Make the table reviewable. refrakt inspect shows a rune's resolved schema row, refrakt contracts describes the structured data it currently omits, and refrakt reference documents it.
700d0159ac9575f847949SPEC-130 puts tooling on items 4–7, riding each migration. That is one step too late, and D5 is why.
D5 dispenses with a validation mechanism on the grounds that a rune only ever emits a type it has a row for. But it says plainly that "no mechanism" is not "no risk": nothing checks a table against schema.org, because refrakt ships no ontology. What replaces validation is visibility — "a wrong row is reviewable".
If the review surface arrives with the last migration, then Groups A, B and C were curated with no review surface at all. Thirty hand-written tables asserting schema.org types on users' pages, reviewed by reading transforms — which is the condition this milestone exists to end.
D2 is the proof it is not hypothetical. organization carries a curated six-value enum, enforced by Markdoc, containing NonProfit — a type schema.org does not have. The enum validates against itself, so nothing noticed. That is first-party code that had been reviewed.
refrakt inspect <rune> shows the resolved schema row — type, property mapping, nested entities, child mappings — the way it already shows BEM classes and data attributes--json carries the same, so it is machine-readableinspect --audit reports a table naming a source the rune does not emit, alongside the CSS coverage it already auditsrefrakt contracts describes each rune's schema.org output, closing the gap between what contracts claims to cover and what it doescontracts --check catches config-to-contract drift on the schema channel, as it does for structurerefrakt reference documents a rune's structured data, and can order the resolved table by schema property regardless of authoring order — the review view need not match the authoring viewdefineRune({ schemaOrgType }) is deleted or fed from the table, so the type is declared onceinspect matters most. It is the per-rune tool an author actually reaches for, and every claim in SPEC-130 was verified with it. Today it shows RDFa only incidentally, as raw HTML in the rendered output.
Delete Rune.schemaOrgType rather than leaving it. It is assigned in packages/runes/src/rune.ts:59 and consumed by nothing — not reference, not contracts, not the language server. It has already drifted: nine catalog entries against thirty emitting runes, and Accordion: 'FAQPage' stopped being unconditional when WORK-552 shipped. Leaving it in place means the second stale declaration outlives the first, which is how this milestone's problems started.
The audit is the piece with teeth. A table naming a data-name the rune does not emit is the failure mode WORK-561 exists to prevent and that every later migration can reintroduce — and unlike a wrong schema.org type, it is mechanically checkable. Ship it here, so Groups A–C land against it.
Accordion: 'FAQPage' is already stalepackages/runes/src/rune.ts:59 — the write-only schemaOrgTypepackages/cli/src/commands/ — inspect, contracts, referenceCompleted: 2026-09-16
Branch: claude/v0.35-parallel-feasibility-eia5le
packages/runes/src/schema-row.ts — describeSchemaRow, bySchemaProperty, collectSchemaRows, tableFor. Lives beside the tables because three consumers need the same answer.packages/runes/src/lib/index.ts — schemaTables, a WeakMap keyed on the Markdoc schema, alongside schemaContentModels and schemaRuneStructures.packages/cli/src/lib/schema-row.ts — auditSchemaSources, the half that needs a rendered tree.packages/cli/src/commands/inspect.ts — a Structured Data section, text and --json.packages/transform/src/contracts.ts — schemaOrg on RuneContract, and generateStructureContract(config, { schemaRows }).packages/runes/src/reference.ts — schemaOrg, ordered by schema.org property rather than authoring order.packages/runes/src/rune.ts + index.ts — defineRune({ schemaOrgType }) deleted, with its nine stale catalog entries.packages/cli/test/schema-row.test.ts — 11 tests.CLAUDE.md — the review mechanism, and D5 stated plainly.schemaTables is the point. Config-derived tooling could not see anything a rune declares in its transform() — the same gap that let four dead CSS rules survive years of review (WORK-564), that kept .rf-lore__title out of the CSS coverage set, and that made regenerating the structure contract show none of WORK-561's new data-names. Recording the table where tooling can reach it is what makes D5's "visibility replaces validation" possible at all.
rune.typeName ?? rune.name; loadPlugin never sets typeName, and the contract is keyed PascalCase, so the lowercase fallback matched nothing. It reported 0 runes with a schema row while both tables were correctly recorded — skipping exactly the 67-of-95 population D7 is about. Now joined on dataRune, the rune's own name on both sides.testimonial's ratingValue <- rating as not emitted, when rating is a declared attribute the canonical fixture simply does not set. Calling a correct table broken because the input was minimal is worse than not checking. A source now resolves against an emitted node, a field-bag entry or a declared attribute, which leaves the failure actually worth catching: a typo'd or renamed source that resolves to nothing ever. Verified both directions by renaming headline to headlinee in event's table and watching it fire.packages/lumina/test/contracts.test.ts regenerates and compares byte-for-byte, and it was right to fail: the CLI's output and the test's regeneration came from different paths. The rows now go into generateStructureContract, so the artifact and its guard are one call with one input. That is also what forced the describing half out of the CLI and into runes, where it belonged.contracts/structures.json and packages/lumina/contracts/structures.json. Both regenerated; a test guards both.schemaOrg is declared on RuneContract rather than cast in: it is genuinely part of the contract's shape now. Typed loosely on purpose — the row's shape is owned by @refrakt-md/runes, and restating it in transform would be a second definition to drift.runes depends on transform and reaching the tables would invert that.defineRune({ schemaOrgType }) broke no build, confirming the item's claim that it was write-only — nine entries against thirty emitting runes.npm test — 370 files, 4538 tests, all passing. npm run format:check clean (run on its own, exit code read directly). npm run seo:baseline:check byte-identical — this item changes no emission.