› Navigation 📂 Forward Orthogonality Matrix Scale Evaluation2026-06-12 Iteration 10 — E1–E3: Leakage-Gated Walk-Forward SKILL GROUNDED
Entry gate: zero leakage — PASS (blocking)
Test Result T1 window disjointness/chronology (sessions excluded) PASS T2 future-perturbation invariance PASS — gates(t) bit-identical under t+1 mutation T3 shuffled-pairing null (blocking) PASS — null AUC CI crypto [0.472,0.522], forex [0.491,0.509] T4 per-cell independence PASS
E1 — forecast skill (t→t+1)
Crypto (2,992 trans) Forex (16,204 trans) Persistence among orthogonal-at-t 95.1% 93.6% redundancy-margin AUC (conditional: who stays) 0.945 (overall 0.997 [0.996,0.998])0.880 (overall 0.983 [0.982,0.985])wold_R / adv_auc / granger (conditional) 0.278(inv) / 0.526 / 0.452 0.623 / 0.585 / —
E2 — conformal coverage (temporal 60/40 split)
Substrate 80% nominal 90% nominal Verdict Crypto 80.9% 88.4% (width ±0.090) well-calibrated Forex 70.4% 83.4% under-covers → needs ACI
E3 — panel error control
Crypto Forex Declared stays-orthogonal (BH=BY, q=0.10) 4 / 16 : aggression_ratio, bar_dispersion_entropy, bar_katz_fd, bar_petrosian_fd72 / 141 PBO (CSCV) 0.00 0.00 Deflated max-z vs null-expected-max 22.15 vs 1.53 12.49 vs 2.45
Confluence: the crypto FDR-controlled declared set = exactly the features the orthogonality audits promoted (aggression_ratio + the newly-backfilled bar_* trio) — the walk-forward did not know which features had been promoted.
Forex anchor set (empirically most-central): duration_us, intra_duration_us, max_gap_duration_us, intra_intensity_qps, intra_burstiness
SSoT: audit folder E123-WALK-FORWARD-RESULTS.md + e123_results.json + e123_walk_forward.py
Campaign: 2026-06-12-forward-orthogonality-matrix-scale-evaluation · generated 2026-06-12 14:13:33 local · read-only · 19,196 transitions