WORK-534 catches authored attributes at validation time. It does not catch values arriving another way — scoped defaults and embed overrides (ADR-027) apply attribute bags to runes that never spelled them out, and those still need a runtime diagnostic.

Acceptance Criteria

reading warns when the resolved register would apply but the rune has no body section to carry it
content-place warns when set on a rune that declares no content-place modifier
scrim-strength warns when set on a rune in cover mode, where the media well cannot honour it — and is consumed either way, so it stops leaking its raw <meta> tag into the rendered tree
Messages match the shape and tone of the four existing warnings
Each goes through the WarningCollector with a dedupeKey, so it reports once per build rather than once per rune instance
Each is asserted as returned FacetWarning data in the facet unit tests, not only through a console.warn spy
The default cases stay silent: an unmarked rune with no reading set must not warn
npm run build, the full repo suite, and refrakt contracts --check pass

The cover-scrim row was mis-stated

The premise "cover scrim* is dropped on a rune that cannot enter cover mode" does not hold. scrim, scrim-type, scrim-strength, scrim-blur and scrim-tone are read by the background layer (bgAxis.contract.inputs lists all five), which raises itself on any rune a scrim is set on. Verified against the engine: a rune config of { block: 'grid' } — no sections, no media slots, no media-position modifier — plus scrim="bottom" produces a full rf-grid__scrim overlay with data-scrim-dir and --scrim-strength. Cover mode does not enable the scrim; it reroutes it to the media well.

Two consequences:

WORK-534 over-narrowed. Filing scrim* under the cover axis — which is gated on a declared media-position modifier — removed a working attribute from 82 of 83 runes. Only card kept it. Fixed here by moving the family to the bg axis, where the facet that implements it lives; cover now owns no author-facing attribute, like density and content-place, and is still reported per rune.
The real drop is the reverse. scrim-strength is the one scrim facet cover mode cannot honour — cover's SCRIM_META omits it, so in cover mode it is neither applied nor consumed, and the unconsumed <meta> survives into the rendered tree.

Approach

Independent of WORK-534 — it can land before or after, and does not block it. Grouped into v0.32.0 because it completes the same "no silent no-ops" goal.

The appliesTo-gated pair needs care: warning from inside a facet that deliberately does not run means either relaxing appliesTo and returning a warning-only result (the shape prominence already uses), or a check outside the facet. Prefer the former — it keeps the diagnostic with the axis that owns it.

References

SPEC-125 — Phase 3
SPEC-124 — the facet warning collector
ADR-027 — scoped attribute bags, the non-authored input path

Resolution

Completed: 2026-09-09

Branch: claude/milestone-v0-31-0-e5ihxr PR: refrakt-md/refrakt#594

What was done

The three gated universal axes that dropped a request in silence now report it.

packages/transform/src/facets/reading.ts — reading-without-body. The facet always resolved and published the register as state; the value only became data-reading once child assembly found a body-role element, so on a body-less rune it vanished with nothing aware it had been asked for. Warns only on an explicit, valid reading=: every unmarked block resolves to the ui default, and a typo falls through the cascade for Markdoc to reject. hasBodySection was hoisted above the facet — it already existed for the contract descriptor at the bottom of the same file.
packages/transform/src/facets/content-place.ts — content-place-undeclared. appliesTo now also fires on the bare attribute, so the facet runs purely to warn on a rune whose config declares no content-place modifier. The existing content-place-outside-cover warning is untouched and stays distinguishable by code.
packages/transform/src/facets/cover.ts — scrim-strength-in-cover, plus consumes. Silent when scrim="none" (the author already said no scrim), and consumed either way so the meta stops leaking.

The third case was not what the item described

The item's premise — "cover scrim* is dropped on a rune that cannot enter cover mode" — does not hold, and I checked before implementing it. scrim* is read by the background layer, which raises itself on any rune: a bare { block: 'grid' } config plus scrim="bottom" produces a full rf-grid__scrim overlay with data-scrim-dir and --scrim-strength. Cover mode does not enable the scrim, it reroutes it to the media well.

Two consequences, both handled here (with the user's agreement to substitute the real defect rather than close the criterion as invalid):

WORK-534 over-narrowed, and this fixes it. Filing scrim* under the cover axis — which is gated on a declared media-position modifier — removed a working attribute from 82 of 83 runes; only card kept it. {% textblock scrim="bottom" %} had become a validation error for markup the engine honours. The family moved to the bg axis, where the facet that implements it lives. cover now owns no author-facing attribute — like density and content-place — and is still reported per rune.
The genuine silent drop runs the other way: scrim-strength is the one scrim facet cover mode cannot honour (cover's SCRIM_META omits it), so it was neither applied nor consumed, and the raw <meta> survived into the rendered tree. That is now the third criterion.

The work item body records the correction and the evidence; the criterion text was rewritten rather than ticked off as-written.

Tests

packages/transform/test/facets/dropped-axes.test.ts (13) — each warning asserted as returned FacetWarning data (code, message, dedupeKey), never through a console spy: the driver owns emission, so a console test would be testing the driver. Covers the silent cases too — no request, a non-register value, a rune that has a body, scrim="none" — since a diagnostic that fires on the default case trains readers to ignore the channel.

Two of my own assertions from WORK-534/535 encoded the over-narrowing and had to be corrected: reference-universals.test.ts listed scrim-tone among grid's absent attributes, and the language-server completion test asserted scrim was not offered on grid. Both now assert the opposite, with the reason.

Notes

Schema narrowing does not make these redundant. WORK-534 catches authored attributes at validation time; scoped defaults and embed overrides (ADR-027) apply attribute bags to runes that never spelled the attribute out, and those never pass through Markdoc validation.
Pre-existing inconsistency, not fixed here: the two older warnings prominence-unsupported and dropcap-off-register carry no dedupeKey, so they still report once per rune instance rather than once per build. The criterion covers the three new ones, which all dedupe; aligning the older pair is a small follow-up worth doing separately since it changes existing output.