GridFlex 1.0.0-rc2 -- the state that produced the deposited dataset The ELMAS correction (#25). The non-residential curves served upstream were the 2018 data labelled 2020, which put the non-residential weekend on the wrong days; building_eload 0.18.0 and buildingdata 0.11.2 read the restaged blob (generation 1793698860990011, config/sources.yaml). The 150 feeders were rerun from the dynamic stage (Slurm 72182) after the smoke run (72155) and the determinism audit (72181) passed on that stack; docs/migration_0.2_to_current.md section 1.8 has the evidence. Zenodo version DOIs, reserved 2026-11-05 and published once every file is verified after transfer: data 10.5281/zenodo.23978159 (concept 10.5281/zenodo.23294795) code 10.5281/zenodo.23978189 (concept 10.5281/zenodo.23334988) 1.0.0-rc1 keeps its own DOIs (data 10.5281/zenodo.23294796, code 10.5281/zenodo.23334989). 43 files, 20 184 651 889 bytes, built 2026-11-04T18:17:21Z (Slurm 73705) by `uv run snakemake all validation --configfile config/config.yaml`. The record's MANIFEST.json names b2af660 as its builder, with a clean tree and uv.lock sha256 1f0134fdad653957d3f3a177df909b8b31dae3967459cee86b80bc9a6eeb30fe. Pinned stack: Python 3.12, gridconnect 0.2.2, buildingmodel 1.13.1, building_eload 0.18.0, buildingcalibration 0.6.3, buildingdata 0.11.2, pandapower 3.2.1, roseau-load-flow 0.15.0, snakemake 9.26.1. 1.0.0 stays reserved for the state that produces both the dataset and the manuscript (plan T3.2/T5.1).