..md, and whether it reads a//or/* */comment block cleanly is not established by this card. That is the measurement a fix owes first.Refs
#15599 (the two repaired instances), #13788 (the second-corpus precedent), #13556 (the ADR migration and its 72.1% one-way lower bound).
Generated by Claude Code