The full training flow — supervision intake, trait discovery, calibration — opens to a private beta first, with deployed models addressable under the submitter's namespace. General availability follows once the beta stabilizes.

Extends: Training.

A guided flow that collects the six elements of a brief through conversation — intent, domain, exemplars, failure modes, edge cases, constraints — and produces a training-ready model from the result. For users who can describe a standard but do not already have labeled samples.

Extends: Briefs.

Automated sample extraction from reference artifacts — policy documents, style guides, exemplar content. The artifact is parsed, positive and negative items are identified from its structure, and the result is consumed as training input.

Extends: the embedded supervision form described on Supervision.

Incremental model updates from signal accumulated during live scoring — acceptances, rejections, downstream corrections. Replaces the need for scheduled retraining to absorb usage feedback.

Extends: the behavioral supervision form described on Supervision.

A provisional model produced from a description and a small unlabeled corpus. Intended as a starting point for a first labeling pass, not a deployment-ready model.

Extends: the training inputs described on Samples and Briefs.

Retained history across retraining runs. Previous trained models remain addressable by version tag, and the differences between versions — shifted breaks, added or removed traits — are recorded.

Extends: the retraining behavior described on Evolution.