German tax-law corpus with historical law versions

Open German tax-law corpus with historical law versions: each norm is stored with its validity period. Dataset on Hugging Face (https://huggingface.co/datasets/sevriukov/german-tax-law-corpus), version 1.1.1 (13.11.2026): 187,025 documents from 17 sources, 3,077,123 fragments split along each document's own structure (paragraph, subsection, sentence; margin numbers). The dataset has been downloaded 286 times (221 in the last month) without promotion. Scripts to collect, split and verify the corpus against the source text are in this repository (v0.4.1, Apache-2.0). The open benchmark KF2 (questions, key, scoring; CC BY 4.0) measures whether retrieval finds the norm version valid on the question date: top-1 80.4 %, top-10 96.8 % (open key v2.1.1, 281 scored questions).