forward-predict orthogonality · living dashboard · campaign-grounded

🔮 Forward — Orthogonal Tomorrow EVALUATING — NOT OPERATIONAL

Will a feature that is orthogonal today STILL be orthogonal tomorrow? The three axes grade a feature today; this probe grades whether the verdict persists. Status: in progress — under evaluation. The walk-forward evidence below is grounded, but the probe is not yet an operational gate: open defects remain and the live monitor is unbuilt. All numbers below are leakage-gated, twice-replicated walk-forward results over 20 named regime slices (10 crypto + 10 forex), probe-SSoT symbols × all production thresholds. Frozen record: campaign spoke (14 iterations) · grounded summary.

94–95%

persistence into next regime

0.95 / 0.88

flip-ranking AUC (crypto/forex)

89.9%

ACI coverage @90% nominal

FDR ≤ 10%

declared-set error bound

The gate ladder (all leakage-gated)

Gate	Question	Grounded result	Status
Entry: zero-leakage	is the walk-forward structurally unable to cheat?	T1 disjoint windows (sessions excluded) · T2 future-perturbation invariance · T3 shuffled-pairing null collapses to AUC≈0.5 (blocking) · T4 cell independence	PASS
E1 forecast skill	does today predict tomorrow?	orthogonal-today features persist 94–95%; redundancy margin ranks flips at conditional AUC 0.945 crypto / 0.877 forex (Δ≤0.003 partial→complete data)	PASS
E2 calibrated uncertainty	do the error bars keep their promise?	crypto 88–90% · forex FIXED by ACI adaptive conformal: 89.9%/79.9% empirical vs 90%/80% nominal (fixed band had failed at 83.4%/70.4%)	PASS
E3 panel error control	is a declaration luck or mining?	BH=BY q=0.10 declared sets · PBO 0.00 crypto / 0.20 forex · deflated max-z passes · survives shuffled-target	PASS
ICP regime-invariance	is anything immune to regime change?	0/92 crypto features invariant — REPLICATED ⇒ every declaration is regime-conditional	PASS
E6 live monitor (CTM)	when does a verdict expire in production?	designed (anytime-valid conformal test martingale); historical-replay validation not yet run	PENDING

The declared set (crypto, FDR ≤ 10%, confirmed on complete data)

aggression_ratio · bar_dispersion_entropy · bar_katz_fd · bar_petrosian_fd
Independently re-discovered by a walk-forward blind to the promotions — the audit-promoted features are exactly the ones that persist across regime transitions with controlled false discovery. Forex: 72/136 declared at q=0.10.

The verdict grammar

ORTHOGONAL-TOMORROW(feature | regime-class R) — every forward declaration carries a regime scope (ICP: nothing is unconditionally invariant) and, in production, a live expiry monitor. The unconditional claim is provably impossible; the campaign measured that ceiling rather than ignoring it.

Open items

forex ICP wiring defect (0 environments returned) — debug pass
wold_R conditional inversion on crypto (0.278, replicates) — mechanism investigation
forex PBO 0.0→0.2 on complete data — watch on next rerun
live-monitor replay validation (E6) — not yet run
~5–6 catalogued matrices still worth testing (DSR-proper, CPSS, nonlinear ICP, Failing-Loudly, VAR, EnbPI variants)

Provenance

Audit twin: 2026-05-26-forward-orthogonality-prediction · results: e123_v3_results.json · harness: e123_v3_walk_forward.py · 50-matrix parameter classification: MATRIX-PARAMETER-CLASSIFICATION.md

← The Gate—

Three-Axis Probes · LIVING reference (regen via build_probes.py) · updated 2026-06-10 · audit twin · campaign spoke