Preflight (A0): 2026-07-10 19:46 โ load1 3.64/32c (steady) ยท no heavy nasimubd jobs (one gh process, trivial) ยท 35 GiB avail, si/soโ0 ยท all four services active โ ALL PASS. Sync: main up-to-date, worktree 0 behind. One capped run (0.43 min), readonly=2 loader, block re-splits of real values (03c).
| Leg | Reading | Ruling |
| E1 โ certain legs | Substitution pipeline bit-equal ยท twin known-answer: s_twin(w) == 1.0 exactly in all 8 windows (identical rank vectors โ x/x = 1.0 โ derived before running, held) | PASS |
| E2 โ the dossier's own anchor = the door law (the min must rise under pseudo-regimes) | Observed floor-gap 0.0804 vs K=99 block re-split null 0.0560 โ 1.44ร; the second honest clearance, and this one has no estimation layer to hide noise in | PASS |
| E3 โ min-vs-mean self-redundancy | ฯ = 0.843 < 0.95 โ the floor is not the average in disguise; moderately high, travels to M3 as a named secondary adjacency | PASS (noted) |
| Verdict (row 94) | EXAM PASS (first form) โ B-04's exam phase complete; the k_v fight decides the batch's close |
Iteration 83 ยท 2026-07-10 ยท CHECKPOINT (row 94) โ H-072 EXAM PASS ยท capped (0.43 min) ยท readonly=2 ยท zero generated values ยท zero code reuse ยท append-only ยท evidence: mrp_exam.py ยท mrp_exam_results.json