Iter 14 — FINAL: E4 verdict — skill, calibration, stamps all CONFIRMED

Iteration 14 — E4 FINAL VERDICT CONFIRMED

Evaluation	Result on complete data
Leakage gate (rerun)	PASS — shuffled null collapses to chance (both substrates)
E1′ forecast skill	CONFIRMED — redundancy-margin conditional AUC 0.945 crypto / 0.877 forex (deltas vs v2 ≤0.003)
E2′ calibration	FIXED — ACI restores forex coverage to 89.9%/79.9% vs 90%/80% promised (was 83.4%/70.4%)
E3′ stamped set	CONFIRMED on full bar_* coverage — same 4 features (aggression_ratio, bar_dispersion_entropy, bar_katz_fd, bar_petrosian_fd); crypto PBO 0.00
ICP regime-invariance	REPLICATED — 0/92 crypto features invariant → declarations are regime-conditional
Open defects	forex ICP wiring (0 envs); wold_R inversion replicates (investigate); forex PBO 0.2 (under 0.5 line, monitor)

Evaluation

Result on complete data

Leakage gate (rerun)

PASS — shuffled null collapses to chance (both substrates)

E1′ forecast skill

CONFIRMED — redundancy-margin conditional AUC 0.945 crypto / 0.877 forex (deltas vs v2 ≤0.003)

E2′ calibration

FIXED — ACI restores forex coverage to 89.9%/79.9% vs 90%/80% promised (was 83.4%/70.4%)

E3′ stamped set

CONFIRMED on full bar_* coverage — same 4 features (aggression_ratio, bar_dispersion_entropy, bar_katz_fd, bar_petrosian_fd); crypto PBO 0.00

ICP regime-invariance

REPLICATED — 0/92 crypto features invariant → declarations are regime-conditional

Open defects

forex ICP wiring (0 envs); wold_R inversion replicates (investigate); forex PBO 0.2 (under 0.5 line, monitor)

Layman's verdict: we re-took every photograph with the complete dataset and re-graded everything. The skill held, the error bars now keep their promise on both markets (the "thermostat" fix worked to within 0.1%), and the four stamped features kept their stamp. The one rule the data insists on: "stays independent tomorrow" is only valid within the kinds of market weather we've seen — which is exactly why declarations come with a regime condition and a live alarm.

SSoT: audit folder E4-FINAL-VERDICT.md + e123_v3_results.json + icp_v3_cross_regime.json

Campaign: 2026-06-12-forward-orthogonality-matrix-scale-evaluation · 2026-06-12 15:17:49 · E4 final · read-only · 4×(2c/2GB) shards