tiny-aya-global-sql-selfheal-bco-replay (BCO + success replay). Deployed.

LoRA adapter for CohereLabs/tiny-aya-global (3.35B) from a fully automated self-healing text-to-SQL loop (Lexsi Labs Hackathon, Track 3, PS #8). Failures are detected by executing the SQL on real Spider databases, classified, corrected by Aya Expanse 32B (every fix verified by execution; the healer never sees the gold query), and retrained through AgentTune. Deployment is decided by AgentTune's DeploymentGate on held-out data.

Recipe: BCO with success replay: 153 verified fixes (True) + 153 wrong queries (False) + the model's own 347 correct pool answers as independent positive rows (impossible under DPO's pairing requirement). 653 rows, 82 steps.

Result: APPROVED by the DeploymentGate on a 400-question held-out gate (59.25% vs 59.0%, recommendation deploy). On the unseen 500-question test split: 47.6% execution accuracy vs 45.6% baseline (+2.0pp, fixed 34 / broke 24), valid-SQL rate 73.0% vs 68.2%. Independently re-verified by AuditKIT (500/500 agreement).

• Project page + code: https://huggingface.co/spaces/HideIron/selfheal-sql
• All run logs and gate decisions: https://huggingface.co/datasets/HideIron/selfheal-sql-runs
• Training data: 158 execution-verified fix pairs (heal/baseline-pool600 in the runs repo)