Iterated the convergence against the live Hoffman originals (no DSCA regression — DSCA33/45
held at 54/56, staged-92 render):
- rendersWithoutSpecs() now covers the mined 8B/5B (54 lacked spec files, same wipe gap as
DSCA33/45) so they render via the template path instead of the missing-specs bail.
- Accuracy formatter unified + driven by units parsed from the accHeader (the miner's accOut
is "?" for Iout models): output mA/mV -> x1000 on calc/meas; input mV/uV -> x1000 on stim;
temperature input -> 2dp; std/freq signed, AC unsigned; Math.fround throughout.
- splitSpecUnit now splits after the last number, so multi-word units survive
("< 17 uA RMS" -> unit "uA RMS", not "RMS").
- Verbatim "Model:" name captured per model (8B/5B prefix inconsistently: "SCM5B49-03" but
"5B45-03D") and rendered directly.
Result: 69/136 validated and marked (24 8B, 21 SCM5B via the template path; 24 SCM7B already
correct via the unchanged legacy 7B path). 67 remain (gated null, safe): 12 need slotMaps the
oracle couldn't derive; the rest are family-divergent tail — 8B/5B Final-Test drops the sign on
negative measured values (opposite of DSCA's keep-sign) needing family-specific measured
formatting, plus per-model "Packing Check List" footer variants. enrich-8b5bscm.js now also
captures modelName.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mike Swanson
1c630552e0