The `spice` field on a digital part is a power and loading macromodel — it draws
the right quiescent current and cannot shift a bit. Showing it under a single
"SPICE model" heading implies the part is simulatable, which for the digital
parts it was not.
`spice_mixed` is a model that performs the part's actual function, verified by
simulation, and it is now shown as its own section with what was verified and
what is not modelled. The download bundles both, so a deck gets the functional
model and keeps the supply behaviour.
With real geometry in, a fixed label size turns multi-character names on fine-pitch
pads into overlapping mush — USB-C's A1/B12 row was unreadable. Scale per pad instead,
bounded by both the pad width (divided across the label) and its height.
The real geometry now arriving from EasyEDA spells a round pad ELLIPSE, which
this test did not match — so every through-hole pad would have drawn as a square
with the right dimensions. OVAL/ROUND are the KiCad-side spellings of the same thing.
Every part rendered "✓ 3D model ✓ Schematic ✓ Footprint ✓ SPICE" as literal markup,
regardless of what it actually had. footprint_pads is absent from all 80 parts, so
the Footprint tab says "no footprint data" while the checklist two inches away
claims a tick — which is how a missing footprint reads as a rendering bug rather
than missing data.
Each item now derives from the field the corresponding tab actually reads.
Two EasyEDA-converted connectors (WJ128V-5.0-3P, HR911105A) shipped with every
material at transparency 1.0. The WRL downloads, parses and renders perfectly --
as nothing: blank tile, a valid PNG of empty space, no error anywhere. The models
are fixed; this stops the next bad vendor material vanishing the same silent way.
