โ€บNavigation

Iteration 46 โ€” H-004 PCMCI+ Passes the Entrance Exam EXAM PASS ยท 12/12 RUNS ยท B-01 EXAMS COMPLETE

In plain words: the last candidate of the batch is a full causal-map drawer: given several data streams, it draws which drives which, including same-instant connections โ€” and unlike most instruments, being built for autocorrelated data is its home turf, not a stretch. Its exam had a hidden trap: the target IS one of the predictors, byte for byte, and testing anything else "given" that predictor leaves literally zero wiggle room โ€” the exact situation that blinded a famous earlier candidate (F014), whose whole falsifier collapsed there. This one walked through cleanly twelve times out of twelve: in every regime tested and at every settings combination, it drew exactly ONE connection to the target โ€” the correct same-instant edge to its byte-twin โ€” and nothing else, with no crashes and no undefined numbers. Its two dials didn't change the answer anywhere. With this exam, every candidate in the first batch has now been examined: six earned certificates, two died honorably, and this one enters the ladder. Field test next.

Preflight (A0): 2026-07-10 02:34 โ€” load1 2.11/32c (โ‰ค24) ยท no heavy jobs ยท 37 GiB avail (si trickle/so=0, scope swap-banned) ยท CH + sidecar + kintsugi active โ†’ ALL PASS. Sync: origin/main unchanged. Provenance (dossier instruction): tigramite 5.2.10.1, GPL-3.0, fetched ephemerally via uv. One capped run, 0.41 min, readonly=2 loader, real rows only (deterministic stride).

Exam results (pre-registered form, pcmci_exam.py)

LegReadingRuling
E1 identification (12/12 runs)Exactly one link involving the twin in every run: the contemporaneous o-o adjacency with duration_us (undirected โ€” PCMCI+ cannot orient a deterministic contemporaneous pair; adjacency was the gated claim)PASS
E2 rejection (12/12 runs)Zero links twin โ†” {intensity, burstiness} at any lag โ€” conditioning on duration_us explains the twin entirely, everywherePASS
The F014 trapEvery CI test conditioned on the byte-twin is exactly degenerate (zero residual variance) โ€” the situation that blinded F014/causalicp. PCMCI+: 0 crashes, 0 NaN p-values across all 12 runsNOT SPRUNG
Dials (swept in-exam, rows 51โ€“53 lesson)ฯ„_max โˆˆ {2, 5} ร— pc_alpha โˆˆ {0.01, 0.05} across the 3 largest FX windows (deterministic env rule) โ€” identical answer at every combinationDIAL-INVARIANT (exam scale)
Verdict (row 57)EXAM PASS โ€” checkpoint; ladder walk unlocked. B-01 exam-level consumption COMPLETE: 10/10 candidates processed (6 certificates incl. candidate A's two legs ยท 2 honorable kills ยท PCMCI+ in ladder)

โ–ถ Next iteration

Iteration 46 ยท 2026-07-10 ยท CHECKPOINT (row 57) โ€” exam PASS ยท capped (0.41 min) ยท readonly=2 ยท zero generated values ยท append-only ยท evidence: pcmci_exam.py ยท pcmci_exam_results.json ยท tigramite 5.2.10.1 GPL-3.0