Preflight (A0): 2026-07-10 05:05 โ load1 2.99/32c (โค24) ยท no heavy jobs ยท 26 GiB avail, si trickle/so=0 ยท CH + sidecar + kintsugi active โ ALL PASS. Sync: origin/main unchanged. One capped run, 0.53 min (per-run 240s wall guards + 12-min budget guard armed, unused), readonly=2 loader, tigramite via ephemeral uv, real rows only.
| Leg | Reading | Ruling |
| E1 identification (12/12) | Exactly one link involving the twin in every run: the contemporaneous o-o PAG adjacency with duration_us | PASS |
| E2 rejection (12/12) | Zero links twin โ {intensity, burstiness} at any lag โ latent-aware conditioning still explains the twin entirely via duration | PASS |
| F014 trap (latent variant) | 0 crashes, 0 NaN p-values across all 12 runs โ the PAG machinery processed the byte-twin's exact-zero conditional residuals cleanly | NOT SPRUNG |
| The dial (the family's killer) | pc_alpha exists with NO model-selection mode (API checked). Exam-scale invariance observed (identical answers at both ฮฑ, both ฯ) โ but the H-004 sibling ALSO showed exam-scale invariance and then died at census scale (row 58). Exam-scale evidence is NOT trusted; the census ฮฑ check is pre-registered DECISIVE. | CENSUS DECIDES |
| Verdict (row 65) | EXAM PASS โ checkpoint; ladder walk unlocked with the ฮฑ question front-loaded |
Iteration 54 ยท 2026-07-10 ยท CHECKPOINT (row 65) โ exam PASS ยท capped (0.53 min) ยท readonly=2 ยท zero generated values ยท append-only ยท evidence: lpcmci_exam.py ยท lpcmci_exam_results.json