UNTESTED registered EVALUATION axis A5 (representation), with training-set size as the pre-registered sweep parameter no run
Everything in this block was recorded in version control before any run. It is rendered above the result on purpose: a prediction a reader meets after the number it predicted is not a prediction.
Health-acoustic pretraining buys sample-efficiency, not ceiling: HeAR leads at small labelled-set size and is overtaken by Whisper-enc as n grows.
Two results in apparent tension. (i) SpeechDx (arXiv:2606.17339) ranks general encoders above domain-specific ones on average. (ii) Sanap, Desikan, Lobaton, 2026, Beyond Classification: A Cough Regression Benchmark for Respiratory Acoustic Foundation Models (arXiv:2606.15436): HeAR and M2D+Resp reach near-full performance at 50 samples where OPERA needs 400. Both can be true only if the crossover exists — and it has never been plotted head-to-head.
Sweep labelled-set size n_train ∈ {25, 50, 100, 200, 400, 800, all}, subject-disjoint. If HeAR never crosses Whisper-enc at any n_train on any of the 3 tasks (Holm-corrected), the "health pretraining is the right prior" claim is falsified and reduces to a low-data convenience argument. If HeAR wins low and loses high, the crossover point is the deliverable.
HeAR leads by 0.04–0.10 AUC at n_train ≤ 50; crossover at n_train ∈ [200, 800]; Whisper-enc leads by 0.01–0.04 AUC at full data.
| tier (as registered) | EVALUATION |
| family size m / seeds n | m = 42 (2 contrasts {HeAR vs Whisper-enc, HeAR vs WavLM} × 7 sizes × 3 tasks) · n = 12 → 0.000488 ≤ 0.05/42 = 0.001190 ✓ |
| datasets | SVD (voice pathology) · Coswara (respiratory) · COUGHVID (respiratory, OOD) |
| cost | One embedding-extraction pass per encoder, then trivial probe training. |
| min attainable paired p | 0.000488 |
| Holm-tightest threshold 0.05/m | 0.001190 |
| arithmetically satisfiable under R6 | ✓ |
| registry status | UNTESTED |
Registry status is read from the ## Summary table of IDEA_TABLE.md, which the author maintains as a set. The per-block status cells are maintained one at a time and disagree with it; one source of truth, and this is it.
UNTESTED No artifact exists for this hypothesis. Under R7 a hypothesis whose falsifier has not been executed is UNTESTED and is never reported as supported. This page exists so that the prediction above is on the record before any result is, and so the debt is visible rather than absent. It is counted as debt on the master dashboard and in the registry table.