🧮 #0 CPCV purge/embargo GROUNDED

The leakage-free scoreboard is grounded — and with it the substrate gate passes (#1 + #0 both GROUNDED). Purge/embargo is decisively load-bearing on real data, and the campaign now proceeds to the realness/utility instruments (P2).

✓ PASS

substrate gate (#1 + #0)

t=169 → 0

leak (naive k-fold → purged)

FPR 0.000

null OOS-IC family FPR

2 / 20

instruments grounded

★ Substrate gate PASSES. Both substrate instruments are grounded — #1 future-perturbation invariance (causality floor) and #0 CPCV purge/embargo (leakage-free partition). Every downstream out-of-sample number now rides on a grounded substrate. The campaign advances to P2.

Gate status — all pass

Gate	Result
Fold generator	0 label-window overlaps survive purge+embargo (naive has many)	PASS
Leakage gate	naive random k-fold straddles 252,512 overlapping pairs, leak t = 169; CPCV+purge leaves 0 straddling	PASS
Positive recovery	look-ahead IC* recovered by CPCV OOS, 0.55% bias, power 1.0	PASS
N_min(H) surface	events/fold 290,101; N_min lag5 ρ0.14→30k, lag20 ρ0.079→100k; dialed-IC → PARK	PASS
Harden (zoo→substrate)	KP recovered · KN silent (0/15 folds) · KD identical	PASS
Null FPR	family-wise OOS-IC FPR 0.000 ≤ 0.05	PASS

Headline: purge/embargo is load-bearing

34.85% of consecutive bars close within the 3s label window, so a naïve random k-fold splits 252,512 overlapping-label pairs across train/test — leaking a torrent of spurious skill (t = 169). The CPCV contiguous-group + purge + embargo partition removes every straddling overlap (0 pairs, 0 contamination) while preserving genuine signal (0.55% recovery bias). Leakage-free and unbiased on a null (FPR 0.000).

The PBO-on-null sub-criterion — descoped to #9 (operator ruling)

The frozen §7 #0 row also listed PBO-on-null ∈ [.45,.55]. On real autocorrelated data this does not calibrate: two pre-registered nulls bracket 0.5 (fixed-rule 0.595 high; faithful i.i.d.-block 0.238 low), and the estimate is seed-unstable (the fixed-rule swung 0.595→0.429 between firings from RNG alone). PBO/CSCV is fundamentally instrument #9's tool — its own graduate row owns "block-permuted null ≈ .5" — so requiring it to calibrate at #0 is circular. Operator ruling 2026-07-22: descope PBO-on-null to #9; #0's null gate is FPR ≤ 0.05 (passes, 0.000), and its leakage-free property stands on the leakage gate + FPR. Both PBO values are kept as #9-deferred diagnostics. This scoping decision was made with all three measurements on the table — not a silent goalpost move.

Bugs caught en route (verify-before-report)

block_perm length bug (iter 3) — permuted block indices used as start positions → a garbage 870,400-length array corrupting PBO. Fixed.
Leakage gate under-powered (iter 3) — nearest-boundary metric gave <8 pairs/combo (M0=0). Reformulated to random-k-fold straddling-overlap correlation (252k pairs, t=169).

Verdict & what's next

GROUNDED — the leakage-free substrate is grounded and the substrate gate passes.

Next iteration → P2 begins: #2 block-permutation shuffled-label null (realness). The realness/utility axis instruments now run on the grounded #0/#1 substrate, each through the §4 battery + its Harden.