Why it is only the rule
SPEC-133 is a draft whose own last open question is "is this worth doing at all?". Its sweep changes HTML across many runes and breaks the component-override interface for the ones whose content markers become slots. None of that belongs in front of an accepted spec that fixes live defects.
But the sweep and the rule are separable, and the rule is free. Stating it costs a docs edit; not stating it costs WORK-561, WORK-568 and WORK-570 each inventing their own answer.
Scope
The rule, as SPEC-133 states it:
refs — any node that survives into the output and should be addressable. Gets data-name and a BEM element class.properties — values only. The value reaches the bag; the carrier is an implementation detail that disappears.
site/content/extend/rune-authoring/output-contract.md currently documents one of the six things data-field does, under a table whose last row says the carrier is "removed from output". 111 of 196 data-fields in the catalog are on elements that survive. Correct the table to describe today's reality and state the rule as the one to author against — the gap between them is the work SPEC-133 is about, and naming it is more useful than pretending it closed.
Acceptance Criteria
output-contract.md states the properties / refs rule as a predicate an author can apply to a node without reading the engine- The same page describes what the catalog emits today, including that
data-field survives on elements, rather than only the meta-carrier case - The postprocess sentinels are named as a documented exception rather than left as a footnote in SPEC-082
- The three schema-relevant idioms are spelled out: a value-only source belongs in
properties, a node the schema stamps in place belongs in refs, and a node that is both needs an entry that gives it a name - No rune transform, config entry or stylesheet changes
Approach
Resist scope. The tempting additions — fixing the kebab/camelCase data-field split, moving the first batch of content markers, deleting the dead Page node — are all SPEC-133's, and pulling any of them in imports its wide HTML diff into a milestone whose risk profile is otherwise "structured data only, HTML unchanged".
The one judgement call worth recording here rather than deferring: a node that is both structural and value-bearing (<time data-field="date">). Under the rule it is a ref, and its value reaches the schema by the applier stamping the node in place — which is exactly the resolution strategy WORK-565 builds. So SPEC-133's third open question has an answer once the applier exists, and this item should say so rather than leave the authoring docs silent on the case that recurs most.
References
- SPEC-133 — the rule, the measurements, and the sweep this defers
- SPEC-082 — carved the postprocess sentinels out by hand
- ADR-008 — the flat namespace both maps share
site/content/extend/rune-authoring/output-contract.md — the page to correct
Resolution
Completed: 2026-09-15
Branch: claude/v0.35-parallel-feasibility-eia5le
What was done
site/content/extend/rune-authoring/output-contract.md — the "Properties vs Refs" section rewritten. Docs only; no transform, config or stylesheet changed.
- The rule, as a predicate. "Does this node survive into the rendered HTML?" Survives and might be addressed →
refs. Exists only to carry a value → properties. Stated before the mechanism table so an author can apply it without reading the engine. - What the catalog actually emits. The existing table's "meta tag removed from output" row describes one case out of several. Most
data-field attributes in the catalog are on elements that survive. Names the trap explicitly: the discriminator is n.name === 'meta', not the author's choice of map, so a non-meta in properties behaves like a ref that forgot its BEM class and nothing warns. - Postprocess sentinels given their own subsection as a documented exception, naming the runes that use it, rather than remaining a footnote in SPEC-082.
- The three schema-relevant idioms as a table: value-only source →
properties; a node the schema stamps in place → refs; both (the <time data-field="date"> case) → refs with a named entry, so the schema reaches it by name rather than by attribute. Records SPEC-133's third open question as answered by the applier WORK-565 builds. - One line carried over from WORK-564: CSS selects the BEM element class, never
[property=...], and that is now enforced by a test.
Notes
- Numbers deliberately not quoted. SPEC-133 measured 196
data-field carriers across 95 runes, 111 surviving. Re-measuring through refrakt inspect today over 126 runes gives 280 carriers, 89 metas, 191 surviving — the same shape, a different catalog scope. Pinning either figure into user-facing docs would just create a second thing to drift, so the page states the shape and gives the command to reproduce it. - Scope held: none of SPEC-133's ~97 node moves, no
contentSection removal, no casing fix. Those carry a wide HTML diff and a breaking component-interface change, which is SPEC-133's character and not this milestone's. - The page tells authors to write against the rule rather than against the catalog, and says migration is in progress — otherwise the "what it emits today" section reads as licence to copy the current shape.
Verification
npm run content:check-links, npm run runes:check-docs, npm run format:check all clean; npx vitest run scripts/ packages/lumina/test/css-coverage.test.ts — 284 tests passing.