Model card

astris-retention-ensemble-v1

Governance-weighted blend of the four retention voters (Cox survival, LightGBM, logistic regression, neural MLP). Voter weights are updated from observed placement outcomes by leave-one-out attribution (M5.1) and an EMA trust-region simplex update (M5.2); a Wald SPRT (M5.3) gates promotion of a challenger blend before it ships.

At a glance

Kind	scoring ensemble
Version	1.0.0
Status	Active — contributing to scores
Ensemble weight	1
Trained at	Not trained yet
Trained on	n/a
Artifact	No artifact (heuristic or pre-training)

Metrics

{
  "voters": [
    "cox",
    "lightgbm",
    "logistic",
    "mlp"
  ],
  "validation": "synthetic_seed",
  "real_cohort": "pending",
  "voter_weights_sum": 1
}

Notes

Internal synthetic validation only; real-cohort benchmark pending. Weights move as placement outcomes enter the governance window — see api/ml/retention_ensemble.py and governance.py.

← Back to the registry