Data and code for "Ranking is not allocating: a two-level validation of a public-sector prioritisation index for 181,065 Brazilian schools"

School-level snapshot (19 October 2026) of a composite water-vulnerability index for Brazilian public schools: 181,065 schools in the 2024 School Census and 176,805 in 2025 (configuration identifier 7067f6005e5c, engine version 2), with the seven component variables, the two axes, the published first-principal-component composite and provenance fields. Also included: the three municipal outcomes used for external validation (per capita water consumption, water and sewerage service coverage, from SNIS), municipal income per head (IBGE), the PCA loadings and the national band cut-points published by the system, the derived municipal tables, results.json with every quantity reported in the manuscript, and the Python code (recompute.py, figs_data.py, figs.py, figs2.py) that reproduces all tables and figures. Primary sources: INEP (School Census), INMET, MapBiomas, IBGE, SNIS. Data under CC BY 4.0; code under the MIT licence (see README.md). Version 1.2.0 accompanies the revised manuscript "Ranking is not allocating: a two-level validation of a public-sector prioritisation index for 181,065 Brazilian schools". It adds the allocation-level routines (relevance.py, relevance2.py, relevance3.py, p70.py, inside_outside.py, fig_weight.py): capture curves under two cost assumptions, paired bootstrap of capture differences, capture across the structural weight, the production scheme, discordant sets and the value of partial outcome information; recompute.py now reads the deposited school files directly. All files are in seps_data_code_v1.2.0.zip.