Skip to the evidence
AI Trading SignalsDefinition / Population / Proof

An AI evaluation manifest

The artefacts to preserve before a model's designated results are inspected.

Frozen identity

Record model and feature versions, data identity, training boundaries, label definition and decision rule. Note when retraining produces a new artefact. Do not let later improvements inherit an earlier model's supposed validation.

Test provenance

Save which data were used for development, which results were viewed and what remains genuinely untouched. Calendar separation is one narrow check, not a full leakage certification.

Result classification

Label simulation, forward observation and account execution separately. Preserve failed runs and missing outputs. A defensible report states exactly what was measured rather than calling an untested plan a validated trading system.

Sources

Dated source record

Research guides

AI data leakage and holdout audit

Audit feature availability, outcome overlap and test exposure with a frozen experiment record and a worked calendar-boundary example.

Backtest, forward and live evidence

Reconcile simulated decisions, prospective instructions and account fills without promoting one evidence layer into another.

Algorithmic versus AI signal claims

Map fixed rules, trained predictions, AI explanations and order routing to the specific consumer evidence each needs.

An exact question beats a vague verified badge

Fictional packet F-9 contains a model label and one score but lacks the event definition and original outcomes. The brief can report what the seller states, not calibration. Its next request names the output contract and population rather than asking for another selected chart. If those records arrive, preserve their source and date alongside the earlier gap.

Keep requested, supplied, inspected and unresolved statuses distinct. A citation is not automatically an inspected run, and a completed workbench record is not authentication of its contents. Remove credentials and private account identifiers before sharing exports. Unknown fields remain meaningful evidence limits rather than blanks to populate with optimistic claims.