results_comparison.md

ID3QNE Sepsis OpenEnv Results

Policy	Mean Score	Density	Steps	Safety
Heuristic	0.9867	1.00	9.7	100%
LLM (gpt-4o-mini)	0.9867	1.00	9.7	100%
ID3QNE	0.9867	1.00	9.7	100%

Statistical Validation

• LLM 10-episode mean score: 0.9867
• LLM 10-episode score std across episode means: 0.0
• LLM global reward density: 1.0
• LLM safety violation rate: 0.0

Key Result

All verified policies achieved dense reward performance with zero safety violations in the local OpenEnv sepsis benchmark.

Notes

• The OpenAI-backed policy was constrained to the environment action schema and guarded against unsupported outputs.
• In this environment, the observed performance ceiling is 0.9867, and both the LLM-controlled run and ID3QNE matched that ceiling.