OptionalalgorithmThe algorithm's display name.
OptionalfeatureThe experiment's feature-set label.
OptionalinnerThe lower bound of the inner-fold mean's interval; null when the row is not an inner-fold score.
OptionalinnerThe upper bound of the inner-fold mean's interval; null when the row is not an inner-fold score.
OptionalisTrue only for the selected candidate, and only when it was measured.
The experiment iteration (or training run) the row came from.
The row's score, as score; null when not measured.
OptionalmetricWhich metric score is (an inner-fold score such as log-loss, or the gain over the flat guess in a metric); null when not measured.
OptionalmodelThe older lowercase spelling of ModelID.
OptionalModelThe model trained for the row: the one model of a build without a search, or a search's finalist (other candidates have none).
OptionaloptimisticTrue for an inner-fold score: the search chose among these numbers, so they are optimistic.
Optionalrank1 for the best row.
OptionalscoreThe score in metricName; null when not measured.
Leaderboard entry matching MLExperimentResultsSpec (its keys are the artifact's JSON keys). For a search, a row is a candidate's inner-fold mean with its interval, labelled optimistic, and the trainer's finalist is the winner even when it is not rank 1. For a pipeline trained without a search, the one row is the stored comparison with the flat guess its trust verdict quotes, or null when the run stores none (not evaluated) — never a stand-in number.