BOCPD Filter — BB_EXTREME_3 robustness (N=30 seeds)
PromotedHypothesis
With 30 shuffled-source backtests instead of 5, the BB_EXTREME_3 z-score either firms up above 1.5 (deploy) or collapses below (skip).
Verdict
**FIRMS UP — DEPLOY** — with N=30 the z-score = +1.52 >= 1.5. BB_EXTREME_3 has a real BOCPD signal. Deploy bocpd_filter@0.3 too.
z_score
+1.5207
real_pnl
+12.2107
shuf_p5_pnl
+9.8599
baseline_pnl
+11.4243
shuf_p95_pnl
+11.6485
shuf_std_pnl
+0.5835
real_lift_pct
+6.8836
shuf_mean_pnl
+11.3234
n_shuffle_seeds
30
shuf_mean_lift_pct
-0.8832
BOCPD Filter — BB_EXTREME_3 robustness (N=30 seeds)
2026-05-19 · status: promoted · 184.6s
Hypothesis: With 30 shuffled-source backtests instead of 5, the BB_EXTREME_3 z-score either firms up above 1.5 (deploy) or collapses below (skip).
Verdict: FIRMS UP — DEPLOY — with N=30 the z-score = +1.52 >= 1.5. BB_EXTREME_3 has a real BOCPD signal. Deploy bocpd_filter@0.3 too.
Key metrics
| metric | value |
|---|---|
| n_shuffle_seeds | 30 |
| baseline_pnl | +11.4243 |
| real_pnl | +12.2107 |
| shuf_mean_pnl | +11.3234 |
| shuf_std_pnl | +0.5835 |
| shuf_p5_pnl | +9.8599 |
| shuf_p95_pnl | +11.6485 |
| real_lift_pct | +6.8836 |
| shuf_mean_lift_pct | -0.8832 |
| z_score | +1.5207 |
Approach
30 shuffled-source backtests on BB_EXTREME_3 with bocpd_filter at θ=0.3. Same setup as bocpd_filter_shuffle_test but more seeds to firm up the z-score (was +1.06 with N=5).
| metric | value |
|---|---|
| baseline_pnl (no filter) | $+11.42 |
| real-source filter pnl | $+12.21 |
| real lift vs baseline | +6.88% |
| shuffled mean pnl | $+11.32 |
| shuffled std pnl | $0.583 |
| shuffled 5/95 percentile | [$+9.86, $+11.65] |
| shuffled mean lift vs baseline | -0.88% |
| real - shuf_mean | $+0.89 |
| z-score (real vs shuffled distrib.) | +1.521 |
Verdict
FIRMS UP — DEPLOY — with N=30 the z-score = +1.52 >= 1.5. BB_EXTREME_3 has a real BOCPD signal. Deploy bocpd_filter@0.3 too.