{"schemaVersion":"agentdepot.eval_plan.v1","asOf":"2026-10-28","storesIpAddresses":false,"query":"multi-turn","matchMode":"matched_task","taskType":"multi-turn","label":"Multi-Turn Customer Support, Voice & Policy-Bound Persona Agents","recommendedBenchmarks":[{"id":"tau-bench","name":"τ-bench (Tau-bench)","officialUrl":"https://github.com/sierra-research/tau-bench","graderType":"Final DB state + policy pass^k check"}],"depotBenchmarkSuite":{"type":"ops_workflow","tasksApi":"/api/benchmark/tasks?type=ops_workflow","submitApi":"/api/benchmark/runs"},"stackPicks":{"evalHarness":{"jtbdId":"synthetic-persona-pre-flight","pickProviderId":"braintrust","coPickProviderId":"promptfoo","runnersUpIds":["promptfoo","patronus"]},"sandboxOrRunner":{"jtbdId":"agent-trajectory-tool-use-evals","pickProviderId":"braintrust","coPickProviderId":"langsmith","runnersUpIds":["langsmith","arize-phoenix"]},"ciGate":{"jtbdId":"agent-regression-leaderboards","pickProviderId":"braintrust","coPickProviderId":"langfuse","runnersUpIds":["langfuse","weights-biases"]},"judgeCalibration":{"jtbdId":"human-labeled-eval-review-queues","pickProviderId":"langsmith","coPickProviderId":"langfuse","runnersUpIds":["langfuse","braintrust"]}},"targetMetrics":["pass^5 policy-compliant resolution","zero unauthorized state mutations","turn count to resolution","costPerSolvedTaskUsd"],"antiContaminationRule":"Report pass^5 (all 5 persona phrasing trials succeeding) alongside pass@1; single-trial scores hide brittle policy compliance.","nextCalls":["/api/decide?jtbd=synthetic-persona-pre-flight","/api/benchmark/tasks?type=ops_workflow","/api/playbooks/agent-evals"]}