Lab · ML Experiments

ML — Pattern Discovery

Inverted workflow: find conditional edges in BTC data first, build strategies second.
55 experiments

CVD Filter — Definitive Shuffle Test

Dropped
2026-05-19 validationshufflecvdfilter
Hypothesis
Lifts identified in cvd_filter_universal are genuine CVD/OFI signal if real-source backtest z-score vs shuffled-source >1.5 over 10 permutation seeds.
Verdict
**TRADE-COUNT EFFECT** — no candidate passes shuffle test.
n_signal
0
n_shuffles
10
n_candidates
3

CVD Filter — Definitive Shuffle Test

2026-05-19 · status: dropped · 250.0s

Hypothesis: Lifts identified in cvd_filter_universal are genuine CVD/OFI signal if real-source backtest z-score vs shuffled-source >1.5 over 10 permutation seeds.

Verdict: TRADE-COUNT EFFECT — no candidate passes shuffle test.

Key metrics

metric value
n_candidates 3
n_signal 0
n_shuffles 10

Approach

For each candidate (strategy, filter, threshold): run baseline + real cvd source + 10 shuffled-cvd sources. Real-source z-score vs shuffled distribution > 1.5 → genuine order-flow signal.

Results

strategy filter baseline_pnl real_pnl shuf_mean shuf_std real_lift_pct z_score
BB_EXTREME_2 align 17.66 18.69 16.98 1.483 +5.9% +1.16
BB_EXTREME_2 div 17.66 32.47 32.47 0 +83.9% +1.00
DONCHIAN_50 div 5.52 7.16 7.16 0 +29.7% +nan

Verdict

TRADE-COUNT EFFECT — no candidate passes shuffle test.