accordion, accordion-item, recipe and how-to set typeof on nodes they did not create and hand-write a <div property="text"> wrapper to supply the value. Replace both with a declaration.
The HTML must not change. That is the whole difficulty.
Tracking started Sep 17 — check back for trends.
The obvious cleanup is to delete the wrapper and teach collectJsonLd to take a typed node's own text as a named property. It would break the HTML.
RDFa Core 1.1 §7.5 step 11 resolves a property's object as: @content → literal; else @typeof present and @about absent → the typed resource; else a plain literal from the text. So an element carrying both property and typeof has its object fixed to the typed resource and its text is unreachable as a literal. Today's output —
<div typeof="Answer" property="acceptedAnswer"><div property="text">…</div></div>
— expresses _:q acceptedAnswer _:a . _:a a Answer . _:a text "…". Without the inner element the third triple is gone. Since SPEC-082 renders the SEO carriers inline, refrakt publishes RDFa and JSON-LD on the same page, so the two would assert different graphs on every accordion, recipe and how-to.
What is wrong is that runes hand-write it. WORK-565's text: moves the wrapping into the applier, and the table declares only the role:
accordionItem: { type: 'Question', properties: { name: 'name' }, entities: { body: { type: 'Answer', property: 'acceptedAnswer', text: 'text' }, }, }
accordion, accordion-item, recipe and how-to declare their schema as tables, with no node.attributes.typeof = … mutation and no hand-written property="text" wrapper left in any transformrecipe's <li>s resolve through their existing data-names (ingredient, step) — no new names needed hererecipe's image source resolves through the name WORK-561 gave itschema="none" still strips the whole subtree on accordion, including the wrapper the applier now emitsproperty and typeof always has an inner carrier for its text — so the wrapper cannot be "optimised away" laterWrite the byte-identical assertion first. These four runes are where a plausible-looking simplification silently changes what the page asserts, and the reasoning that leads there is short enough that someone will re-derive it. The test is the only durable form of the argument; the prose above is a comment that will not run.
accordion is also the rune that motivated the whole spec, by way of WORK-548 — rendering a list of universal attributes as accordion items publishes roughly 970 fabricated Question entries across 88 pages. WORK-552 already shipped schema="none" as the author-side out. What this item adds is that the suppression and the emission are now declared in the same place, so the relationship between them is visible.
Note accordion's comment at packages/runes/src/tags/accordion.ts:111 — "Done here rather than in a later pass because extractSeo reads the …" — is the same constraint WORK-565 is built around. Update it rather than deleting it; it is the only place in the source that records why this cannot live in the engine.
Question entriesschema="none", shippedpackages/runes/src/tags/accordion.ts:111 — the comment to carry forwardCompleted: 2026-09-16
Branch: claude/v0.35-parallel-feasibility-eia5le
accordion, accordion-item, how-to and recipe declare tables. No transform sets typeof on a node it did not create, and none hand-writes an RDFa carrier: text: moves the wrapping into the applier and the rune declares only the role.
The HTML does not move, which was the whole difficulty. Captured before and after through refrakt inspect for all four runes: identical byte count, same elements, same attributes, same values, same document positions. Eight elements serialise their attributes in a different order (typeof now follows data-name; a rebuilt meta's property precedes its content), because the applier sets them after createComponentRenderable rather than before. Attribute order is not part of an HTML document's meaning, so "byte-identical" is the one claim in that criterion I cannot make literally — stated rather than glossed.
The JSON-LD diff is empty. All 48 baseline fixtures unchanged.
textTag is new on a row: how-to and recipe wrap an <li>'s content in a <p>, and a <div> there would change the page's margins. A rendering choice, declared so the applier reproduces what each rune already rendered.
Two defects in the applier this surfaced, both from recipe:
stamp took only the first node bearing a name, so recipe published one of its six ingredients and silently dropped the rest. It now stamps every match — which is also what makes recipeIngredient an array, as the baseline has it.character maps name, and so does every character-section inside it, so a character's name became ["Veshra", "Backstory", "Abilities"]. findAllByName now stops at another rune's node — ADR-008's flat namespace is unique per rune, and the lookup has to respect the same boundary. Caught by the baseline, which is what it is for.The RDFa argument is now a test. rdfa-text-wrapper.test.ts (split across packages/runes and plugins/learning, since the two plugin runes only resolve with the plugin loaded) asserts that any element carrying both property and typeof has an inner carrier for its text, with a hand-built counter-example so the check cannot pass vacuously, and recipe's untyped ingredients as the control — an untyped property node is not wrapped, which is what makes it a rule rather than "wrap everything". schema="none" is covered too: it must reach the wrapper the applier now emits, not only the types.
lists: on these four, deliberately. D6 is defensible for mainEntity, step, tool, recipeIngredient and recipeInstructions, but this item's entire evidence is an empty diff, and declaring a list changes a one-item collection from an object to an array. Filed as WORK-574 so it is a deliberate change rather than a side effect of a refactor.accordion's comment about why this cannot live in the engine was carried forward into the table's own documentation rather than deleted, as the item asked.imperative-schema-migration.test.ts's PENDING list is empty — SPEC-130's closing condition, mechanically checked.