madhuishere-123/datacenter-cooling-env

fix: restore HF Spaces metadata header in README

polished README with scores, incident system docs

v4: NL incident system - LLM reads weather/load/maintenance warnings

v3: physics thermostat, mean score 0.7954

Fix task_critical: reduce difficulty so episode runs full 150 steps

v3: physics-aware thermostat, stability-maximizing, crash-preventing

fix: raise emergency threshold to 31C

fix: emergency override prevents thermal shutdown

fix: reduce LLM calls to every 5 steps to save tokens

fix: remove task_rogue_detection from tasks.py

fix: remove rogue_detection from yaml and tasks

fix: compliant logging format, per-task steps, drop rogue detection

Add LLM-as-judge overseer, reward hacking detection, fix task lengths

Fix openenv validate: pyproject.toml, uv.lock, server/app.py - all checks pass

Update openenv.yaml: add new tasks, carbon fields, overseer endpoints

Fix task_critical: reduce difficulty so model can actually learn

Fix indentation in _compute_reward method

Fix overseer detection thresholds for rogue task

Update inference with rogue detection task

Add AI Safety overseer + rogue detection task

Add AI Safety: Overseer rogue agent detection + task_rogue_detection

Fix: add required [START][STEP][END] log format for evaluation

v2: carbon awareness, cascading failures, task_critical, rich rewards

v2: carbon awareness, cascading failures, task_critical, rich rewards

Force refresh: updated reward logic

v2: carbon awareness, cascading failures, sensor noise, task_critical, rich rewards

Fix Docker build

Fix HF Space metadata

Add complete datacenter cooling environment

initial commit