EmpathRAG / eval /README_multiturn_eval.md
MukulRay's picture
Strict cleanup: remove internal handoff docs, neutralize teammate/sprint naming in public docs
091059d
|
Raw
History Blame Contribute Delete
586 Bytes
# Multi-Turn Evaluation Harness
This harness evaluates the current EmpathRAG fast backend without loading the full local LLM stack.
Run:
```powershell
$env:PYTHONIOENCODING='utf-8'
.\venv\Scripts\python.exe eval\run_multiturn_eval.py
```
Inputs:
- `eval/multiturn_scenarios.jsonl`
Default output:
- `eval/multiturn_results.json`
Metrics:
- route accuracy
- safety tier accuracy
- missed escalation count
- unsafe generation count
- pure-validation/no-action count
- ungrounded action count
- average latency
This is a deterministic project harness, not a clinical validation.