Run detection analysis on a task
detected: Whether loops were detectedpatterns: Number of loop patterns foundhighest_severity: low, medium, or hightotal: Total number of failuresexplicit: Explicit failures (errors returned)semantic: Semantic failures (wrong results)silent: Silent failures (no result)status_mismatches: Status mismatch counttotal: Total context failuresforgotten_preferences: Count of forgotten user preferencesrepeated_questions: Count of repeated questionscontradicted_facts: Count of contradicted factsviolated_constraints: Count of violated constraintslost_context: Count of lost context instanceshighest_severity: low, medium, or hightotal_calls: Total tool callsmatched: Successfully matchedunmatched: Tool calls without resultsorphaned_results: Results without matching callstask.complete(). Use this endpoint to run detection without completing the task.