docs: the correction entries answer a caller, and the record has one home (#472)
#464's third slice, grouped by subject because subject decides the destination:
`page_corrected` (156 lines), `page_correction_failed` (118), `page_correction_recheck`
(69) and `page_redrawn` (95) → 117 / 99 / 58 / 38.
design-notes already held the correction path's *verdict* side — the binding recheck, the
discard decision, `rechecks.binding_error`, `correction_discarded` — and #429's doubt
words. Neither covered what a correction DOES or what its fields can see, so this adds
`## Correcting a page against its image` between them, with four subsections, pointing at
the two existing homes rather than restating them.
What stayed in API.md is what a caller does with a log line: the firing condition, every
field and what it holds, the remedy, and the limits to know before counting a corpus off
one. What moved is the evidence and the reasoning — the alt fields' asymmetric matching,
the six-name budgets, the correction ceiling's multiple and floor, #293's two truncations
argued both ways, #288's sample read as a rate and the census that answers it, and the
redraw rule's whole corpus.
docs/API.md 5,821 -> 5,695
docs/design-notes.md 2,391 -> 2,611
Nothing deleted, and it is checked rather than read: 108 figures, every issue number, 105
of 105 identifiers still in API.md, and link targets compared by citation count.
Three review rounds, six notes. Three of round 1's four were about the DESTINATION, not
the moved text: design-notes' own index listed neither this section nor #471's, a bullet
restated a sentence API.md owns, and a corpus was widened from "every bench round" to
"every bench log". Measuring what a destination holds is not measuring what it requires.
Round 2 then caught the count my round 1 fix introduced — deleting a claim can introduce
one, because a summarising lead-in is a claim and the trimmed text had none to check it
against. Round 3 clean, no new findings.
npm test 1707/0 - tsc clean - e2e ALL ENDPOINTS PASSED - anchors 418 links / 0 bad
Refs #464
Co-Authored-By: Blake Bertuccelli-Booth <bbertucc@users.noreply.github.com>
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>