{
  "model_details": {
    "provider": "inception",
    "model_requested": "mercury-2",
    "model_for_requests": "mercury-2",
    "api_base_url": "https://api.inceptionlabs.ai/v1",
    "chat_completions_endpoint": "https://api.inceptionlabs.ai/v1/chat/completions"
  },
  "run_config": {
    "task_name": "OE number",
    "prompt_layout": "standard",
    "task_description": "Determining  grammatical number (sg./pl.) in Old English nouns.",
    "tags": "open source;Old English;morphology;number"
  },
  "source_input_csv": "",
  "source_output_csv": "data/output/OE_number__inception__mercury2__2026-03-04-23-43.csv",
  "source_labels_csv": "",
  "cache_padding": {
    "enabled": false,
    "target_shared_prefix_tokens": 0,
    "calibration_shared_prefix_tokens": null,
    "target_prompt_tokens": 0,
    "calibration_prompt_tokens": null,
    "calibration_example_id": null,
    "applied_padding_tokens_estimate": 0,
    "examples_with_padding_applied": 0
  },
  "request_control_summary": {
    "configured": {},
    "attempts_total": 1384,
    "attempts_with_control_telemetry": 1214,
    "per_control": {
      "reasoning_effort": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "thinking_level": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "effort": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "verbosity": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "prompt_cache_key": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "gemini_cached_content": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "requesty_auto_cache": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      }
    }
  },
  "usage_metadata_summary": {
    "attempts_total": 1384,
    "attempts_with_usage_metadata": 1214,
    "attempts_with_cached_token_signals": 1214,
    "cached_tokens_total_estimate": 493320,
    "cache_read_tokens_total": 493320,
    "cache_write_tokens_total": 0,
    "cache_token_fields_totals": {
      "usage.cached_input_tokens": 493320
    },
    "attempts_with_gemini_cached_content_token_signals": 0,
    "gemini_cached_content_token_count_total": 0,
    "gemini_cached_content_token_fields_totals": {}
  },
  "token_usage_totals": {
    "attempts_total": 1384,
    "attempts_with_token_usage": 1214,
    "attempts_with_output_tokens": 1214,
    "attempts_with_cached_input_tokens": 1214,
    "attempts_with_thinking_tokens": 1214,
    "input_tokens_total": 831843,
    "cached_input_tokens_total": 493320,
    "non_cached_input_tokens_total": 338523,
    "output_tokens_total": 602970,
    "thinking_tokens_total": 461805,
    "output_tokens_definition": "total_tokens - prompt_tokens (or completion_tokens + thinking_tokens fallback)"
  },
  "mode": "metrics_only",
  "truth_label_count": 1200,
  "prediction_count": 1200,
  "evaluated_example_count": 1200,
  "calibration_metrics": {
    "available": true,
    "sample_count": 1200,
    "bin_count": 10,
    "ece": 0.11454166666667637,
    "mce": 0.4,
    "brier_score": 0.15047608333333334
  },
  "first_prompt_timestamp": "2026-03-04T22:49:36.410032Z",
  "last_prompt_timestamp": "2026-03-05T00:04:21.706204Z",
  "overall_time_seconds": 4485.296172,
  "overall_time_human": "1h 14m 45s",
  "truth_source": "output_csv_truth_column",
  "accuracy": 0.8308333333333333,
  "macro_precision": 0.19772743275594987,
  "macro_recall": 0.18644318673484225,
  "macro_f1": 0.1913929035433865,
  "per_label": {
    "accusative": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "dative": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "genitive": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "nominative": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "noun": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "plural": {
      "precision": 0.8125,
      "recall": 0.8540145985401459,
      "f1": 0.8327402135231317,
      "support": 274
    },
    "singular": {
      "precision": 0.9670468948035488,
      "recall": 0.8239740820734341,
      "f1": 0.889795918367347,
      "support": 926
    },
    "unclassified": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "verb": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    }
  },
  "labels": [
    "accusative",
    "dative",
    "genitive",
    "nominative",
    "noun",
    "plural",
    "singular",
    "unclassified",
    "verb"
  ],
  "label_count": 9,
  "total_examples": 1200,
  "confusion_matrix": {
    "accusative": {
      "accusative": 0,
      "dative": 0,
      "genitive": 0,
      "nominative": 0,
      "noun": 0,
      "plural": 0,
      "singular": 0,
      "unclassified": 0,
      "verb": 0
    },
    "dative": {
      "accusative": 0,
      "dative": 0,
      "genitive": 0,
      "nominative": 0,
      "noun": 0,
      "plural": 0,
      "singular": 0,
      "unclassified": 0,
      "verb": 0
    },
    "genitive": {
      "accusative": 0,
      "dative": 0,
      "genitive": 0,
      "nominative": 0,
      "noun": 0,
      "plural": 0,
      "singular": 0,
      "unclassified": 0,
      "verb": 0
    },
    "nominative": {
      "accusative": 0,
      "dative": 0,
      "genitive": 0,
      "nominative": 0,
      "noun": 0,
      "plural": 0,
      "singular": 0,
      "unclassified": 0,
      "verb": 0
    },
    "noun": {
      "accusative": 0,
      "dative": 0,
      "genitive": 0,
      "nominative": 0,
      "noun": 0,
      "plural": 0,
      "singular": 0,
      "unclassified": 0,
      "verb": 0
    },
    "plural": {
      "accusative": 3,
      "dative": 5,
      "genitive": 0,
      "nominative": 2,
      "noun": 2,
      "plural": 234,
      "singular": 26,
      "unclassified": 1,
      "verb": 1
    },
    "singular": {
      "accusative": 11,
      "dative": 9,
      "genitive": 4,
      "nominative": 9,
      "noun": 11,
      "plural": 54,
      "singular": 763,
      "unclassified": 62,
      "verb": 3
    },
    "unclassified": {
      "accusative": 0,
      "dative": 0,
      "genitive": 0,
      "nominative": 0,
      "noun": 0,
      "plural": 0,
      "singular": 0,
      "unclassified": 0,
      "verb": 0
    },
    "verb": {
      "accusative": 0,
      "dative": 0,
      "genitive": 0,
      "nominative": 0,
      "noun": 0,
      "plural": 0,
      "singular": 0,
      "unclassified": 0,
      "verb": 0
    }
  },
  "label_metrics_available": true,
  "estimated_cost_usd": 0.54919125,
  "pricing": {
    "status": "priced",
    "status_label": "priced",
    "provider_key": "inception",
    "matched_key": "mercury-2",
    "resolved_key": "mercury-2",
    "pricing_tier": "standard",
    "estimated_cost_usd": 0.54919125,
    "catalog_updated_at": "2026-05-27T15:36:49Z",
    "missing_buckets": []
  }
}