Phase 8 · Predictive readiness

Machine-learning readiness

Every holding is exposed as an analysis-ready feature table: one row per entity, stable identifiers, typed numeric columns in fixed units and derived features computed in the database. Model runs are versioned, so predictive layers can be published without changing the schema.

Profiling feature tables…

Model run registry

Each training run is recorded with its target, feature source, algorithm, version, hyperparameters and metrics; predictions are stored per location and linked to the run that produced them. Published runs are visible here; unpublished runs stay with reviewers and administrators.

Loading runs…