Clamp all remaining score leak paths: /state, step_rewards, demo SSE

Clamp every grader return value strictly inside (0, 1)

Defensive score clamping at all emission points

Widen score epsilon to 0.01 so :.3f formatting stays in (0, 1)

Fix reward.value to always be task_score in (0,1) exclusive

Fix task scores to be strictly in (0, 1) exclusive per OpenEnv spec

fix

fix

fix

Fix IndexError: skip empty-choices chunks from HF Router streaming

fix: GEPA current_generation, task_id mapping, Connect DB button, remove difficulty from header

fix: default LLM to HF Router + Qwen2.5-72B, no custom Space variables needed

Initial submission: SQL Agent OpenEnv for Meta+HF hackathon