Model Versions

Roadmap

SwarmFixer model training and versioning is planned. No SwarmFixer model has been trained or deployed. The scheme below is design intent.

Models would be versioned SwarmFixer-vN.M. Each version would be logged with: cumulative training-pair count · per-mode repair-lift baseline · holdout test scores.

In keeping with house doctrine, versions would never be auto-promoted. An operator approves each promotion against operator acceptance gates — the specific gate set is design intent and not yet defined from a real source. The discipline is the point: operator approval, not automatic promotion.

---

🐝 Operator-grade · books and records · to the shed.