Coverage thresholds — where the ratio and the floor come from

Issue #235: a coverage claim (“312 URLs, all covered”) means nothing without an independent denominator to be a ratio against, and a ratio needs a published bar, not a constant buried in a script — the same argument #70 already made for klink/nosepass/orbeetle. This is that page for the migration lane’s coverage rows.

Provenance. Seeded verbatim from nkp-studio’s packages/capture/src/gates/completeness.ts DEFAULT_COMPLETENESS_CONFIG — a completeness gate built on a denominator, tuned from real client captures rather than from taste. Two are wired into a swanna row today; the rest are published for the same reason the two are — a number anyone can question needs to be a number anyone can read — and are candidates for a future row rather than a silent gap.

Setting	Value	Gated by	Enforced today?
`minPageRatio`	0.95	`coverage-ratio` — captured URL count (ariados) ÷ expected total (yamask’s WordPress REST denominator) must be at or above this	Yes
`thinWordFloor`	50	`no-thin-pages` — a target page’s body word count, absolute, independent of the source	Yes
`maxThinRatio`	0.30	the fraction of pages tolerated below `thinWordFloor` before the book (not just the one page) is worth flagging	Not yet — `no-thin-pages` grades per page today, not the book-wide fraction
`maxBodyTierRatio`	0.50	the fraction of pages whose content came from a last-resort extraction tier	Not yet — no row reads an extraction-tier signal
`minImageResolveRatio`	0.90	the fraction of images referenced in body content that resolve in the archive/inventory	Not yet
`maxDanglingLinkRatio`	0.10	the fraction of internal links resolving to a captured slug	Not yet
`maxDuplicateTitleRatio`	0.15	the fraction of pages sharing a duplicate or templated title	Not yet

The denominator itself

denominator-established (yamask, capture group) sums X-WP-Total from every viewable, REST-exposed WordPress post type (/wp-json/wp/v2/types, then /wp-json/wp/v2/<rest_base>?per_page=1 per type) — the database’s own count, available only while the source still exists. “Viewable” is WordPress’s own metadata (is_post_type_viewable(), mirrored in the REST response), not a naming guess: it is what makes a WooCommerce product type count toward the total on a store while nav_menu_item/revision/other internal plumbing never does, without per-client configuration.

A section deliberately out of scope for a given migration (--exclude-type "<slug>: <reason>" on yamask) is removed from the denominator entirely and named on the sheet — a capped section is not a coverage failure, and the exclusion is a recorded decision, never a silent shrink of the bar.

No denominator available (REST disabled, the source not WordPress) is reported as SKIPPED with the reason, on both denominator-established and coverage-ratio — never a silent PASS on a number nobody checked.

Reading the ratio/floor together

A page thin on both sides passes the ratio-based ariados/doduo comparisons wholesale (38 words next to a source that was already 40 is a 5% drop, far under any comparison tolerance) — no-thin-pages is the one row that grades a page’s substance in absolute terms, independent of whatever the source looked like. coverage-ratio answers a different question at the URL-set level: not “is each known page still good” but “do we know about every page at all.”