app/evaluation/explainability_eval.py

"""Explanation grounding evaluation."""

from __future__ import annotations

from app.evaluation.simulator_rollouts import run_rollouts


def explainability_eval() -> dict[str, float]:
    rows = run_rollouts(episodes=8, difficulty="medium")
    if not rows:
        return {"grounding_rate": 0.0}
    grounding_scores = [
        float((row.get("reward_breakdown", {}) or {}).get("explanation_grounding_score", 0.0))
        for row in rows
    ]
    return {"grounding_rate": round(sum(grounding_scores) / max(1, len(grounding_scores)), 6)}