{
  "model_details": {
    "provider": "vertex",
    "model_requested": "gemini-3-flash-preview",
    "model_for_requests": "google/gemini-3-flash-preview",
    "api_base_url": "https://aiplatform.googleapis.com/v1/projects/gen-lang-client-0461875984/locations/global/endpoints/openapi",
    "chat_completions_endpoint": "https://aiplatform.googleapis.com/v1/projects/gen-lang-client-0461875984/locations/global/endpoints/openapi/chat/completions",
    "vertex_cache_model": "projects/gen-lang-client-0461875984/locations/global/publishers/google/models/gemini-3-flash-preview"
  },
  "cache_padding": {
    "enabled": true,
    "target_shared_prefix_tokens": 1200,
    "calibration_shared_prefix_tokens": 496,
    "target_prompt_tokens": 1200,
    "calibration_prompt_tokens": 496,
    "calibration_example_id": "water-3",
    "applied_padding_tokens_estimate": 302,
    "examples_with_padding_applied": 1200
  },
  "request_control_summary": {
    "configured": {
      "thinking_level": "medium"
    },
    "attempts_total": 1202,
    "attempts_with_control_telemetry": 1200,
    "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": "medium",
        "requested_attempts": 1200,
        "sent_attempts": 1200,
        "accepted_attempts": 1200,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": 1.0,
        "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": 1202,
    "attempts_with_usage_metadata": 1200,
    "attempts_with_cached_token_signals": 0,
    "cached_tokens_total_estimate": 0,
    "cache_read_tokens_total": 0,
    "cache_write_tokens_total": 0,
    "cache_token_fields_totals": {},
    "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": 1202,
    "attempts_with_token_usage": 1200,
    "attempts_with_output_tokens": 1200,
    "attempts_with_cached_input_tokens": 0,
    "attempts_with_thinking_tokens": 1200,
    "input_tokens_total": 1621959,
    "cached_input_tokens_total": 0,
    "non_cached_input_tokens_total": 1621959,
    "output_tokens_total": 1021633,
    "thinking_tokens_total": 849822,
    "output_tokens_definition": "total_tokens - prompt_tokens (or completion_tokens + thinking_tokens fallback)"
  },
  "truth_label_count": 1200,
  "prediction_count": 1200,
  "evaluated_example_count": 1200,
  "calibration_metrics": {
    "available": true,
    "sample_count": 1200,
    "bin_count": 10,
    "ece": 0.01512500000000378,
    "mce": 0.85,
    "brier_score": 0.01994375
  },
  "first_prompt_timestamp": "2026-03-10T21:35:33.887375Z",
  "last_prompt_timestamp": "2026-03-10T22:17:10.737263Z",
  "overall_time_seconds": 2496.849888,
  "overall_time_human": "41m 37s",
  "accuracy": 0.9783333333333334,
  "macro_precision": 0.9739820643686303,
  "macro_recall": 0.9641184909586795,
  "macro_f1": 0.968936678614098,
  "per_label": {
    "plural": {
      "precision": 0.9661654135338346,
      "recall": 0.9379562043795621,
      "f1": 0.9518518518518518,
      "support": 274
    },
    "singular": {
      "precision": 0.9817987152034261,
      "recall": 0.990280777537797,
      "f1": 0.9860215053763441,
      "support": 926
    }
  },
  "confusion_matrix": {
    "plural": {
      "plural": 257,
      "singular": 17
    },
    "singular": {
      "plural": 9,
      "singular": 917
    }
  },
  "labels": [
    "plural",
    "singular"
  ],
  "total_examples": 1200,
  "label_metrics_available": true,
  "run_config": {
    "task_name": "OE number",
    "prompt_layout": "standard",
    "task_description": "Determining  grammatical number (sg./pl.) in Old English nouns.",
    "tags": "Old English;morphology;number"
  },
  "source_input_csv": "",
  "source_output_csv": "data/output/OE_number__vertex__gemini3flashpreview__2026-03-10-22-35.csv",
  "source_labels_csv": ""
}