Research

Tr8dr labels — solid y, partial OOS permission

Amplitude-based segments on EURUSD TIMB: D(h) peak +0.55, C4 stack permission R=0.56 valid → 0.34 OOS 2024.

In brief. Before rich models, you need a stable label . Tr8dr amplitude-based (15 bps / Tinactive 60) on EURUSD TIMB: 48.6% signed occupancy, peak +0.549 at H=32. Context stack C4 yields live permission R=0.558 on valid — 0.344 OOS 2024. Solid foundation, not yet a strategy.

Why I ran this study

Everything downstream (process gates, HMM, events) rests on a frozen directional segment y. If the label is coding noise or not economically separable, the rest is decoration.

Pipeline: L0 census → validity → univariate T2/T5 screening → C4 stack → OOS 2024 lock.

Definition (summary)

Up / neutral / down segments via min amplitude + inactivity timeout:

L0 metricValue
Signed occupancy48.6%
Segments / day~5.2
Duration p50 / p9083 / 206 bars

Key results

Validité économique D(h) — peak +0,549 à H=32 bars (EURUSD TIMB, 2023)

P1 screening: T5 termination AUC ~0.63–0.69 (range, vol); T2 onset modest (~0.55–0.65) — conditioning, not standalone entry edge. T3 direction: several inverted signals (AUC < 0.5) → careful with orientation.

Context C4: best config C_bps_slim q=0.5 → R=0.558, live rate 50%, AUC 0.606 valid.

Permission live C4 — R valid 0,56 → OOS 2024 : 0,34 (dégradé mais positif)
MetricValidOOS 2024
R0.5580.344
Live AUC0.6060.587
Live rate0.4970.442

Takeaways

QuestionAnswer
Sound label?Yes
C4 permission?Yes in-sample
OOS 2024?Partial — R 0.34, signal present but degraded
Standalone strategy?No — regime gate

Recommended use: frozen y + C score as permission in the lab stack (HMM, process, events).


Research note — Aug 2026. EURUSD · TIMB · 2023–2024.