CODEC/FOCI cascade — Selection verdict

2026-06-05T03:28:20Z · machinery-only on the 325-cell leakage-free grid · preflight PASS

Selection-partial — ξ REPRODUCES the Pearson-derived duplicate/orthogonal labels with stable, leakage-free walk-forward Jaccard (S1/S2-ξ/S3/S5-ξ machinery PASS). NO incremental value over Pearson/Spearman is demonstrated (S4 deferred — circular labels); 2 of 3 cascade stages (CODEC, O-info) have NO analytic null (HELD); the S5 headline pairs are degenerate (3 near-deterministic dups ξ≈0.97–0.999, 4 near-zero ξ≈0.02–0.17, nothing in the contested middle).
Deployment 🔒 BLOCKED on Terry.

Provenance: S3/S5 aggregate a grid generated 2026-06-03 (PRE-S0, by the committed Exploration code); the Selection gates RE-AGGREGATE, they do not re-measure. test_leakage.py validates the MACHINERY on synthetic fixtures, not these 325 cached cells. Threshold-tightening would require regenerating the grid under a fresh pre-registration.

GateStatusDetail
S0FROZENpre-registration in candidates/codec-foci-selection-s0.md
S1PASSmutation oracle: kill_rate 0.9606 over 127 non-equivalent mutants; 0 boundary survivors
S2-xiPASSMARGIN/SPECIFICITY check — every orthogonal pair sits ≥0.30 below the 0.80 flag (worst-epoch confident-flag rate 0.0); NOT a boundary-calibrated false-flag rate (no orthogonal pair lives near 0.80).
S2-CODECHELDno faithful FOSS analytic null (never faked)
S2-OinfoHELDno faithful FOSS analytic null (never faked)
S3PASSDESCRIPTIVE stability — mean Jaccard 0.8811 / non-canonical 0.8393 (6 ρ≈1 dups removed); MIN-over-epochs 0.8523; worst-epoch PBO 0.0 < 0.50; leakage_free True. Floor 0.20 NON-BINDING (worst cell 0.4821).
S4DEFERREDcircular: Y labels are Pearson/Spearman-derived
S5-xiPASS7 headline pairs; arctanh-transformed cluster jackknife + t-multiplier (g=6, 5 dof). Reporting gate, not coverage-validated.

A clean pass validates the ξ sieve internally; it does not claim value-add over baseline (S4 deferred) or forward predictivity.