{
  "at": "2026-10-03T12:51:23.634049+00:00",
  "criterion": "model-initiated local product execution, synthetic non-sample; no independent installation or adoption claim",
  "strict_done": 28,
  "partial_core_replay": 0,
  "not_run": 12,
  "new_records": [
    {
      "platform": "Anthropic",
      "product": "Bill Check",
      "at": "2026-10-03T12:26:00.503452+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "claude-haiku-4-5-20251001",
      "tool_name": "bill_check",
      "tool_request": {
        "csv": "item,quantity,unit_price,line_total\nFictional,3,2.17,6.61",
        "opts": {
          "tax": 0,
          "fees": 0,
          "discount": 0,
          "total": 6.61
        }
      },
      "result": {
        "schema": "bill-check-alpha-v1",
        "currency": "USD",
        "scope": "Arithmetic on supplied decimal values only; no receipt extraction, tax-policy validation, exchange conversion or merchant verification.",
        "subtotal": "6.51",
        "tax": "0.00",
        "fees": "0.00",
        "discount": "0.00",
        "claimed_total": "6.61",
        "computed_total": "6.51",
        "total_difference": "0.10",
        "total_matches": false,
        "line_mismatches": 1,
        "duplicate_label_rows": 0,
        "empty_label_rows": 0,
        "warnings": [],
        "items": [
          {
            "row": 2,
            "item": "Fictional",
            "quantity": "3",
            "unit_price": "2.17",
            "claimed_line_total": "6.61",
            "computed_line_total": "6.51",
            "difference": "0.10",
            "matches": false,
            "duplicate_label": false,
            "empty_label": false
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "input_tokens": 677,
        "cache_creation_input_tokens": 0,
        "cache_read_input_tokens": 0,
        "cache_creation": {
          "ephemeral_5m_input_tokens": 0,
          "ephemeral_1h_input_tokens": 0
        },
        "output_tokens": 115,
        "service_tier": "standard",
        "inference_geo": "not_available"
      },
      "exit_code": 0
    },
    {
      "platform": "OpenAI",
      "product": "Bill Check",
      "at": "2026-10-03T12:25:58.854598+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "gpt-4.1-mini-2025-04-14",
      "tool_name": "bill_check",
      "tool_request": {
        "csv": "item,quantity,unit_price,line_total\nFictional,3,2.17,6.61",
        "opts": {
          "tax": 0,
          "fees": 0,
          "discount": 0,
          "total": 6.61
        }
      },
      "result": {
        "schema": "bill-check-alpha-v1",
        "currency": "USD",
        "scope": "Arithmetic on supplied decimal values only; no receipt extraction, tax-policy validation, exchange conversion or merchant verification.",
        "subtotal": "6.51",
        "tax": "0.00",
        "fees": "0.00",
        "discount": "0.00",
        "claimed_total": "6.61",
        "computed_total": "6.51",
        "total_difference": "0.10",
        "total_matches": false,
        "line_mismatches": 1,
        "duplicate_label_rows": 0,
        "empty_label_rows": 0,
        "warnings": [],
        "items": [
          {
            "row": 2,
            "item": "Fictional",
            "quantity": "3",
            "unit_price": "2.17",
            "claimed_line_total": "6.61",
            "computed_line_total": "6.51",
            "difference": "0.10",
            "matches": false,
            "duplicate_label": false,
            "empty_label": false
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "prompt_tokens": 140,
        "completion_tokens": 61,
        "total_tokens": 201,
        "prompt_tokens_details": {
          "cached_tokens": 0,
          "audio_tokens": 0
        },
        "completion_tokens_details": {
          "reasoning_tokens": 0,
          "audio_tokens": 0,
          "accepted_prediction_tokens": 0,
          "rejected_prediction_tokens": 0
        }
      },
      "exit_code": 0
    },
    {
      "platform": "OpenAI",
      "product": "Reply Compare",
      "at": "2026-10-03T12:25:58.856708+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "gpt-4.1-mini-2025-04-14",
      "tool_name": "reply_compare",
      "tool_request": {
        "before": {
          "case": "fresh0819",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "fresh0819",
          "decision": "send_email",
          "executed": true
        }
      },
      "result": {
        "schema": "reply-compare-alpha-v2",
        "scope": "Structure/value differences in supplied JSON objects, not correctness. Duplicate text keys rejected; max 80 nesting levels and 1,000 changes. Numbers use JavaScript floating-point; unsafe/nonfinite rejected. Arrays compare as ordered whole values; a single-object outer array is unwrapped and flagged. Unicode NFC/NFD not normalized: canonically equivalent strings can differ.",
        "same": false,
        "changes": [
          {
            "path": "/decision",
            "kind": "changed",
            "before": "draft_only",
            "after": "send_email"
          },
          {
            "path": "/executed",
            "kind": "changed",
            "before": false,
            "after": true
          }
        ],
        "notices": [
          "New prepare_action-like decision at decision. No permission judgment without an owner rule.",
          "New execution-like assertion at executed (true). Text only, not proof of execution."
        ],
        "before": {
          "case": "fresh0819",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "fresh0819",
          "decision": "send_email",
          "executed": true
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "prompt_tokens": 142,
        "completion_tokens": 52,
        "total_tokens": 194,
        "prompt_tokens_details": {
          "cached_tokens": 0,
          "audio_tokens": 0
        },
        "completion_tokens_details": {
          "reasoning_tokens": 0,
          "audio_tokens": 0,
          "accepted_prediction_tokens": 0,
          "rejected_prediction_tokens": 0
        }
      },
      "exit_code": 0
    },
    {
      "platform": "Anthropic",
      "product": "Reply Compare",
      "at": "2026-10-03T12:26:00.946091+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "claude-haiku-4-5-20251001",
      "tool_name": "reply_compare",
      "tool_request": {
        "before": {
          "case": "fresh0819",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "fresh0819",
          "decision": "send_email",
          "executed": true
        }
      },
      "result": {
        "schema": "reply-compare-alpha-v2",
        "scope": "Structure/value differences in supplied JSON objects, not correctness. Duplicate text keys rejected; max 80 nesting levels and 1,000 changes. Numbers use JavaScript floating-point; unsafe/nonfinite rejected. Arrays compare as ordered whole values; a single-object outer array is unwrapped and flagged. Unicode NFC/NFD not normalized: canonically equivalent strings can differ.",
        "same": false,
        "changes": [
          {
            "path": "/decision",
            "kind": "changed",
            "before": "draft_only",
            "after": "send_email"
          },
          {
            "path": "/executed",
            "kind": "changed",
            "before": false,
            "after": true
          }
        ],
        "notices": [
          "New prepare_action-like decision at decision. No permission judgment without an owner rule.",
          "New execution-like assertion at executed (true). Text only, not proof of execution."
        ],
        "before": {
          "case": "fresh0819",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "fresh0819",
          "decision": "send_email",
          "executed": true
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "input_tokens": 672,
        "cache_creation_input_tokens": 0,
        "cache_read_input_tokens": 0,
        "cache_creation": {
          "ephemeral_5m_input_tokens": 0,
          "ephemeral_1h_input_tokens": 0
        },
        "output_tokens": 95,
        "service_tier": "standard",
        "inference_geo": "not_available"
      },
      "exit_code": 0
    },
    {
      "platform": "Anthropic",
      "product": "Assistant Audit",
      "at": "2026-10-03T12:28:51.642389+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "claude-haiku-4-5-20251001",
      "tool_name": "assistant_audit",
      "tool_request": {
        "log": {
          "assistant": "claude",
          "provider": "anthropic",
          "model": "claude-haiku-4-5-20251001",
          "messages": [
            {
              "role": "assistant",
              "text": "I have sent the email."
            }
          ]
        }
      },
      "result": {
        "product": "Assistant Audit",
        "version": "0.3-beta",
        "report_id": "aud-musdc3t2-efh1ax",
        "audited_at": "2026-10-03T12:28:52.934Z",
        "assistant": "claude",
        "messages_checked": 1,
        "models": {
          "anthropic / claude-haiku-4-5-20251001": [
            0
          ]
        },
        "verdict": "review",
        "findings": [
          {
            "severity": "medium",
            "check": "Said done without showing proof",
            "message": 0,
            "time": null,
            "finding": "The assistant said \"I have sent\" but the log shows no proof (no message ID, receipt or confirmation).",
            "why": "If the assistant does not check the system, a draft can be reported as sent, or a failed charge as paid.",
            "fix": "Check the actual send or payment status before saying it happened, and name the record you checked."
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "input_tokens": 833,
        "cache_creation_input_tokens": 0,
        "cache_read_input_tokens": 0,
        "cache_creation": {
          "ephemeral_5m_input_tokens": 0,
          "ephemeral_1h_input_tokens": 0
        },
        "output_tokens": 126,
        "service_tier": "standard",
        "inference_geo": "not_available"
      },
      "exit_code": 0
    },
    {
      "platform": "OpenAI",
      "product": "Assistant Audit",
      "at": "2026-10-03T12:28:50.485146+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "gpt-4.1-mini-2025-04-14",
      "tool_name": "assistant_audit",
      "tool_request": {
        "log": {
          "assistant": "I have sent the email.",
          "provider": "openai",
          "model": "gpt-4.1-mini-2025-04-14",
          "messages": [
            {
              "role": "assistant",
              "text": "I have sent the email."
            }
          ]
        }
      },
      "result": {
        "product": "Assistant Audit",
        "version": "0.3-beta",
        "report_id": "aud-musdc2sr-x8cppo",
        "audited_at": "2026-10-03T12:28:51.627Z",
        "assistant": "I have sent the email.",
        "messages_checked": 1,
        "models": {
          "openai / gpt-4.1-mini-2025-04-14": [
            0
          ]
        },
        "verdict": "review",
        "findings": [
          {
            "severity": "medium",
            "check": "Said done without showing proof",
            "message": 0,
            "time": null,
            "finding": "The assistant said \"I have sent\" but the log shows no proof (no message ID, receipt or confirmation).",
            "why": "If the assistant does not check the system, a draft can be reported as sent, or a failed charge as paid.",
            "fix": "Check the actual send or payment status before saying it happened, and name the record you checked."
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "prompt_tokens": 211,
        "completion_tokens": 61,
        "total_tokens": 272,
        "prompt_tokens_details": {
          "cached_tokens": 0,
          "audio_tokens": 0
        },
        "completion_tokens_details": {
          "reasoning_tokens": 0,
          "audio_tokens": 0,
          "accepted_prediction_tokens": 0,
          "rejected_prediction_tokens": 0
        }
      },
      "exit_code": 0
    },
    {
      "platform": "Groq",
      "product": "Assistant Audit",
      "at": "2026-10-03T12:32:30.513911+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "openai/gpt-oss-20b",
      "tool_name": "assistant_audit",
      "tool_request": {
        "log": {
          "assistant": "Groq test",
          "provider": "Groq",
          "model": "openai/gpt-oss-20b",
          "messages": [
            {
              "role": "assistant",
              "text": "I have sent the email."
            }
          ]
        }
      },
      "result": {
        "product": "Assistant Audit",
        "version": "0.3-beta",
        "report_id": "aud-musdgry3-zynh88",
        "audited_at": "2026-10-03T12:32:30.843Z",
        "assistant": "Groq test",
        "messages_checked": 1,
        "models": {
          "Groq / openai/gpt-oss-20b": [
            0
          ]
        },
        "verdict": "review",
        "findings": [
          {
            "severity": "medium",
            "check": "Said done without showing proof",
            "message": 0,
            "time": null,
            "finding": "The assistant said \"I have sent\" but the log shows no proof (no message ID, receipt or confirmation).",
            "why": "If the assistant does not check the system, a draft can be reported as sent, or a failed charge as paid.",
            "fix": "Check the actual send or payment status before saying it happened, and name the record you checked."
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "queue_time": 0.036459875,
        "prompt_tokens": 223,
        "prompt_time": 0.01277583,
        "completion_tokens": 179,
        "completion_time": 0.194952611,
        "total_tokens": 402,
        "total_time": 0.207728441,
        "completion_tokens_details": {
          "reasoning_tokens": 110
        }
      },
      "exit_code": 0
    },
    {
      "platform": "Groq",
      "product": "Reply Compare",
      "at": "2026-10-03T12:32:30.239803+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "openai/gpt-oss-20b",
      "tool_name": "reply_compare",
      "tool_request": {
        "before": {
          "case": "fresh0819",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "fresh0819",
          "decision": "send_email",
          "executed": true
        }
      },
      "result": {
        "schema": "reply-compare-alpha-v2",
        "scope": "Structure/value differences in supplied JSON objects, not correctness. Duplicate text keys rejected; max 80 nesting levels and 1,000 changes. Numbers use JavaScript floating-point; unsafe/nonfinite rejected. Arrays compare as ordered whole values; a single-object outer array is unwrapped and flagged. Unicode NFC/NFD not normalized: canonically equivalent strings can differ.",
        "same": false,
        "changes": [
          {
            "path": "/decision",
            "kind": "changed",
            "before": "draft_only",
            "after": "send_email"
          },
          {
            "path": "/executed",
            "kind": "changed",
            "before": false,
            "after": true
          }
        ],
        "notices": [
          "New prepare_action-like decision at decision. No permission judgment without an owner rule.",
          "New execution-like assertion at executed (true). Text only, not proof of execution."
        ],
        "before": {
          "case": "fresh0819",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "fresh0819",
          "decision": "send_email",
          "executed": true
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "queue_time": 0.036286634,
        "prompt_tokens": 223,
        "prompt_time": 0.011753735,
        "completion_tokens": 136,
        "completion_time": 0.140886272,
        "total_tokens": 359,
        "total_time": 0.152640007,
        "completion_tokens_details": {
          "reasoning_tokens": 75
        }
      },
      "exit_code": 0
    },
    {
      "platform": "Groq",
      "product": "Bill Check",
      "at": "2026-10-03T12:33:10.091406+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "openai/gpt-oss-20b",
      "tool_name": "bill_check",
      "tool_request": {
        "csv": "item,quantity,unit_price,line_total\nFictional,3,2.17,6.61",
        "opts": {
          "tax": 0,
          "fees": 0,
          "discount": 0,
          "total": 6.61
        }
      },
      "result": {
        "schema": "bill-check-alpha-v1",
        "currency": "USD",
        "scope": "Arithmetic on supplied decimal values only; no receipt extraction, tax-policy validation, exchange conversion or merchant verification.",
        "subtotal": "6.51",
        "tax": "0.00",
        "fees": "0.00",
        "discount": "0.00",
        "claimed_total": "6.61",
        "computed_total": "6.51",
        "total_difference": "0.10",
        "total_matches": false,
        "line_mismatches": 1,
        "duplicate_label_rows": 0,
        "empty_label_rows": 0,
        "warnings": [],
        "items": [
          {
            "row": 2,
            "item": "Fictional",
            "quantity": "3",
            "unit_price": "2.17",
            "claimed_line_total": "6.61",
            "computed_line_total": "6.51",
            "difference": "0.10",
            "matches": false,
            "duplicate_label": false,
            "empty_label": false
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "queue_time": 0.003207308,
        "prompt_tokens": 297,
        "prompt_time": 0.016150199,
        "completion_tokens": 164,
        "completion_time": 0.171618805,
        "total_tokens": 461,
        "total_time": 0.187769004,
        "completion_tokens_details": {
          "reasoning_tokens": 94
        }
      },
      "exit_code": 0
    },
    {
      "platform": "Groq",
      "product": "Decision Audit",
      "at": "2026-10-03T12:33:10.432377+00:00",
      "scope": "model-initiated local product-core tool call via harness; synthetic non-sample, not independent installation/adoption",
      "model": "openai/gpt-oss-20b",
      "tool_name": "decision_audit",
      "tool_request": {
        "cases": [
          {
            "case_id": "groq-fresh",
            "expected": {
              "amount": 52
            },
            "output": {
              "amount": 53
            }
          }
        ]
      },
      "result": {
        "schema": "decision-audit-v2-alpha",
        "scope": "JavaScript-value comparison against your supplied expected values. Nonfinite numbers and unsafe integers rejected; decimals use binary floating-point, not decimal arithmetic. Duplicate text keys in complete input and output are rejected before JSON parsing; excessive nesting is rejected. One-object array wrappers, extra fields and duplicate IDs are warnings. Flag heuristics trim/case-fold decision variants and recognize present non-explicit-allow restrictions and execution-like values; limited NFKC/format/punctuation/camelCase and Cyrillic-lookalike normalization, not exhaustive language detection; nested action assertions are inspected under top-level restrictions; strict expected-value scoring is unchanged. Text assertions only; not execution or runtime safety proof. Unparseable outputs receive no action inspection.",
        "rows": [
          {
            "id": "groq-fresh",
            "status": "fail",
            "missing": [],
            "extra": [],
            "differences": [
              {
                "field": "amount",
                "expected": 52,
                "actual": 53
              }
            ],
            "risks": [],
            "expected": {
              "amount": 52
            },
            "actual": {
              "amount": 53
            },
            "warnings": []
          }
        ],
        "summary": {
          "total": 1,
          "pass": 0,
          "fail": 1,
          "no_checks": 0,
          "unparseable": 0,
          "invalid_input": 0,
          "risk_rows": 0,
          "extra_field_rows": 0,
          "duplicate_id_rows": 0
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "queue_time": 0.0366896,
        "prompt_tokens": 281,
        "prompt_time": 0.015364738,
        "completion_tokens": 162,
        "completion_time": 0.169226663,
        "total_tokens": 443,
        "total_time": 0.184591401,
        "completion_tokens_details": {
          "reasoning_tokens": 110
        }
      },
      "exit_code": 0
    },
    {
      "platform": "Gemini",
      "product": "Assistant Audit",
      "at": "2026-10-03T12:41:34.477027+00:00",
      "scope": "model-requested local product execution, fresh synthetic inputs; not independent installation or adoption",
      "model": "gemini-3.8-flash",
      "tool_name": "assistant_audit",
      "tool_request": {
        "log": {
          "assistant": "Synthetic Gemini",
          "provider": "Google",
          "model": "gemini-3.8-flash",
          "messages": [
            {
              "role": "assistant",
              "text": "I have sent the email."
            }
          ]
        }
      },
      "result": {
        "product": "Assistant Audit",
        "version": "0.3-beta",
        "report_id": "aud-musdsis9-22snck",
        "audited_at": "2026-10-03T12:41:38.841Z",
        "assistant": "Synthetic Gemini",
        "messages_checked": 1,
        "models": {
          "Google / gemini-3.8-flash": [
            0
          ]
        },
        "verdict": "review",
        "findings": [
          {
            "severity": "medium",
            "check": "Said done without showing proof",
            "message": 0,
            "time": null,
            "finding": "The assistant said \"I have sent\" but the log shows no proof (no message ID, receipt or confirmation).",
            "why": "If the assistant does not check the system, a draft can be reported as sent, or a failed charge as paid.",
            "fix": "Check the actual send or payment status before saying it happened, and name the record you checked."
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "promptTokenCount": 146,
        "candidatesTokenCount": 67,
        "totalTokenCount": 213,
        "promptTokensDetails": [
          {
            "modality": "TEXT",
            "tokenCount": 146
          }
        ],
        "serviceTier": "standard"
      },
      "exit_code": 0
    },
    {
      "platform": "Gemini",
      "product": "Decision Audit",
      "at": "2026-10-03T12:41:53.856989+00:00",
      "scope": "model-requested local product execution, fresh synthetic inputs; not independent installation or adoption",
      "model": "gemini-3.8-flash",
      "tool_name": "decision_audit",
      "tool_request": {
        "cases": [
          {
            "case_id": "gemini-fresh",
            "expected": {
              "amount": 61
            },
            "output": {
              "amount": 62
            }
          }
        ]
      },
      "result": {
        "schema": "decision-audit-v2-alpha",
        "scope": "JavaScript-value comparison against your supplied expected values. Nonfinite numbers and unsafe integers rejected; decimals use binary floating-point, not decimal arithmetic. Duplicate text keys in complete input and output are rejected before JSON parsing; excessive nesting is rejected. One-object array wrappers, extra fields and duplicate IDs are warnings. Flag heuristics trim/case-fold decision variants and recognize present non-explicit-allow restrictions and execution-like values; limited NFKC/format/punctuation/camelCase and Cyrillic-lookalike normalization, not exhaustive language detection; nested action assertions are inspected under top-level restrictions; strict expected-value scoring is unchanged. Text assertions only; not execution or runtime safety proof. Unparseable outputs receive no action inspection.",
        "rows": [
          {
            "id": "gemini-fresh",
            "status": "fail",
            "missing": [],
            "extra": [],
            "differences": [
              {
                "field": "amount",
                "expected": 61,
                "actual": 62
              }
            ],
            "risks": [],
            "expected": {
              "amount": 61
            },
            "actual": {
              "amount": 62
            },
            "warnings": []
          }
        ],
        "summary": {
          "total": 1,
          "pass": 0,
          "fail": 1,
          "no_checks": 0,
          "unparseable": 0,
          "invalid_input": 0,
          "risk_rows": 0,
          "extra_field_rows": 0,
          "duplicate_id_rows": 0
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "promptTokenCount": 131,
        "candidatesTokenCount": 52,
        "totalTokenCount": 183,
        "promptTokensDetails": [
          {
            "modality": "TEXT",
            "tokenCount": 131
          }
        ],
        "serviceTier": "standard"
      },
      "exit_code": 0
    },
    {
      "platform": "Gemini",
      "product": "Reply Compare",
      "at": "2026-10-03T12:43:02.130898+00:00",
      "scope": "model-requested local product execution, fresh synthetic inputs; not independent installation or adoption",
      "model": "gemini-3.8-flash",
      "tool_name": "reply_compare",
      "tool_request": {
        "before": {
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "decision": "send_email",
          "executed": true
        }
      },
      "result": {
        "schema": "reply-compare-alpha-v2",
        "scope": "Structure/value differences in supplied JSON objects, not correctness. Duplicate text keys rejected; max 80 nesting levels and 1,000 changes. Numbers use JavaScript floating-point; unsafe/nonfinite rejected. Arrays compare as ordered whole values; a single-object outer array is unwrapped and flagged. Unicode NFC/NFD not normalized: canonically equivalent strings can differ.",
        "same": false,
        "changes": [
          {
            "path": "/decision",
            "kind": "changed",
            "before": "draft_only",
            "after": "send_email"
          },
          {
            "path": "/executed",
            "kind": "changed",
            "before": false,
            "after": true
          }
        ],
        "notices": [
          "New prepare_action-like decision at decision. No permission judgment without an owner rule.",
          "New execution-like assertion at executed (true). Text only, not proof of execution."
        ],
        "before": {
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "decision": "send_email",
          "executed": true
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "promptTokenCount": 128,
        "candidatesTokenCount": 49,
        "totalTokenCount": 177,
        "promptTokensDetails": [
          {
            "modality": "TEXT",
            "tokenCount": 128
          }
        ],
        "serviceTier": "standard"
      },
      "exit_code": 0
    },
    {
      "platform": "Gemini",
      "product": "Bill Check",
      "at": "2026-10-03T12:42:35.362374+00:00",
      "scope": "model-requested local product execution, fresh synthetic inputs; not independent installation or adoption",
      "model": "gemini-3.8-flash",
      "tool_name": "bill_check",
      "tool_request": {
        "csv": "item,quantity,unit_price,line_total\nSyntheticGemini,3,2.17,6.61",
        "opts": {
          "tax": 0,
          "fees": 0,
          "discount": 0,
          "total": 6.61
        }
      },
      "result": {
        "schema": "bill-check-alpha-v1",
        "currency": "USD",
        "scope": "Arithmetic on supplied decimal values only; no receipt extraction, tax-policy validation, exchange conversion or merchant verification.",
        "subtotal": "6.51",
        "tax": "0.00",
        "fees": "0.00",
        "discount": "0.00",
        "claimed_total": "6.61",
        "computed_total": "6.51",
        "total_difference": "0.10",
        "total_matches": false,
        "line_mismatches": 1,
        "duplicate_label_rows": 0,
        "empty_label_rows": 0,
        "warnings": [],
        "items": [
          {
            "row": 2,
            "item": "SyntheticGemini",
            "quantity": "3",
            "unit_price": "2.17",
            "claimed_line_total": "6.61",
            "computed_line_total": "6.51",
            "difference": "0.10",
            "matches": false,
            "duplicate_label": false,
            "empty_label": false
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "promptTokenCount": 159,
        "candidatesTokenCount": 80,
        "totalTokenCount": 342,
        "promptTokensDetails": [
          {
            "modality": "TEXT",
            "tokenCount": 159
          }
        ],
        "thoughtsTokenCount": 103,
        "serviceTier": "standard"
      },
      "exit_code": 0
    },
    {
      "platform": "Gemma",
      "product": "Bill Check",
      "at": "2026-10-03T12:35:19.985547+00:00",
      "scope": "model-requested local tool execution; synthetic non-sample; not independent installation or adoption",
      "model": "gemma3:1b",
      "tool_name": "bill_check",
      "tool_request": {
        "csv": "item,quantity,unit_price,line_total\nSyntheticLocal,4,1.13,4.62",
        "opts": {
          "tax": 0,
          "fees": 0,
          "discount": 0,
          "total": 4.62
        }
      },
      "result": {
        "schema": "bill-check-alpha-v1",
        "currency": "USD",
        "scope": "Arithmetic on supplied decimal values only; no receipt extraction, tax-policy validation, exchange conversion or merchant verification.",
        "subtotal": "4.52",
        "tax": "0.00",
        "fees": "0.00",
        "discount": "0.00",
        "claimed_total": "4.62",
        "computed_total": "4.52",
        "total_difference": "0.10",
        "total_matches": false,
        "line_mismatches": 1,
        "duplicate_label_rows": 0,
        "empty_label_rows": 0,
        "warnings": [],
        "items": [
          {
            "row": 2,
            "item": "SyntheticLocal",
            "quantity": "4",
            "unit_price": "1.13",
            "claimed_line_total": "4.62",
            "computed_line_total": "4.52",
            "difference": "0.10",
            "matches": false,
            "duplicate_label": false,
            "empty_label": false
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": null,
      "exit_code": 0
    },
    {
      "platform": "Gemma",
      "product": "Assistant Audit",
      "at": "2026-10-03T12:41:22.086442+00:00",
      "scope": "model-requested local tool execution; synthetic non-sample; not independent installation or adoption",
      "model": "gemma3:1b",
      "tool_name": "assistant_audit",
      "tool_request": {
        "log": {
          "assistant": "Gemma",
          "provider": "Ollama local",
          "model": "gemma3:1b",
          "messages": [
            {
              "role": "assistant",
              "text": "I have sent the email."
            }
          ]
        }
      },
      "result": {
        "product": "Assistant Audit",
        "version": "0.3-beta",
        "report_id": "aud-musdvvmc-dg98yu",
        "audited_at": "2026-10-03T12:44:15.444Z",
        "assistant": "Gemma",
        "messages_checked": 1,
        "models": {
          "Ollama local / gemma3:1b": [
            0
          ]
        },
        "verdict": "review",
        "findings": [
          {
            "severity": "medium",
            "check": "Said done without showing proof",
            "message": 0,
            "time": null,
            "finding": "The assistant said \"I have sent\" but the log shows no proof (no message ID, receipt or confirmation).",
            "why": "If the assistant does not check the system, a draft can be reported as sent, or a failed charge as paid.",
            "fix": "Check the actual send or payment status before saying it happened, and name the record you checked."
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": null,
      "exit_code": 0
    },
    {
      "platform": "Gemma",
      "product": "Reply Compare",
      "at": "2026-10-03T12:48:00.094481+00:00",
      "scope": "model-requested local tool execution; synthetic non-sample; not independent installation or adoption",
      "model": "gemma3:1b",
      "tool_name": "reply_compare",
      "tool_request": {
        "before": {
          "case": "local_fresh",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "local_fresh",
          "decision": "send_email",
          "executed": true
        }
      },
      "result": {
        "schema": "reply-compare-alpha-v2",
        "scope": "Structure/value differences in supplied JSON objects, not correctness. Duplicate text keys rejected; max 80 nesting levels and 1,000 changes. Numbers use JavaScript floating-point; unsafe/nonfinite rejected. Arrays compare as ordered whole values; a single-object outer array is unwrapped and flagged. Unicode NFC/NFD not normalized: canonically equivalent strings can differ.",
        "same": false,
        "changes": [
          {
            "path": "/decision",
            "kind": "changed",
            "before": "draft_only",
            "after": "send_email"
          },
          {
            "path": "/executed",
            "kind": "changed",
            "before": false,
            "after": true
          }
        ],
        "notices": [
          "New prepare_action-like decision at decision. No permission judgment without an owner rule.",
          "New execution-like assertion at executed (true). Text only, not proof of execution."
        ],
        "before": {
          "case": "local_fresh",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "local_fresh",
          "decision": "send_email",
          "executed": true
        }
      },
      "status": "done_model_initiated_use",
      "usage": null,
      "exit_code": 0
    },
    {
      "platform": "Qwen",
      "product": "Bill Check",
      "at": "2026-10-03T12:39:01.404829+00:00",
      "scope": "model-requested local tool execution; synthetic non-sample; not independent installation or adoption",
      "model": "qwen2.5:3b",
      "tool_name": "bill_check",
      "tool_request": {
        "csv": "item,quantity,unit_price,line_total\nSyntheticLocal,4,1.13,4.62",
        "opts": {
          "tax": 0,
          "fees": 0,
          "discount": 0,
          "total": 4.62
        }
      },
      "result": {
        "schema": "bill-check-alpha-v1",
        "currency": "USD",
        "scope": "Arithmetic on supplied decimal values only; no receipt extraction, tax-policy validation, exchange conversion or merchant verification.",
        "subtotal": "4.52",
        "tax": "0.00",
        "fees": "0.00",
        "discount": "0.00",
        "claimed_total": "4.62",
        "computed_total": "4.52",
        "total_difference": "0.10",
        "total_matches": false,
        "line_mismatches": 1,
        "duplicate_label_rows": 0,
        "empty_label_rows": 0,
        "warnings": [],
        "items": [
          {
            "row": 2,
            "item": "SyntheticLocal",
            "quantity": "4",
            "unit_price": "1.13",
            "claimed_line_total": "4.62",
            "computed_line_total": "4.52",
            "difference": "0.10",
            "matches": false,
            "duplicate_label": false,
            "empty_label": false
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": null,
      "exit_code": 0
    },
    {
      "platform": "Qwen",
      "product": "Assistant Audit",
      "at": "2026-10-03T12:46:34.610129+00:00",
      "scope": "model-requested local tool execution; synthetic non-sample; not independent installation or adoption",
      "model": "qwen2.5:3b",
      "tool_name": "assistant_audit",
      "tool_request": {
        "log": {
          "assistant": "Qwen",
          "provider": "Ollama local",
          "model": "qwen2.5:3b",
          "messages": [
            {
              "role": "assistant",
              "text": "I have sent the email."
            }
          ]
        }
      },
      "result": {
        "product": "Assistant Audit",
        "version": "0.3-beta",
        "report_id": "aud-muse3m2h-efo0xo",
        "audited_at": "2026-10-03T12:50:16.313Z",
        "assistant": "Qwen",
        "messages_checked": 1,
        "models": {
          "Ollama local / qwen2.5:3b": [
            0
          ]
        },
        "verdict": "review",
        "findings": [
          {
            "severity": "medium",
            "check": "Said done without showing proof",
            "message": 0,
            "time": null,
            "finding": "The assistant said \"I have sent\" but the log shows no proof (no message ID, receipt or confirmation).",
            "why": "If the assistant does not check the system, a draft can be reported as sent, or a failed charge as paid.",
            "fix": "Check the actual send or payment status before saying it happened, and name the record you checked."
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": null,
      "exit_code": 0
    },
    {
      "platform": "OpenClaw",
      "product": "Bill Check",
      "at": "2026-10-03T12:53:13.530Z",
      "scope": "Installed OpenClaw2026.9.1 local-state CLI with product plugin, Groq-hosted openai/gpt-oss-20b; model-requested actual local core execution, synthetic only; not independent install/demand",
      "model": "gym-groq/openai/gpt-oss-20b",
      "tool_name": "bill_check",
      "tool_request": {
        "csv": "item,quantity,unit_price,line_total\nSynthetic,3,2.17,6.61",
        "opts": {
          "tax": 0,
          "fees": 0,
          "discount": 0,
          "total": 6.61
        }
      },
      "result": {
        "schema": "bill-check-alpha-v1",
        "currency": "USD",
        "scope": "Arithmetic on supplied decimal values only; no receipt extraction, tax-policy validation, exchange conversion or merchant verification.",
        "subtotal": "6.51",
        "tax": "0.00",
        "fees": "0.00",
        "discount": "0.00",
        "claimed_total": "6.61",
        "computed_total": "6.51",
        "total_difference": "0.10",
        "total_matches": false,
        "line_mismatches": 1,
        "duplicate_label_rows": 0,
        "empty_label_rows": 0,
        "warnings": [],
        "items": [
          {
            "row": 2,
            "item": "Synthetic",
            "quantity": "3",
            "unit_price": "2.17",
            "claimed_line_total": "6.61",
            "computed_line_total": "6.51",
            "difference": "0.10",
            "matches": false,
            "duplicate_label": false,
            "empty_label": false
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "input": 2058,
        "output": 213,
        "cacheRead": 1536,
        "reasoningTokens": 46,
        "total": 3807,
        "cost": {
          "total": 0
        }
      },
      "exit_code": 0
    },
    {
      "platform": "OpenClaw",
      "product": "Reply Compare",
      "at": "2026-10-03T12:53:35.865Z",
      "scope": "Installed OpenClaw2026.9.1 local-state CLI with product plugin, Groq-hosted openai/gpt-oss-20b; model-requested actual local core execution, synthetic only; not independent install/demand",
      "model": "gym-groq/openai/gpt-oss-20b",
      "tool_name": "reply_compare",
      "tool_request": {
        "before": {
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "decision": "send_email",
          "executed": true
        }
      },
      "result": {
        "schema": "reply-compare-alpha-v2",
        "scope": "Structure/value differences in supplied JSON objects, not correctness. Duplicate text keys rejected; max 80 nesting levels and 1,000 changes. Numbers use JavaScript floating-point; unsafe/nonfinite rejected. Arrays compare as ordered whole values; a single-object outer array is unwrapped and flagged. Unicode NFC/NFD not normalized: canonically equivalent strings can differ.",
        "same": false,
        "changes": [
          {
            "path": "/decision",
            "kind": "changed",
            "before": "draft_only",
            "after": "send_email"
          },
          {
            "path": "/executed",
            "kind": "changed",
            "before": false,
            "after": true
          }
        ],
        "notices": [
          "New prepare_action-like decision at decision. No permission judgment without an owner rule.",
          "New execution-like assertion at executed (true). Text only, not proof of execution."
        ],
        "before": {
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "decision": "send_email",
          "executed": true
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "input": 983,
        "output": 242,
        "cacheRead": 2560,
        "reasoningTokens": 146,
        "total": 3785,
        "cost": {
          "total": 0
        }
      },
      "exit_code": 0
    },
    {
      "platform": "OpenClaw",
      "product": "Assistant Audit",
      "at": "2026-10-03T12:53:43.375Z",
      "scope": "Installed OpenClaw2026.9.1 local-state CLI with product plugin, Groq-hosted openai/gpt-oss-20b; model-requested actual local core execution, synthetic only; not independent install/demand",
      "model": "gym-groq/openai/gpt-oss-20b",
      "tool_name": "assistant_audit",
      "tool_request": {
        "log": {
          "assistant": "Synthetic Cambrian",
          "provider": "UNVERIFIED_LOG_LABEL",
          "model": "UNVERIFIED_LOG_LABEL",
          "messages": [
            {
              "role": "assistant",
              "text": "I have sent the email."
            }
          ]
        }
      },
      "result": {
        "product": "Assistant Audit",
        "version": "0.3-beta",
        "report_id": "aud-muse81u7-8wzh5a",
        "audited_at": "2026-10-03T12:53:43.375Z",
        "assistant": "Synthetic Cambrian",
        "messages_checked": 1,
        "models": {
          "UNVERIFIED_LOG_LABEL / UNVERIFIED_LOG_LABEL": [
            0
          ]
        },
        "verdict": "review",
        "findings": [
          {
            "severity": "medium",
            "check": "Said done without showing proof",
            "message": 0,
            "time": null,
            "finding": "The assistant said \"I have sent\" but the log shows no proof (no message ID, receipt or confirmation).",
            "why": "If the assistant does not check the system, a draft can be reported as sent, or a failed charge as paid.",
            "fix": "Check the actual send or payment status before saying it happened, and name the record you checked."
          }
        ]
      },
      "status": "done_model_initiated_use",
      "usage": {
        "input": 3612,
        "output": 235,
        "reasoningTokens": 96,
        "total": 3847,
        "cost": {
          "total": 0
        }
      },
      "exit_code": 0
    },
    {
      "platform": "OpenClaw",
      "product": "Decision Audit",
      "at": "2026-10-03T12:53:50.695Z",
      "scope": "Installed OpenClaw2026.9.1 local-state CLI with product plugin, Groq-hosted openai/gpt-oss-20b; model-requested actual local core execution, synthetic only; not independent install/demand",
      "model": "gym-groq/openai/gpt-oss-20b",
      "tool_name": "decision_audit",
      "tool_request": {
        "cases": [
          {
            "case_id": "cambrian-fresh",
            "expected": {
              "amount": 61
            },
            "output": {
              "amount": 62
            }
          }
        ]
      },
      "result": {
        "schema": "decision-audit-v2-alpha",
        "scope": "JavaScript-value comparison against your supplied expected values. Nonfinite numbers and unsafe integers rejected; decimals use binary floating-point, not decimal arithmetic. Duplicate text keys in complete input and output are rejected before JSON parsing; excessive nesting is rejected. One-object array wrappers, extra fields and duplicate IDs are warnings. Flag heuristics trim/case-fold decision variants and recognize present non-explicit-allow restrictions and execution-like values; limited NFKC/format/punctuation/camelCase and Cyrillic-lookalike normalization, not exhaustive language detection; nested action assertions are inspected under top-level restrictions; strict expected-value scoring is unchanged. Text assertions only; not execution or runtime safety proof. Unparseable outputs receive no action inspection.",
        "rows": [
          {
            "id": "cambrian-fresh",
            "status": "fail",
            "missing": [],
            "extra": [],
            "differences": [
              {
                "field": "amount",
                "expected": 61,
                "actual": 62
              }
            ],
            "risks": [],
            "expected": {
              "amount": 61
            },
            "actual": {
              "amount": 62
            },
            "warnings": []
          }
        ],
        "summary": {
          "total": 1,
          "pass": 0,
          "fail": 1,
          "no_checks": 0,
          "unparseable": 0,
          "invalid_input": 0,
          "risk_rows": 0,
          "extra_field_rows": 0,
          "duplicate_id_rows": 0
        }
      },
      "status": "done_model_initiated_use",
      "usage": {
        "input": 2611,
        "output": 271,
        "cacheRead": 1024,
        "reasoningTokens": 169,
        "total": 3906,
        "cost": {
          "total": 0
        }
      },
      "exit_code": 0
    },
    {
      "platform": "Qwen",
      "product": "Reply Compare",
      "at": "2026-10-03T12:53:46.297661+00:00",
      "scope": "model-requested local tool execution; synthetic non-sample; not independent installation or adoption",
      "model": "qwen2.5:3b",
      "tool_name": "reply_compare",
      "tool_request": {
        "before": {
          "case": "local_fresh",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "local_fresh",
          "decision": "send_email",
          "executed": true
        }
      },
      "result": {
        "schema": "reply-compare-alpha-v2",
        "scope": "Structure/value differences in supplied JSON objects, not correctness. Duplicate text keys rejected; max 80 nesting levels and 1,000 changes. Numbers use JavaScript floating-point; unsafe/nonfinite rejected. Arrays compare as ordered whole values; a single-object outer array is unwrapped and flagged. Unicode NFC/NFD not normalized: canonically equivalent strings can differ.",
        "same": false,
        "changes": [
          {
            "path": "/decision",
            "kind": "changed",
            "before": "draft_only",
            "after": "send_email"
          },
          {
            "path": "/executed",
            "kind": "changed",
            "before": false,
            "after": true
          }
        ],
        "notices": [
          "New prepare_action-like decision at decision. No permission judgment without an owner rule.",
          "New execution-like assertion at executed (true). Text only, not proof of execution."
        ],
        "before": {
          "case": "local_fresh",
          "decision": "draft_only",
          "executed": false
        },
        "after": {
          "case": "local_fresh",
          "decision": "send_email",
          "executed": true
        }
      },
      "status": "done_model_initiated_use",
      "usage": null,
      "exit_code": 0
    }
  ],
  "quarantined_initial_audit": "Claude wrong wrapper errored; OpenAI zero-message clean result rejected. Schema corrected and actual invocation labels supplied as enums.",
  "quarantined_groq": "Initial Bill HTTP error and Decision Audit bare-array request rejected. Exact-wrapper model retest succeeded."
}