HF_SPACE_DEBUG.md

HF Space State Persistence Issue

Problem

RAM usage and rewards stay the same after pressing step in HF Space app, regardless of action type/intensity.

Root Cause Analysis

✅ Backend is working correctly - direct environment tests confirm:

• Step 1 (reduce_ram 0.8): 80% → 72% ✓
• Step 2 (reduce_ram 0.8): 72% → 64% ✓
• Rewards update correctly: 0.080 each step ✓

❌ Issue is in HF Space app request handling:

• Each HTTP POST request may be creating a new environment instance
• State is not persisting across requests
• Session management might not be working correctly

Solution

For HF Space Users:

1. Clear browser cache - Sometimes old state is cached
2. Use WebSocket /ws endpoint instead of REST (this should be automatic)
3. Check browser console for HTTP 422 errors
4. Refresh the page completely before starting new episode

For Developers:

The OpenEnv framework manages this automatically via:

POST /reset → Creates new environment instance
WebSocket /ws → Maintains session state across multiple steps
POST /step → Updates persistent environment state

The 422 errors in server logs indicate malformed requests from HF Space. Check:

1. Request body format matches EnergyOptimizationAction schema
2. Content-Type header is application/json
3. Session/connection is properly maintained

Verification (Local Testing)

✅ Direct Python test: PASSED ✅ Environment state updates: CONFIRMED ✅ Reward calculation: WORKING ✅ RAM reduction logic: FUNCTIONAL

Test Summary:
- Reset: RAM=80%, Energy=8.0
- Step1: RAM→72%, Reward=0.08
- Step2: RAM→64%, Reward=0.08
- Status: ✅ WORKING CORRECTLY

Next Steps

1. Rebuild HF Space Docker image to ensure latest code
2. Test with curl/Postman to verify API works
3. Check WebSocket connection in browser dev tools