This page documents the construction and evaluation workflow for the positioning model.
We benchmark level reactions against a matched baseline to see whether the market follows through more at the model’s identified areas than it does under identical conditions picked at random.
Event Detection Intraday price and participation are scanned against session norms. Only statistically abnormal events are kept.
Level Formation Events inside a volatility-scaled band become a single level. Width and spacing are measured, not hand drawn.
Forward Testing Touches are replayed in sequence at 5 to 240 minute horizons. Outcomes use the same thresholds that run live.
Scoring & Selection Expected value blends win rate, payoff profile, and sample depth. Shrinkage keeps small samples near neutral until they earn weight.
Instrument: ES (S&P 500 E-mini). Baseline: matched sessions with identical time windows, no lookahead applied.