Nova Origins

ML/adaptive learning

ML Learning Foundation

ML-D1 is infrastructure only. Baseline heuristic scores are operational prioritization — not scientific confidence. No discovery model is trained. No ML output implies anomaly, biosignature, habitability, atmosphere, or life.

ML-D1 infrastructure only. Training disabled. Autonomous inference disabled. Reinforcement learning deferred. External model APIs disabled. Human validation required.

ML Policy

Foundation: enabled

Baseline heuristic: enabled

Training: disabled

Autonomous inference: disabled

Reinforcement learning: disabled

Scientific claims: disabled

External model APIs: disabled

Adaptive Learning Roadmap

1. Heuristic scoring ✓ (current)

2. Human validation ✓ (current)

3. Supervised learning (future)

4. Active learning (future)

5. Contextual bandit (future)

6. True RL (future)

Feature Store

targetarchive_recordobservation_recordobservation_quality_reviewtarget_archive_quality_review

Feature snapshots capture operational metadata only. Source records are never mutated.

Baseline Scorer

Heuristic review priority scoring (0-100).

Does not score life, anomaly, biosignature, or habitability.

Reinforcement Learning Deferred

Reinforcement learning and contextual bandit optimization are deferred until sufficient validated outcomes, training labels, and reward events exist.

No autonomous reinforcement update occurs. No model is retrained. No reward event triggers automated behavior.

Scientific Caution

No ML output means anomaly, biosignature, habitability, atmosphere, or life detection.

Baseline scores are operational prioritization only.

All ML outputs require human validation.

No external model APIs are enabled.

Learning Core Preview

ml_learning.feature_snapshot_createdml_learning.training_label_createdml_learning.reward_event_recordedml_learning.baseline_prediction_loggedml_learning.model_run_recorded

All ML-D1 events include: operational_ml_only, no_scientific_interpretation, reinforcement_learning_deferred.