{
  "analysis_id": "taipei-border-premium",
  "created_utc": "2026-05-14T09:02:03+00:00",
  "purpose": "Public minimal result-reproduction bundle for the 500m Taipei/New Taipei local conditional premium regression.",
  "public_files": [
    {
      "path": "model_input.csv.gz",
      "role": "Statsmodels-ready model frame for the main 500m property/time/segment regression.",
      "rows": 1322,
      "columns": [
        "log_unit_price_ping",
        "is_taipei",
        "dist_100m",
        "area_building_ping",
        "building_age",
        "floor",
        "total_floor",
        "has_parking_int",
        "building_type",
        "main_use",
        "material",
        "year_month",
        "boundary_segment_id"
      ]
    },
    {
      "path": "expected_results.json",
      "role": "Expected coefficient transform, confidence interval, sample counts, and QA gate pass expectations."
    },
    {
      "path": "reproduction-manifest.json",
      "role": "Bundle inventory, source artifact origins, and privacy exclusions."
    }
  ],
  "source_artifact_origins": {
    "model_input.csv.gz": "data/processed/formal_analysis_sample_history.parquet",
    "expected_results.json": [
      "outputs/estimation/final_estimate_summary.json",
      "outputs/qa/qa_gate_summary.json",
      "outputs/estimation/model_results_by_bandwidth.csv"
    ],
    "qa_context": [
      "outputs/qa/formal_robustness_report.json",
      "outputs/qa/formal_placebo_report.json"
    ]
  },
  "privacy_notes": [
    "The bundle is derived from the formal matched analysis sample, not from the full geocoded panel.",
    "The CSV keeps only variables needed to rerun the published statsmodels regression.",
    "Address strings, address keys, raw transaction dates, projected coordinates, and geocode QA fields are excluded.",
    "Fixed effects are limited to year_month and boundary_segment_id; these identify time and reviewed border segment only, not individual addresses.",
    "Rows still contain transaction attributes and segment/month cells, so they should be treated as a reproduction dataset rather than an anonymized microdata release."
  ],
  "not_included_fields": [
    "address_raw",
    "address_key",
    "tx_date",
    "source_season",
    "city",
    "district",
    "x_3826",
    "y_3826",
    "latitude",
    "longitude",
    "lat",
    "lon",
    "n_address_rows",
    "n_coordinate_pairs",
    "coordinate_spread_m",
    "coordinate_status",
    "geocode_status",
    "distance_to_border_m",
    "signed_distance_m",
    "boundary_position_m",
    "formal_match_id",
    "formal_candidate_id",
    "private_repo_url",
    "local_filesystem_path"
  ],
  "reproduction_formula": "log_unit_price_ping ~ is_taipei + dist_100m + is_taipei:dist_100m + area_building_ping + I(area_building_ping ** 2) + building_age + floor + total_floor + C(building_type) + C(main_use) + C(material) + C(year_month) + C(boundary_segment_id) + has_parking_int",
  "expected_result_file": "expected_results.json"
}