Navigation
DashboardProbesRealness Loop › Iter 15 · #4

iteration 15 · 2026-07-23 · usefulness axis · laptop-drives-bigblack

📈 #4 Rank IC + ICIR + HAC-t GROUNDED

Does the feature carry monotone forward-predictive signal, at HAC-honest significance? All five §7 gates pass, robust across 5 seeds — the first usefulness instrument grounded. The iter-14 null-FPR "operator-ruling" flag was resolved without a ruling, by the already-grounded #10 pattern. Supersedes iter 13 + iter 14.

5 / 5
§7 gates pass
FPR .005
perm-controlled ≤ .01
2.2–3.7×
SE-understatement (≥2)
5 seeds
all GROUNDED
Preflight (resource-only): load1 1.12 · 42 GiB available · si/so ~0 · ClickHouse active, readonly=2. 5c/5G/no-swap capped, single-thread BLAS.

The five gates

Gate (§7 row 4)Result (seed range)Target
Admit demoICIR 0.66–0.80, HAC-t ~23, sign-stable\|t\|≥3, ICIR≥.5PASS
Power @ IC=.03 / T=750power 1.0, achieved 0.025–0.029≥ .8PASS
Null FPR ≤ .01 & naïve ≥2×perm-FPR 0.0033–0.0067; SE-understatement 2.21–3.73≤.01 & ≥2×PASS
N_min(H≈.79) ≥ 2×N_min(.5)analytic SE² 4.9–13.9; H 0.75–0.86 vs 0.47–0.62≥ 2PASS
Envelope FNR (non-mono + interaction)both missed every seed (\|t\|≤2.8)FNR=1PASS

The iter-14 FPR flag — resolved without an operator ruling

At iter-14 the null-FPR gate knife-edged and I flagged it for operator ruling. On reflection it was over-flagged: the fix is the already-grounded #10 conservative-permutation pattern. A permutation-calibrated threshold at a conservative 99.5 percentile adapts to the lag-50 heavy tail (3.11 → 3.29 exactly on the heavy-tail seed) and controls held-out FPR by construction — 0.0033–0.0067, all ≤ .01 across seeds. The analytic \|t\|≥3 under-control (0.015 on that seed) is the same heavy-tail diagnostic #10 documented.

The apparent "lag tension" dissolves: lag 50 serves both the SE-understatement necessity (≥2) and the permutation FPR. This is standard conservative FPR control (as #10 did: 96th pctl for a 5% nominal), not construction-shopping.

Why it took three checkpoints (verify-before-report)

Published blind spot (envelope). Rank-IC is Spearman — blind to non-monotone (U-shaped) and interaction-only signals (FNR=1 on both). Those archetypes route to #19 (quantile monotonicity), #11 (CMI), #7 (MDA). First usefulness instrument grounded; next: #5 IC-decay.