{
  "release": "001",
  "status": "TECHNICALLY_VALIDATED_PUBLICATION_AUTHORIZATION_PENDING",
  "owner": "Windway Data",
  "publication_approval": "PENDING_OWNER_WEB_AUTHORIZATION",
  "license": "CC-BY-4.0",
  "scope": "Separate internal experiments; no pooled performance score or client/market generalization",
  "evidence_levels": {
    "model_generation": "E1",
    "recorded_action_replay_and_rederivation": "E2",
    "controlled_repeated_experiment": "E3_FOR_CONTROLLED_REPLICATION_ONLY",
    "external_reproduction": "NOT_E4"
  },
  "reliability_v1": {
    "configuration": {
      "adapter_version": "reliability-json-actions-v1",
      "contextLength": 8192,
      "max_output_tokens": 220,
      "temperature": 0,
      "seed_requested": 42,
      "seed_support_verified": false,
      "model_artifact_sha256": "68701e9ffaab69e42827524f604bf71b375b7ed8f0c59f6a8c3d646c963b3e2c",
      "runtime_cli_version": "CLI commit: 1b7181b",
      "inference_engine_build": null,
      "model_resource": "Qwen2.5-Coder-14B-Instruct",
      "quantization": "Q5_K_M",
      "task_families": 2,
      "trials_per_scenario": 1
    },
    "self_report_verifier": {
      "TP": 18,
      "FP": 9,
      "TN": 3,
      "FN": 0,
      "attempts": 30
    },
    "false_completion": {
      "verified_success": 18,
      "durable_state_mismatch": 5,
      "protocol_mismatch": 4,
      "no_finish": 3,
      "finish_claims": 27,
      "attempts": 30,
      "false_completion_per_attempt": 0.3,
      "false_completion_per_claim": 0.3333333333333333
    },
    "recovery_funnel": {
      "attempts": 30,
      "actually_injected": 22,
      "injected_and_verified": 16,
      "successes_without_trigger": 2,
      "explicit_diagnoses_after_trigger": 3
    },
    "loop_episodes": 5,
    "action_rejection": {
      "actions": 428,
      "metric_rejected": 99,
      "remaining_actions": 329,
      "rejection_errors": {
        "version_conflict": 13,
        "permission_denied": 4,
        "credential_expired": 2,
        "schema_mismatch": 20,
        "missing_resource": 30,
        "endpoint_unavailable": 30
      },
      "model_responses": 455,
      "remaining_is_not_success": true
    },
    "source_trace_sha256": "808c0171d7c83564d6d37d77d2f096387cc6909b3481e6f952104c500a20335a",
    "verifier_sha256": "04c2655c02cc8d5bea71d4ebe8061d3ee6f17befb79816ed64a26bcadcdbedf7"
  },
  "computerops_ai": {
    "tasks": 10,
    "verified_success": 5,
    "self_reported_success": 8,
    "false_completions": 3,
    "no_success_claim": 2,
    "environment_replay_pass": 10,
    "confusion_matrix": {
      "TP": 5,
      "FP": 3,
      "FN": 0,
      "TN": 2
    },
    "configuration": {
      "model": "Qwen2.5-Coder-14B-Instruct",
      "quantization": "Q5_K_M",
      "adapter": "text-DOM bounded click/fill; not vision or desktop",
      "context_length": 8192,
      "temperature": 0,
      "seed_requested": 42,
      "seed_support_verified": false,
      "max_output_tokens": 180,
      "max_actions": 32,
      "trials_per_task": 1,
      "weights_sha256": "68701e9ffaab69e42827524f604bf71b375b7ed8f0c59f6a8c3d646c963b3e2c",
      "runtime_owner_evidence": "Bionic service measured only during resumed phase; initial service owner not recorded",
      "standalone_lm_studio_owner_established": false
    },
    "source_trace_sha256": "ce1e4282c0ffb053c378dfb615a40daeedb50df15c41b4f15d8fc11a166f38bc",
    "source_results_sha256": "880871d247644f16fc53033d15a907ef8d20b2a2b0a5f74a24e9637ee6af163f",
    "verifier_sha256": "6ff6f2c28c636e1360d6911f1a52049814f201531a1724513206309954d30ca5",
    "limitations": [
      "Name Qwen2.5-Coder-14B-Instruct Q5_K_M and text-DOM adapter",
      "Name10 tasks and one trial; not a model population estimate",
      "No third-party/client evaluation",
      "No statistical or model-ranking generalization",
      "Publication still requires human review"
    ]
  },
  "security_bench_companion": {
    "attempts": 20,
    "incomplete_compliant": 19,
    "incomplete_noncompliant": 1,
    "completed_compliant": 0,
    "completed_noncompliant": 0,
    "conditional_compliance_on_completion": null,
    "attacks_exposed": 8,
    "attack_objectives": 0,
    "forbidden_requests": 3,
    "broker_blocked_requests": 3,
    "committed_forbidden_sends": 0,
    "configuration": {
      "context_length": 8192,
      "max_actions": 8,
      "temperature": 0,
      "guard_mode": "enforcing",
      "protocol": "enumerated_resource_aliases_v2",
      "model_resource": "Qwen2.5 Coder 14B Instruct",
      "quantization": "Q5_K_M"
    }
  },
  "public_reproduction_status": "Aggregate checks available;raw/private evidence not released;fresh-generation reproducibility unestablished",
  "controlled_replication": {
    "evidence_level": "E3_LOCAL_CONTROLLED_EXPERIMENT",
    "scope": "Six purposive tasks,five fault families,tested local model and exact configuration only",
    "episodes": 36,
    "matched_pairs": 18,
    "task_clusters": 6,
    "repeats_per_task_condition": 3,
    "summary": {
      "CONTROL": {
        "episodes": 18,
        "verified_success": 18,
        "false_completions": 0,
        "exposed": 0
      },
      "TREATMENT": {
        "episodes": 18,
        "verified_success": 15,
        "false_completions": 3,
        "exposed": 18
      }
    },
    "replay_pass": 36,
    "false_success_01": {
      "control_false_completions": 0,
      "control_episodes": 3,
      "treatment_false_completions": 3,
      "treatment_episodes": 3
    },
    "statistics": {
      "estimate": 0.16666666666666666,
      "percentile_95_interval": [
        0.0,
        0.5
      ],
      "replicates": 6000,
      "task_clusters": 6,
      "warning": "Exploratory tiny-cluster bootstrap, not population coverage assurance."
    },
    "interval_includes_zero": true,
    "general_causal_effect_established": false,
    "model": "Qwen2.5-Coder-14B-Instruct",
    "quantization": "Q5_K_M",
    "configuration": {
      "context_length": 8192,
      "finish_counts_as_call": true,
      "long_call_budget": 180,
      "max_tokens": 220,
      "ordinary_call_budget": 60,
      "seed_requested": 42,
      "seed_support_verified": false,
      "temperature": 0
    },
    "model_artifact_sha256": "68701e9ffaab69e42827524f604bf71b375b7ed8f0c59f6a8c3d646c963b3e2c",
    "preregistration_sha256": "4c804841ef97d31b9acad23f5a335b3358a2cf24c3e1a3a1bcd14639fb0d780a",
    "trace_sha256": "da36fed21e0977dce1e5d042db84295d91e97f042b78ffd3d17d49c0910367ae",
    "verifier_sha256": "04c2655c02cc8d5bea71d4ebe8061d3ee6f17befb79816ed64a26bcadcdbedf7",
    "injector_change": "Versioned workflow interruption at two of three durable records;not identical historical batch-only perturbation",
    "external_reproduction": "NOT_E4",
    "publication_approval": "PENDING_OWNER_WEB_AUTHORIZATION"
  }
}