Data and code for "Correcting air quality forecasts beyond a fixed station offset: the case of CAMS-MOS"

Data and code accompanying the manuscript "Exploitable error survives CAMS-MOS calibration" (Nir Levy, independent researcher, Berlin), an evaluation of what the operational CAMS-MOS calibration leaves in the residuals of the Copernicus regional air quality ensemble, over four pollutants (NO2, O3, PM10, PM2.5) and four operational forecast slots at about 500-1,100 European monitoring stations per pollutant. The deposit contains every table from which a number in the article is drawn: the frozen metrics tables of the headline evaluation and its robustness checks for both panel designs, the static-bound, efficiency and oracle tables, a 5,000-resample bootstrap check, and the competitor, nested and placebo-calibrated tables produced during peer review. It also contains the code that produced them, the five figures as published, and the manuscript assembly script, which reads the tables and interpolates every reported number into the text. Running that script from this deposit reproduces the submitted manuscript byte-for-byte, and the figure script reproduces all five figures byte-for-byte; both were verified on 3 November 2026. A SHA-256 manifest covers every file, and TABLE_INDEX.md maps each table to the part of the article it supplies. The two primary data sources are third-party and are not redistributed: CAMS regional ensemble forecasts, raw and MOS-calibrated, distributed by the Copernicus Atmosphere Monitoring Service through the Atmosphere Data Store (https://doi.org/10.24381/2cce4067) under the Copernicus licence; and European Environment Agency air quality e-Reporting hourly surface observations, obtained from the EEA air quality download service under the EEA's own terms. Both are free to any user under those terms. The retrieval scripts and vintage identifiers needed to rebuild the matched panel are included. Code is licensed MIT; the derived tables and figures are licensed CC BY 4.0. Neither licence extends to the CAMS or EEA source data. Version 1.0.1 (4 November 2026) updates the manuscript assembly script and the reference table to match the submitted version of the article, following a final pre-submission review and a retitling. The frozen metrics tables and analysis outputs are unchanged from version 1.0.0.