Blog/Research & Analysis
Research notes · provenance-first

Plain-language analysis of deployed models.

Every note ships with the model card, feature attribution, calibration plot, and the timestamped forecasts it was generated against.

13articles
9archetypes
2asset classes
13 of 13 articles
Title
Bid-Ask Imbalance: Volume-Weighted Spread Changes as Trading Signals
How a percentile-regime model reads BTC and when it stands aside
How a composite predictive signals model traded ETH 94 times and missed four gates
The Percentile-Regime Archetype: Ranking a Market State to Decide When to Go Long, Short, or Stand AsideReusable method · applies across assets
The Volatility-Regime Reversion Archetype: Fading a Deviation Only When the Volatility Regime Supports ReversionReusable method · applies across assets
The Conditional Z-Score Archetype: Standardizing a Market State and Trading Only When a Gate AgreesReusable method · applies across assets
The Signal-Ensemble Archetype: how a Composite Predictive model trades agreement, not any one inputReusable method · applies across assets
The Correlation-Signal Archetype: reading one move through another, until the link breaksReusable method · applies across assets
The Put-Skew Regime Archetype: trading the price of fear, when it is overdone and when it is earnedReusable method · applies across assets
The Z-Score Reversion Onset Archetype: standardizing a market-state input, and acting on the turn rather than the extremeReusable method · applies across assets
The Volatility Surface Archetype: reading the shape of the options surface, not the price, to infer directionReusable method · applies across assets
The Open-Interest Positioning Regime Archetype: reading where the options crowd is positioned, not the price, to infer directionReusable method · applies across assets
Inside QBTS's open-interest positioning gate: 66 trades, a failed drawdown limit, and a trailing benchmarkequity
Transparent. Probabilistic. Built for real decisions.

Every note links to the deployed model card, the feature attribution, and the live forecast log.

Read our methodology  ↗