from typing import Any, Dict from server.tasks.base_task import BaseTask class IncidentRoutingTask(BaseTask): def grade(self, final_state: Dict[str, Any], org_config: Dict[str, Any], scenario: Dict[str, Any]) -> float: score = 0.0 expected = scenario["expected"] msgs = final_state["chat"]["messages_posted_this_episode"] tickets = final_state["ticketing"]["tickets"] new_tickets = [t for t in tickets if t.get("created_this_episode")] correct_msgs = [m for m in msgs if m.get("channel") == expected.get("channel")] wrong_msgs = [m for m in msgs if m.get("channel") != expected.get("channel")] if correct_msgs: score += 0.30 if new_tickets: t = new_tickets[0] if t.get("label") == expected.get("label"): score += 0.20 if t.get("priority") == expected.get("priority"): score += 0.20 if t.get("assigned_team") == expected.get("team"): score += 0.15 if t.get("linked_pr"): score += 0.05 score -= 0.15 * len(wrong_msgs) return max(0.0, min(1.0, score))