The six
| Selector | State | Why |
|---|
pricing.css:65 — .rf-tier h1[property="name"] | live | nameTag is in both refs and schema |
plot.css:2 — .rf-plot > span[property="name"] | live | titleTag is in both |
pricing.css:71 — .rf-tier p[property="price"] | dead | property="price" lands on parsedPriceMeta, a <meta> — not on the <p> |
lore.css:14 — .rf-lore > span[property="title"] | dead | the span carries property="headline" |
bond.css:6 — .rf-bond > span[property="from"] | dead | bond has no schema: map at all |
bond.css:7 — .rf-bond > span[property="to"] | dead | as above |
None of the four dead rules has a BEM equivalent elsewhere in its stylesheet, so a lore title, a bond endpoint and a tier price render unstyled today.
The CSS coverage test does not catch this: it checks that config-derived selectors exist, not that hand-written ones match.
Acceptance Criteria
- No stylesheet in
packages/lumina/styles/ or plugins/*/styles/ selects on property= - Each of the six rules is expressed against the rune's BEM element class instead, and the four dead ones are verified to now apply — a screenshot or a rendered-HTML assertion, not an assumption
- A CSS coverage assertion fails if a
property= selector is reintroduced anywhere lore's title, bond's endpoints and tier's price are styled after this lands, having been unstyled before- BUG-015 is resolved, with the four dead rules recorded as the visual change they are
- Any visual diff is deliberate: a rule that was dead and now applies changes the rendering, and the item says what changed rather than treating it as a no-op
Approach
The four dead rules are the interesting half. Restoring them is a visual change to three runes, not a refactor, and it should be reviewed as such — someone has been looking at unstyled bond endpoints for long enough that the styled version may read as a regression.
The rule this establishes: presentation never selects on the schema.org channel. The refs already produce BEM element classes for every one of these nodes (.rf-tier__name, .rf-lore__title), which is what the CSS should have used from the start.
The coupling matters beyond tidiness, and it is the same argument ADR-028 makes pointed the other way. A theme may not redefine a rune's schema.org output because emission is a claim about content, not about skin. A theme that styles off emission makes the rune's appearance depend on its SEO channel — so a correction to the structured data becomes a visual regression. Every rename in WORK-565's tables would be one.
schema="none" is the sharpest case: stripSchemaOrg deletes property wholesale, so on pricing it would unstyle the tier heading. It is benign on accordion by luck rather than design, and WORK-552 has already shipped.
References
- BUG-015 — the defect, with each selector verified against
refrakt inspect - SPEC-130 — "Constraint: styling already selects on this channel"
- ADR-028 — emission is not theme configuration
- WORK-552 —
schema="none", shipped
Resolution
Completed: 2026-09-15
Branch: claude/v0.35-parallel-feasibility-eia5le
What was done
packages/lumina/styles/runes/{bond,lore,plot,pricing}.css — the six catalogued selectors moved to the BEM element classes the refs emit (.rf-bond__from, .rf-bond__to, .rf-lore__title, .rf-plot__title, .rf-tier__name, .rf-tier__price).packages/skeleton/styles/runes/{annotate,bento,comparison,lore,map,plot, reveal,storyboard}.css — eight more the item's survey missed (see below).packages/lumina/test/css-coverage.test.ts — the guard. Walks every stylesheet under packages/ and plugins/ and fails on any rule selecting [property], in either the ="value" or bare-presence form. Verified to fail by reintroducing a selector, not assumed.plugins/storytelling/test/schema-channel-styling.test.ts and plugins/marketing/test/schema-channel-styling.test.ts — 8 rendered-HTML assertions through createTransform, proving the BEM classes the new CSS selects are really on those nodes.
The survey was short by eight
The item catalogues six selectors in packages/lumina/styles/. There were fourteen: packages/skeleton/styles/ — the second home for rune CSS, which the coverage test already scans — carried eight more.
- Two value selectors,
.rf-lore > span[property="title"] and .rf-plot > span[property="name"], exact duplicates of their Lumina twins. Same defect, same fix. - Six bare
span[property] presence selectors (reveal, storyboard, bento, annotate, comparison, map), each paired with a live meta { display: none }. All six are provably dead: those runes emit no <span> carrying property — annotate, bento and storyboard never mention property in their transforms at all, and reveal/comparison only set contentSection on the block wrapper. Dropped the dead half, kept the meta half.
Eleven of the fourteen selectors were already dead.
Visual changes — three runes gain styling they had lost
lore title — font-weight: bold, margin-bottom: 0.5rem, text colour, plus display: block from the skeleton. Was an unstyled inline span.bond from / to — font-weight: semibold, white-space: nowrap, text colour.tier price — 4xl, bold, tight tracking, margin-bottom: 1.5rem, tight leading. The one a user would actually notice.
tier's name and plot's title were live and move to the same nodes, so no change there.
The finding: the channel hid these rules from another guard
npm test failed on prominence-reach.test.ts — "no rune stylesheet pins a title font-size outside the prominence chain" — the moment the selectors were renamed. That guard (WORK-538) derives its targets from config.sections, as .rf-{block}__{slot} for every title-role slot, so it had never been able to see these two rules: they named the schema.org channel rather than the BEM slot. Styling off that channel did not only risk breakage, it made the rules invisible to the theme's own invariant.
Underneath it was a live bug:
lore — the old rule was dead, nothing was pinned, prominence worked by accident.plot — the old rule was live and pinned font-size on the title element, severing the chain. {% plot prominence="display" %} did nothing, and the density ramp was inert too — precisely the failure WORK-538 exists to prevent, hidden from its guard by a selector name.
Fixed by dropping font-size from both rules rather than exempting them. sections.css gives [data-section="title"] var(--rf-prominence-size, var(--rf-title-size, var(--rf-text-2xl))), and density.css sets the resting --rf-title-size at full density to var(--rf-text-2xl) — the value those rules pinned. Appearance at full density is unchanged; the prominence axis and density ramp are restored.
Notes
- The two guards now compose: reverting to a
[property=] selector fails the new guard first, instead of silently blinding the prominence one as before. schema="none" is not yet the hazard the item describes for tier: stripSchemaOrg has exactly one caller (accordion), so the attribute does nothing on pricing today. A test asserting it here passed vacuously and was replaced. WORK-565 extends it to every rune carrying a table, which is when it becomes real.tier's <h1> legitimately carries property="name" and the BEM class — which is why selecting on it looked reasonable and stayed alive, while the identical idiom on the price (no property at all) went dead unnoticed.- No plugin ships CSS today, so the AC's
plugins/*/styles/ clause is vacuous now; the guard walks that tree anyway so a future plugin stylesheet is covered. - No changeset: Lumina and skeleton CSS ship in published packages, but this is a defect fix with no API change — worth one if the release notes should mention the three runes' restored styling.
Verification
npm test — 365 files, 4468 tests, all passing. npm run format:check clean.