Spaces:
Sleeping
Sleeping
| """Deterministic state verifier — no LLM-as-judge.""" | |
| from typing import Any, Dict | |
| from server.tasks.triage_task import TriageTask | |
| from server.tasks.incident_routing_task import IncidentRoutingTask | |
| from server.tasks.release_notes_task import ReleaseNotesTask | |
| from server.tasks.dep_update_task import DepUpdateTask | |
| _GRADERS = { | |
| "triage": TriageTask(), | |
| "incident_routing": IncidentRoutingTask(), | |
| "release_notes": ReleaseNotesTask(), | |
| "dep_update": DepUpdateTask(), | |
| } | |
| def verify(final_state: Dict[str, Any], org_config: Dict[str, Any], scenario: Dict[str, Any]) -> Dict[str, Any]: | |
| task_type = scenario.get("type", "") | |
| grader = _GRADERS.get(task_type) | |
| if not grader: | |
| return {"score": 0.0, "error": f"Unknown task type: {task_type}"} | |
| score = grader.grade(final_state, org_config, scenario) | |
| return {"score": score, "task_type": task_type} | |