Versioned scan: returns PublicReportV1 (stable allowlist, JSON Schema at /api/v1/schema)

Observer build, vantage, rule version, observed and stored times, sealed manifest

combined {all_rows_fixed, all_findings_cleared, both} and next_letter {target, reachable, ...} are recomputed, never summed. What each decision moved. rows[]: every failing scored row with points = the score recomputed with that row passing minus the score now (same formula, Reach + 15 ceiling included), points_section_average, and why a row moves nothing; withheld (rows_withheld) where the fix list is. Findings move no points: each finding's score_effect states the letter cap it sets, whether it binds, and the letter without it.

Counts and ratios behind the reading: sections_scored, identities_answered, fetches_kept, stage_failures, non_observations, vantages. No composite number, because the parts do not combine into one honestly

Each stated fact with two times: when CrawlCheck observed it (observed_from, observed_last, removed_at) and when the site says it became true (valid time, null when the site does not say)

Each value carries locators[] (one per source): bytes [start, end) in the page this scan received, line, col, scope (meta[og:title], script[ld+json]#0, title, page text) and match (value | property | normalised | partial), or {found:false, why}; check them with /api/locate. Each business fact (name, phone, address, email, founder) with every value the site stated and where it stated it, and whether the statements agree