2026-09-26 sponsor questions

The library

Rendered from docs/reports/2026-09-26_sponsor-questions.md; the markdown file is the record.

One-time draft review

Approved by Nathan, Sat Sep 26, ~08:25.

Payload: submissions/20260926-0813_M4_sweep_E_on_A_alpha0.2_our_direction.tsv, copied to ~/Downloads/HEARSAY_predictions.tsv. The two copies are byte-identical.
Rule: M1b v3 80% + handcrafted v5 20% (rank), with M3 used only for false-alarm suppression. Pre-declared in docs/reports/2026-09-26_fusion-sweep-predeclared.md.
Validation:
- holdout minDCF 0.014
- In-the-Wild 0.260 (brief cost) / 0.267 (sponsor-code cost)
- test share above 0.5: 27.4%
Ask NSA for: minDCF, plus P_FA, P_miss and EER.
Pre-flipped copy: ..._FLIPPED_only_if_NSA_scores_inverted.tsv.

Decoding the returned minDCF (per the fusion consult response):

Returned minDCF	Action
0.00–0.20	Ship unchanged.
0.20–0.45	Switch to the most FA-robust candidate.
0.45–0.90	Don't flip; fall back to M1b v3 alone.
0.95–1.00	NSA's code is reading our scores inverted: submit the pre-flipped file.

If NSA returns EER instead, 2–5% means our direction and 95–98% means theirs.

Sent: Sat 12:30 by Discord DM (team Cross Exam), superseding the 08:13 payload above: CrossExam_predictions.tsv = submissions/20260926-0914_M4_sweep_A3_w0.2_E_CANDIDATE_our_direction.tsv (rule A3 w0.2 + E, sha256 fb783076…, submissions/log.csv row 12:30:37).

Returned (Sat, ~15:30, relayed by Nathan): minDCF 0.0733, EER 3.534%. No P_FA, P_miss, threshold, cost weighting or direction convention was given.

Decoding: 0.0733 is in the 0.00–0.20 band, so the shipped rule stays; EER 3.5% (not 96.5%) confirms 1.0 = synthetic is being read as intended. No flip, no fallback.
Against our proxies: holdout 0.0065, In-the-Wild 0.228; the test set sits at 11× the holdout and a third of In-the-Wild, next to M1b v3 alone on the holdout (0.072).
Field (draft review, Sat ~16:15): one team ahead at minDCF 0.0584 / EER 2.5%; under points = (1 − ours)/(1 − best) × 60 that is a 0.95-point gap on the 60-point detection score.
Status: the number is the fixed baseline for any further work before the Sunday 05:00 freeze. The follow-up consult on what to build next is docs/consults/2026-09-26_post-draft-review_CONSULTATION.md.