test(bfld): measured §3.6 separability + audit's cardiac-alone negative result
Deterministic synthetic-data tests producing reproducible, honestly-labeled
numbers (MEASURED-on-synthetic, explicitly NOT real-person identification):
- same_person_scores_higher_than_cross_person: self-match ≈1.0000,
cross-person ≈0.8088 (full channels) — a real but modest ~0.19 margin.
- cardiac_alone_cannot_separate_identity_matches_audit (centerpiece): with the
decisive channels (AETHER 0.35, subcarrier 0.20) absent, cardiac (0.15) +
respiratory (0.10) alone give same=1.0000 cross=0.9995, gap=0.0005 — no
threshold fits, so the matcher correctly refuses to lock identity. Proves the
audit's claim 'your heartbeat alone overlaps too much' with real numbers.
- Graceful degradation, zero-norm/NaN safety, insufficient-channels typed
result, empty-enrolled-set, threshold boundary, min-channels gate.
13 new tests; full crate suite 364 passed / 0 failed.
Co-Authored-By: claude-flow <ruv@ruv.net>
e2864bbd52