Wave 7 — 3D Package Model Coverage

Coverage manifest for the wave-7 3D package model sourcing pass (issue #145) — every published footprint package's sourcing outcome, the one LCSC data-quality substitution found, and the two architecture decisions the pass made.

This page is the coverage manifest required by issue #145 (S7 of the components-docs-restructure epic). It records, for every one of the 27 published footprint packages, whether a reviewed .wrl/.step 3D model pair was sourced and from which LCSC id — the deliverable the 3D-viewer work (#146) consumes to know which records carry a resolved model.

Result: 27 of 27 sourced, zero unsourced

Every published package — including the 3 that previously carried no (model …) line at all — now resolves to a reviewed, committed 3D model in footprints/kicad/zudo-pd.3dshapes/. easyeda2kicad was run once per package against the LCSC id its record[s] use in .claude/skills/component-spec-audit/references/inventory.json (or, for scripts/schgen/board_{a,b}_spec.py-only fields, the same LCSC number those spec modules use). No package needed to be recorded as unsourced.

Every .wrl file passed the same VRML 2.0 / no-traversal / no-executable-node / allowed-node-set validation references.ts's validateVrml() enforces at generation time (checked standalone before committing, then proven again by pnpm test:components against the real corpus). All 27 .wrl files are individually well under the 2 MB per-model cap (REFERENCE_LIMITS.modelBytes), and the sum of all 27 is ~2.6 MB against the 8 MB aggregate cap.

The one substitution: a supplier data-quality mismatch, not a package error

CAP-SMD_BD10.0-L10.3-W10.3-LS11.0-FD has two inventory lines: C2983319 (GVT1E477M0810CNVC) and C22387780 (EFVH035ADA471M10B0). Fetching C2983319 first — the natural "first line wins" choice — returns a 3D model EasyEDA itself names CAP-SMD_BD8.0-L8.3-W8.3-LS9.1-FD: an 8.3 mm package geometry wired to a part whose own footprint is the 10.3 mm BD10.0 outline. This is upstream EasyEDA/LCSC library data being wrong for that one LCSC id (a known class of issue for electrolytic caps on LCSC, where a part's linked 3D asset can be copied from a similar-but-different sibling) — not a mistake in this project's package assignment. It was caught by cross-checking the model's own reported dimensions against the footprint name's declared dimensions before wiring anything in, and confirmed by re-running pnpm generate:footprint-previews after the fact: dimension mismatches on this scale would have been visible immediately in the rendered silkscreen/courtyard outline sitting inside a wrongly-sized model, had the page rendered one.

C22387780's own EasyEDA data returns a correctly BD10.0-sized model (CAP-SMD_BD10.0-L10.4-W10.4-LS11.2-FD — 10.4 mm, matching within normal manufacturer-to-manufacturer footprint tolerance), so it was used instead. The .kicad_mod's existing (offset (xyz …))/(rotate (xyz …)) transform was also replaced, not just the path, because it is specific to whichever geometry file it is paired with, and the two LCSC ids' models are not the same underlying geometry.

Two architecture decisions this pass made

Directory name: followed the code, not the issue text. Issue #145 and the wave-7 brief both say the target directory is footprints/kicad/zudo-power.3dshapes/. The consuming code (doc/component-docs/adapters/circuit/paths.ts's MODEL_ROOT, references.ts's MODEL_PREFIX) already hard-codes footprints/kicad/zudo-pd.3dshapes/ from an earlier wave. Using the issue text's name would have made every model silently unresolved (readPackage() would treat the (model …) line as "not a reviewed local asset" even though the file exists on disk, since the prefix check would never match) — so this pass followed the code.

Flipped reference.model.* from DENY to PUBLISH. matrix.ts states outright that these four fields "flip when their consumer lands — not before." Sourcing the assets made entry.model resolve for the first time, which made projectFootprint()'s previously-dead publishRequired branch live — and it failed loudly exactly as designed, breaking pnpm build, pnpm generate:components, and pnpm test:components. Investigating showed the consumer already existed: ui/component-references.tsx's ModelCard / ResolvedModel (a static "Open the VRML model" link — explicitly not the interactive WebGL viewer, which its own comment calls out as "a separate port" that "needs assets that do not exist here yet") has rendered both the resolved and unresolved states since an earlier wave. With the emitter already in place and 27/27 packages now resolved, this pass flipped the four fields — mirroring wave 6's identical reference.footprint.path flip — and updated the two test files whose assertions described the old zero-assets state. The interactive WebGL viewer itself remains out of this pass's scope.

Scope note: the other ~30 footprints in the library

footprints/kicad/ holds footprints beyond the 27 the manifest publishes (parts removed from the design, superseded revisions, hand-created pads, etc.). Their (model …) lines were left untouched — re-sourcing them is outside what the manifest or any published record needs, and several no longer correspond to a current inventory line at all.