DEGENERATE — harness defect (crypto-named active set)
crypto-validated
redundancy
INPUT-ONLY (rank-corr is permutation-invariant → null gate inapplicable). Anchors read 1.0 on every combo ✓
the input series
stab_freq
6/21
0/19
unreliable — drop or redesign
bootci_w
MISCALIBRATED
MISCALIBRATED
rank-derived → null-invariant
ICP cross-regime
91 features → 0 invariant
0 features (same harness defect)
see finding
Verdict: CONDITIONAL — recorded without inflation. Strict pre-registered count = 1 cross-substrate USEFUL (PIVOT territory); but two failures are attributable harness defects (forex active-set mapping) and one metric is INPUT-ONLY by construction. Within valid scopes the discriminating scales replicate perfectly (40/40, 19/19, 21/21).
Substantive findings
ICP: zero of 91 crypto features have a regime-invariant relationship to the active set — empirical confirmation that any "orthogonal tomorrow" declaration must be regime-conditional (consistent with the audit's ~8/10 structural ceiling).
Rank-derived statistics cannot serve as forward gates (permutation-invariant null).
The full-grid loop is cheap & safe to re-run: ~2 h, 251 MB, read-only, resumable.
Named defects → next iteration
Define the forex active set (fxview column names) → unlocks granger_echo + ICP on forex.