get_insight

Scope: read · Signature: get_insight(experiment_id: str, kind: str, params: dict[str, Any] | None = None) -> Any

Use for computed analysis artifacts beyond charts and model evaluation.

kind="kinds" lists the kinds that apply, their meanings and allowed query_params. Families: evaluation (leaderboard, scores, diagnose, decision_layer, composition); trust & explain (trust_summary, fairness, drift, error_tree, case = per-row why, engineered_features, surrogate, …); archetypes (incl. archetype_map, archetype_diagnostics); forecast_; unsupervised_; association_; recommendation_ / ranking / ltr. params fills path placeholders and query options (case needs row_index; association_audience / association_evidence need rule_id; text_insight needs encoder + feature). A surface that is not computed or does not apply returns a graceful note (with the backend's reason); a malformed param is an error naming it. Unsupported options are rejected, never silently ignored. Meanings: get_playbook("explainability").

Powered by: mli explanations, fairness, drift & trust (mli/docs/guides/explanations.md).