Mechanism diagram · Rule / Decile
CROSS-SECTIONAL RULE / DECILE Feature value funding-z or OI velocity Cross-sectional rank at each bar Decile partition top decile vs bottom Weights no model training · single feature

How it works

At each bar, each coin's funding rate is normalised to a z-score against its own history.

Coins are ranked cross-sectionally by funding z. The top decile and bottom decile define the long/short positions.

No model training. The rule is the architecture.

Pros and cons on this universe

Pros

  • Most orthogonal arm on the screen (distinctness 0.66) — useful diversifier.
  • Rank #20 IC — positive despite being a single-feature rule.
  • Zero training cost; reproducible from raw data.

Cons / failure modes

  • Not actually a “neural architecture.”
  • Weak vs TimeMixer alone — rules decay as their statistical bias erodes.
  • STRAT-17 deployment failed (walk-forward −0.527 Sharpe). The rule does not survive standalone deployment.

References