AI evaluation

A probability needs context before it becomes useful.

Models add evidence to a decision alongside data quality, strategy and risk checks. Read their available calibration and provenance rather than interpreting a raw score as a promise.

RESEARCH · REVIEW · DECIDE
01Know the evidence

Know the evidence

A model output depends on its feature definitions and training history.

  • Inspect available model version and evidence timestamps.
  • Distinguish raw scores from calibrated probabilities.
  • Missing model context must remain visible.
02Validate the result

Validate the result

Out-of-sample behavior matters more than an appealing training score.

  • Look for untouched evaluation data and leakage checks.
  • Compare performance by asset and market regime.
  • Use sufficient outcome evidence to evaluate uncertainty.
03Keep decisions governed

Keep decisions governed

AI research does not grant execution permission.

  • Account risk, data admission and kill switches remain independent.
  • Experiment history informs review rather than silently changing live rules.
  • No model can guarantee future profitability.
Disclosure. Trading can result in losses. Features depend on your plan, market coverage and account permissions. Paper activity is simulated; broker execution requires independent verification and authorization.
Continue reading: Explore the platform Open secure account