Move the remaining lettered sub-steps of transformRune into facets: density (1e), width / spacing / inset (1f), reading and dropcap (1f-bis), content-measure, and reveal / stagger. Individually these are small; together they are most of what remains between step 1 and step 2, and clearing them is what lets the lettered numbering go away.
9bba8b128d33eca859ee6cf489ad70b55505d3998d629dbde07f6c3d487755adensity, width, spacing, inset, reading, dropcap, content-measure and motion (reveal + stagger) are facetsdropcap declares after: ['reading'], replacing the current adjacency-only guarantee, and keeps its off-register warndensity continues to resolve author attribute ▸ parent childDensity ▸ rune default ▸ full, reading the parent rune's config through the facet contextmotion keeps data-reveal / data-stagger emission and the --rf-reveal-index stamping on cascade items — the latter via postAssemble, since it walks assembled childrenwidth, spacing and inset keep their default-suppression rules (content, default) so unmarked output stays byte-identical1b…1h) remain in transformRunereading → dropcap gatenpm run build, the full repo suite, and refrakt contracts --check passThese are the axes the WORK-517 pattern was designed for, so most are a direct port. Three need care.
density reads the parent rune's childDensity, so the facet context must expose the parent config lookup — the first facet to need more than the current rune's own config.
stagger stamps an index on cascade items in document order, which is a children walk and therefore postAssemble, not resolve.
content-measure is config-derived rather than author-driven (config.contentMeasure === 'anchored'), a reminder that not every axis reads an attribute.
Each facet takes its vocabulary with it — READING_REGISTERS and READING_CAPABILITIES already live in reading.ts and can stay there, re-exported by the facet rather than duplicated.
Completed: 2026-08-27
Branch: claude/transform-package-refactor-7mxi8a
packages/transform/src/facets/box.ts — width, content-measure, spacing, inset. Grouped in one module because they are the same family and each is a handful of lines; they remain separate facets so the registry still orders them individually.packages/transform/src/facets/density.ts — the density axis and DENSITY_VALUES.packages/transform/src/facets/reading.ts — reading and dropcap, re-exporting the vocabulary from ../reading.ts rather than duplicating it.packages/transform/src/facets/motion.ts — reveal / stagger plus stampStaggerIndex, whose index stamping runs in postAssemble.packages/transform/src/facets/types.ts — added FacetContext.parentConfig.packages/transform/src/engine.ts — steps 1e, 1f, 1f-bis, the reveal/stagger block and step 7c all gone. Down to 1463 lines from 2223 before the facet work began.packages/transform/test/facets/scalars.test.ts — 38 tests covering default suppression per axis, the reading→dropcap gate, density inheritance precedence and stagger stamping.Ordering is now a list, not an accident. Registry order reproduces the original emission sequence, so class order, modifierValues key insertion order and inline-style declaration order all follow from one ordered declaration. That also fixes the benign output difference WORK-519 introduced: --tinted is back ahead of --width/--spacing/--inset, and output is byte-identical again.
dropcap declares after: ['reading'] — the capability gate reads the resolved register, a constraint that previously held only because the two blocks sat five lines apart.
Three axes publish state rather than emitted axes, because the engine owns their emission point: reading and dropcap are applied to the body section during child assembly, and data-density is unconditional at a fixed position. density is also the first axis to read config other than its own, so FacetContext gained parentConfig — the resolved parent config rather than the rune registry, so a facet cannot reach arbitrarily into the rune graph for one lookup.
A mistake worth recording. Extracting stampStaggerIndex I first rewrote it from memory and got it wrong three ways: dropped the data-field match, mutated in place instead of replacing the array element, and invented a rune-boundary check that does not exist. Caught by diffing against the original before running anything; the committed version is verbatim. None of the three would have been caught by the suite — the data-field path has no direct coverage — which is the argument for copying rather than retyping during a behaviour-preserving move.
"No lettered sub-steps (1b…1h) remain in transformRune" is not satisfied: 1b (context modifiers) and 1c (static modifiers) remain, and both are explicitly WORK-523's scope — its own criteria name steps 1, 1b and 1c. The criterion was mis-scoped when written; everything in this item's actual scope (1d–1h) is cleared. Left unchecked rather than reworded, so the overlap stays visible.