{"_id": {"$oid": "69d4eb12b877594bc5718d0b"}, "test_id": "T0602", "benchmark": "personnel_cards", "date": "2026-03-23", "tags": {"document-type": ["index-card"], "writing": ["handwritten", "typed", "printed"], "century": [20], "layout": ["table", "form"], "task": ["transcription", "document-understanding", "data-correction"], "language": ["de", "fr"]}, "contributors": ["tabea_wullschleger", "maximilian_hindermann"], "hidden": false, "config": {"provider": "mistral", "model": "mistral-medium-2508", "dataclass": "Table", "temperature": 0.0, "role_description": "You are a historian specializing in 20th century personnel records and administrative documents", "prompt_file": "prompt.txt", "legacy_test": false}, "prompt": "Extract all information from this personnel card table and return it as a structured JSON object. The card contains rows documenting employment history with positions, locations, salary information, dates, and remarks.\n\nREQUIRED JSON STRUCTURE:\n```json\n{\n  \"rows\": [\n    {\n      \"row_number\": 1,\n      \"dienstliche_stellung\": {\n        \"diplomatic_transcript\": \"exact text from card\",\n        \"interpretation\": \"expanded/standardized form or null\",\n        \"is_crossed_out\": false\n      },\n      \"dienstort\": {\n        \"diplomatic_transcript\": \"exact text from card\",\n        \"interpretation\": \"expanded/standardized form or null\",\n        \"is_crossed_out\": false\n      },\n      \"gehaltsklasse\": {\n        \"diplomatic_transcript\": \"exact text from card\",\n        \"interpretation\": \"expanded/standardized form or null\",\n        \"is_crossed_out\": false\n      },\n      \"jahresgehalt_monatsgehalt_taglohn\": {\n        \"diplomatic_transcript\": \"exact text from card\",\n        \"interpretation\": \"standardized numeric form or null\",\n        \"is_crossed_out\": false\n      },\n      \"datum_gehalts\u00e4nderung\": {\n        \"diplomatic_transcript\": \"exact text from card\",\n        \"interpretation\": \"YYYY-MM-DD format or null\",\n        \"is_crossed_out\": false\n      },\n      \"bemerkungen\": {\n        \"diplomatic_transcript\": \"exact text from card\",\n        \"interpretation\": \"expanded/standardized form or null\",\n        \"is_crossed_out\": false\n      }\n    }\n  ]\n}\n```\n\nCOLUMN DEFINITIONS:\n1. **dienstliche_stellung**: Official position or job title (e.g., \"Assistent\", \"Professor\", \"Sekret\u00e4r\")\n2. **dienstort**: Place of service or work location (e.g., \"Basel\", \"Z\u00fcrich\")\n3. **gehaltsklasse**: Salary class or grade (e.g., \"III\", \"IV\", roman numerals or numbers)\n4. **jahresgehalt_monatsgehalt_taglohn**: Salary amount (annual/monthly/daily wage)\n5. **datum_gehalts\u00e4nderung**: Date of salary change or effective date\n6. **bemerkungen**: Remarks, notes, or additional comments\n\nFIELD EXTRACTION RULES:\n\n**diplomatic_transcript** (REQUIRED for all fields):\n- Transcribe EXACTLY as written on the card\n- Include all abbreviations, punctuation, and formatting as they appear\n- Preserve original capitalization and spacing\n- Include currency symbols and separators (e.g., \"Fr. 2'400.-\", \"3.700.-\")\n- For dates, copy the exact format (e.g., \"1. Jan. 1946\", \"1.April 1945\")\n- Use empty string \"\" for empty cells\n- Be sure to escape ditto marks (repetition marks such as `\"` indicating \"same as above\")\n- DO NOT expand abbreviations or standardize formats\n- Do not transcribe checkmarks\n\n**interpretation** (OPTIONAL - use null if not applicable):\n- For ditto marks (repetition marks such as `\"`): Replace with the actual repeated value from the previous row in the same column\n  - For example, if previous row's dienstliche_stellung is \"Hilfsleiterin\" and current row has `\"`, interpret as \"Hilfsleiterin\"\n  - Do NOT add explanatory text like \"wie oben\" or similar - just provide the repeated value\n- Never expand abbreviations\n- For dates: Convert to ISO format YYYY-MM-DD\n  - \"1. Jan. 1946\" \u2192 \"1946-01-01\"\n  - \"1.April 1945\" \u2192 \"1945-04-01\"\n- For salary amounts: Extract numeric value only (remove currency symbols, separators)\n  - \"Fr. 2'400.-\" \u2192 \"2400\"\n  - \"3.700.-\" \u2192 \"3700\"\n  - \"5.094.-\" \u2192 \"5094\"\n  \"6,30.-\" \u2192 \"6.3\"\n- For salary class: Convert roman numerals to arabic if clear\n  - \"III\" \u2192 \"3\"\n  - \"IV\" \u2192 \"4\"\n- Do not convert roman numerals to arabic if not salary, date, or salary class\n- Use null if no interpretation/expansion is needed or if the field is empty\n- \"+ {word}\" where {word} is a word does not need interpretation\n\n**is_crossed_out** (REQUIRED for all fields):\n- Set to true if text in this field is crossed out, struck through, or deleted\n- Set to false if text is normal (not crossed out)\n- Empty cells should have is_crossed_out: false\n\nROW HANDLING:\n- Number rows sequentially starting from 1 (top to bottom)\n- Include ALL rows that have ANY content in ANY column\n- Empty rows (all cells empty) should be omitted\n- A row with only one filled cell should still be included\n\nIMPORTANT NOTES:\n- Return ONLY the JSON object, no additional text or explanation\n- Every field must have all three sub-fields: diplomatic_transcript, interpretation, is_crossed_out\n- diplomatic_transcript must never be null (use empty string \"\" for empty cells)\n- interpretation can be null when no expansion/standardization applies\n- Process the entire table from top to bottom\n- Maintain consistent row numbering throughout\n", "results": {"text": "", "model": "mistral-medium-2508", "provider": "mistral", "finish_reason": "error", "usage": {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0}, "duration": 2162.00714635849, "timestamp": "2026-03-23T15:27:57.665834", "score": {"f1_score": 0.0, "precision": 0.0, "recall": 0.0, "true_positives": 0, "false_positives": 0, "false_negatives": 0, "field_scores": {"": {"response": null, "ground_truth": null, "score": 1.0}}, "total_fields": 1}}, "scoring": {"f1_micro": 0.00860832137733142, "f1_macro": 0.0013114754098360656, "micro_precision": 0.044444444444444446, "micro_recall": 0.004765687053216839, "total_instances": 61, "total_tp": 12, "total_fp": 258, "total_fn": 2506, "cost_summary": {"total_input_tokens": 1854, "total_output_tokens": 5036, "total_tokens": 6890, "input_cost_usd": 0.0007416, "output_cost_usd": 0.010072, "total_cost_usd": 0.0108136}}, "normalized_score": 0.860832137733142}