ML Governance: Quarantine (W11-T12)

Outcome reports flagged as statistical outliers, held out of calibration training pending human adjudication. Dual control on release: the staff member who proposes a release cannot also approve it (DB-enforced CHECK).

Report	Reason	Status	Adjudicated By	Flagged	Actions
No quarantine entries. No statistical-outlier detector is wired up yet to populate this table automatically (disclosed gap, not a fabricated queue).

Releasing sets included_in_training = TRUE on the underlying outcome report, feeding it back into the real calibration training pool (idx_outcome_fit_lookup). No automated outlier detector exists yet to populate this queue — this is a real, tested review surface over a currently-empty table until that statistical engine is built.