OptionalevaluationThe MJ: ML Evaluation Protocols row (exact version) the run was evaluated under.
OptionalevaluationThat protocol row's Version.
OptionalgateThe decision-table branch the stored outcome came from (for example b:keep-simple-baseline or c:no-signal).
OptionalgateThe stored gate outcome.
OptionalgateThe stored reason for the outcome, in words generated from the stored comparisons.
OptionalholdoutJSON of the test-set scores measured once on the locked holdout ({ "<metric>": value, … }): the produced
model's stored scores when a model exists (the member the gate ships), otherwise the trainer's finalist scores.
True when one feature dominates the finalist's predictions (possible target leakage); a flagged model stays Draft and needs human sign-off.
OptionalmodelId of the produced (Draft) MJ: ML Models row; absent when the gate decision ships no model (no-signal, or the existing rule kept).
OptionalshippedThe member the gate ships (the model, a simple baseline or the existing rule); absent under no-signal.
Lifecycle status of the produced model (Draft), NoSignal when the gate stored no-signal, or KeptExistingRule when it kept the existing rule.
Id of the MJ: ML Training Runs row recording this attempt.
OptionaltrainingThe MJ: ML Training Snapshots row the run trained on.
OptionalversionMonotonic model version under the pipeline; absent when no model is produced.
Output of
PredictiveStudio.TrainModel: the run's stored gate decision, its Training Snapshot and Evaluation Protocol, and the produced model when the decision ships one.