{
  "method": "hindsearch",
  "seed": 42,
  "episodes": 120,
  "benchmark": "planbench-mini",
  "metrics": {
    "answer_accuracy": 1.0,
    "plan_success": 1.0,
    "joint_success": 1.0,
    "average_cost": 0.7800000000000008
  },
  "axis_metrics": {
    "cross_episode_execution": 1.0
  },
  "diagnostics": {
    "episodes": 120,
    "memory_size": 24,
    "tool_evictions": 0,
    "reused_plans": 0,
    "reasoning_steps": 0,
    "actions": 360,
    "reflections": 24,
    "skills_created": 0,
    "skills_reused": 0,
    "verification_retries": 0,
    "tree_nodes_expanded": 0,
    "search_rollouts": 0,
    "backtracks": 0,
    "tool_call_candidates": 0,
    "tool_calls_accepted": 0,
    "refinements": 0,
    "plans_created": 0,
    "worker_calls": 0,
    "agent_messages": 0,
    "critic_rounds": 0,
    "plan_explorations": 0,
    "policy_updates": 120,
    "router_calls": 0,
    "symbolic_expert_calls": 0,
    "model_matches": 0,
    "dependency_edges": 0,
    "memories_retrieved": 0,
    "reflection_syntheses": 0,
    "archival_writes": 0,
    "page_ins": 0,
    "interrupts": 0,
    "browser_queries": 0,
    "references_collected": 0,
    "rejection_candidates": 0,
    "affordance_checks": 0,
    "infeasible_skills_filtered": 0,
    "programs_generated": 0,
    "interpreter_calls": 0,
    "task_examples_retrieved": 0,
    "generation_pauses": 0,
    "task_library_updates": 0,
    "hindsight_skills": 24,
    "dense_credit_updates": 72,
    "solver_value_queries": 0,
    "turn_credit_updates": 0,
    "rollout_turns_saved": 0,
    "skill_graph_nodes": 0,
    "skill_graph_edges": 0,
    "atomic_ops_reused": 0,
    "episodic_routes": 0,
    "parametric_routes": 0,
    "memory_consolidations": 0,
    "search_queries": 360,
    "retrieved_tokens_masked": 0,
    "outcome_rewards": 0,
    "trajectory_rollouts": 0,
    "trajectory_filters": 0,
    "critic_baseline_updates": 0,
    "gradient_clips": 0,
    "echo_trap_events": 0,
    "reasoning_rewards": 0,
    "off_policy_reuses": 0,
    "leave_one_out_updates": 0,
    "per_token_clips": 0,
    "context_compressions": 0,
    "parallel_trajectory_groups": 0,
    "multi_turn_group_updates": 0,
    "simulated_user_turns": 0,
    "intent_refinements": 0,
    "real_tool_responses": 0,
    "task_completion_rewards": 0,
    "tools_exposed": 0,
    "tools_available": 0,
    "tool_exposure_reduction": 0.0,
    "cost_aware_stops": 0,
    "regret_weighted_labels": 0,
    "skill_document_updates": 0,
    "cross_task_skill_reuses": 0,
    "downstream_credit_updates": 0,
    "step_value_queries": 0,
    "step_gae_updates": 0,
    "step_sequence_ratios": 0,
    "intra_group_advantages": 0,
    "inter_group_advantages": 0,
    "transition_targets": 0,
    "transition_accuracy": 0.0,
    "reflective_groups": 0,
    "success_failure_contrasts": 0,
    "privileged_guidance_updates": 0,
    "world_rehearsals": 0,
    "rolewise_advantage_updates": 0,
    "private_rehearsals": 0,
    "recursive_belief_updates": 0,
    "pivotal_turns": 0,
    "observation_calibrations": 0,
    "scaffold_ablations": 0,
    "local_verifier_calls": 0,
    "global_verifier_calls": 0,
    "memory_operations": 0,
    "coevolution_alternations": 0,
    "router_updates": 0,
    "memory_bank_updates": 0,
    "fidelity": "mechanism reproduction on deterministic benchmark mini-suites"
  },
  "schema_version": 2,
  "manifest_ref": "agent-research:planbench-mini-seed42",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "provenance": {
    "artifact_path": "docs/agent-research/2608.01597-hindsearch/metrics/planbench-mini-seed42.json",
    "original_code_commit": "working tree",
    "dataset_fingerprint": "deterministic built-in mini-suite",
    "historical_migration": "historical-metrics-v2-2026-08-09"
  }
}