Preflight (A0): 2026-07-10 15:45 โ load1 spiked 10.23 (sharply rising vs load15 5.37; heavy external I/O ~1.1M blk/s) โ re-sampled 45 s later per the spare-only rule: 7.69 and FALLING, si/soโ0, 27 GiB avail, all four services active, external python job 1-core (under bar) โ PASS on re-sample, both samples recorded. Sync: main up-to-date, worktree 0 behind. One capped run (4.56 min), readonly=2 loader, permutation-free (e-process readout).
| Check | Reading | Bar | Ruling |
| 1 ยท NSUB dial law | ฯ(@1000, @1500) = 0.994 | ≥ 0.90 | PASS |
| 2 ยท MMD shadow (DECISIVE โ named in row 81 before the fight) | ฯ(bet_mean, mmd_mean) = 0.982 on the same 45-pair grid | |ฯ| < 0.95 | FAIL โ echo of certificate #13 |
| The echo signature (corroboration) | Every correlate mirrors the MMD readout's row-79 profile: k_v โ0.358 (MMD: โ0.35) ยท IPP โ0.671 (โ0.679) ยท calgap 0.794 (0.786) ยท eicp_log_min_e 0.831 (0.842) ยท CD d_min 0.804 (0.804) | โ | TOTAL MIRROR |
| Mechanism | The bettor stakes on the SAME kernel-embedding witness the distance meter reads โ its wealth is a sequential MMD estimate; anytime-validity (real: exam rejection at pair 110/1500) is not a ranking delta | โ | STRUCTURAL |
| Verdict (row 82) | EXCLUDE(echo of certificate #13 โ |ฯ| 0.982 โฅ 0.95 vs mmd_mean; a different inference wrapper around the same witness is the same instrument โ H-053 precedent, second sighting) โ KILL #16, BONEYARD #31 |
Boneyard guidance recorded: do not re-propose kernel-witness betting variants; a betting form with a genuinely different witness must name its non-MMD delta at M0. The exam's wealth-identity certain case (row 81 โ the cleanest E1 of the campaign) stays on record, reusable for any future betting-family form.
Iteration 71 ยท 2026-07-10 ยท VERDICT (row 82) โ EXCLUDE ยท kill #16 ยท boneyard #31 ยท capped (4.56 min) ยท readonly=2 ยท zero generated values ยท append-only ยท evidence: betting_m3_census.py ยท betting_m3_census_results.json