You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
test(characterisation): re-baseline drifted snapshots after 2026-06 adviser reworks (ABT #10)
The [characterisation] suite had 11 stale snapshots (all captured 2026-05-19)
after the saturation reclassification + isat gate, the proximity-aware
single-winding re-rank, and the loss/saturation recompute landed on main.
Each drift was verified as an intended consequence before re-pinning:
- MagneticFilter SATURATION valid true->false: the 40+20-turn fixture is now
classified as an inductor (no isolation sides) and gated by gap-aware isat;
an ungapped ferrite at ~1.73 A correctly fails. ENERGY_STORED/TEMPERATURE
drift from the same reclassified flux recompute.
- CoreAdviser AVAILABLE/STANDARD POWER: proximity-aware re-rank promotes EP 20
over PQ 20/20 / PQ 27/15 on real winding-proximity loss (intended, ~1.6%).
- MagneticAdviser 3W 507 kHz: the under-turned E 8.3/4 (10 turns) now fails the
saturation gate; the 20-turn E 10/3 wins. Valid pool thinned to 2 so slot 2
is an INVALID-penalised candidate (documented thin-pool fallback).
- CoreCrossReferencer DEFAULT/SAME_MATERIAL/ONLY_TDK: powder re-shuffle below a
stable top-1 ferrite + sub-0.3% score nudges.
- CoilAdviser flyback top[1]: benign within-class reinforced-wire re-rank
(ABT #3).
- CoreAdviser AVAILABLE x INTERFERENCE_SUPPRESSION: re-baselined to the SAME
three XFlux 26 toroids as the 2026-05-19 baseline now that the suppression
returns-0 fix restores candidate flow (see preceding 3 commits); scores
nudged < 0.6%.
Full [characterisation] suite green (314 assertions, 55 cases); [smoke-test]
gate unaffected.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
0 commit comments