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