# Search Ranking Regression Report Local source edition 1.0.0. Compare two search result exports against judged document IDs. Find which queries lost recall, first-hit rank or correct empty-result behavior. ## Run locally Extract the complete ZIP. Use Python 3.12 from a terminal in the extracted directory. The tool uses the Python standard library. python run.py example.json results Use a new output folder for each run. Start with example.json and replace its synthetic values with your own data. Preserve field names, types and units. Files and reports are processed on your computer. ## Inputs Supply cutoff k (1–100) and up to 1000 queries. Each needs a unique id and three arrays of distinct document IDs, relevant, baseline and candidate. Arrays can contain 0–100 IDs. Put results in rank order and use anonymized identifiers. An empty relevant list specifically means the correct response is empty. ## Results QUERIES.csv compares reciprocal rank and recall at k for each judged query. LOSSES.csv includes any decline in either metric or a new false positive on an expected-empty query. The summary averages over queries with at least one relevant document. When both revisions return results for an expected-empty query, both fail that test and change is unchanged. The results folder includes summary.json and the output files produced by this input. The sample produces: QUERIES.csv, LOSSES.csv. ## Español · Informe de regresiones de búsqueda Compara dos exportaciones con resultados evaluados e identifica consultas que pierden cobertura o posición. Extrae el ZIP completo y abre una terminal en esa carpeta con Python 3.12. El programa usa la biblioteca estándar de Python. Ejecuta: python run.py example.json resultados Copia example.json para preparar tus datos. Conserva los nombres y tipos de los campos, usa una carpeta de salida nueva y consulta summary.json junto con los archivos generados. La descripción detallada de los campos está en inglés. The included LICENSE covers use and modification for your own personal and commercial projects. This edition includes its source and can run locally after setup.