Performance & conviction calibration
Does a higher conviction score mean the AI is right more often? This page tracks it, using every completed analysis saved in the app.
Conviction vs. directional win rate
Loading calibration…
How this is measured
- Every completed analysis saves a snapshot that cannot be edited afterwards: symbol, verdict, conviction score, the price when it was generated, a summary of the agents' reasoning, the scoring model version and the time.
- An outcome job then records the daily closing price 1, 5 and 20 trading days later. A trading day is a daily bar published by the data provider, so weekends and market holidays are skipped. The job is safe to run repeatedly: each outcome is recorded once.
- Right direction means the price moved the way the call pointed: up for BUY / long, down for SELL / short. Forex and CFD "no trade" results count in the direction of their better-scoring side. A stock HOLD has no direction and is not scored.
- A win rate is only shown once a bucket has at least 20 scored outcomes. Below that the result would mostly be noise, so the table says Not enough data yet.
- Direction only: the size of the move, trading costs, stops and targets are not part of this score.
- Scoring model: conviction-v1 (head-trader self-reported, 0-100). Conviction is the AI decision-maker's self-reported confidence, not a probability. See what conviction means.
Past results in a practice app do not predict future results. Virtual money only; this is not financial advice.