Webull Calibration Method
Purpose
This page explains how Webull strategy buckets become calibration evidence. It is intentionally short: the binding stop rule, Wilson lower bound, and Kelly-form break-even are defined in the F7 whitepaper.
Bucket Identity
Each closed paper trade with a non-empty webull_strategy_id contributes to exactly one bucket, such as long_call, bull_put_ladder, or short_box_spread. The gate does not pool unrelated structures.
Observation Fields
The gate reads closed rows from cv_paper_trades. The key fields are status='closed', webull_strategy_id, pnl_usd, and realized_r. A row without realized PnL is not calibration evidence.
Probability Estimate
p_hat is the bucket's observed win rate. p_lower is the Wilson 95% lower confidence bound used by the gate so the UI does not size or judge from the optimistic point estimate.
Break-Even
b is the empirical R-multiple win/loss ratio for the bucket. The probability break-even is break_even = 1 / (1 + b). A bucket clears only if the lower bound beats this threshold after the sample gate is met.
Verdicts
INSUFFICIENT_SAMPLE means the bucket has not reached the required closed-trade count. EDGE_REJECTED means the sample exists but the edge did not clear. EDGE_CONFIRMED means the paper bucket earned operator review; it is not an automatic live submission.