From the single right provision to a solution space: multi-norm retrieval, clarifying questions and point-in-time law for German tax questions

Retrieval systems for legal questions are usually scored by one number: how often the "right" provision appears among the top results. I argue that this number describes the wrong object for German tax law. In a blind assessment (40 questions × the first 10 retrieved places, two model reviewers, agreement 94 %), a single life situation was governed by two to four provisions at once: a rule, an exception, a procedure and a sanction. The answer is therefore a space bounded by several provisions, not one provision. I describe a working database that searches all 104,207 provisions in force without pre-filtering. On 225 control questions the target provision ranks first in 83.6 %, within the first five in 96.9 % and among the candidates in 100 % (fusion of five signals by reciprocal rank fusion, 26–27 October 2026). I show with § 14c UStG how one clarifying question ("who issued the invoice?") separates Abs. 1 from Abs. 2. I add the counter-party to the map (§ 320 and § 273 BGB) and propose ordering questions by information gain; this is a principle and a manual prototype, not a function. A second axis is time, which has been studied for German statutes (Prior, Schultz & Grabmair 2026) and French tax law (Cymbler et al. 2026); here it is applied to German tax law: 104,910 provisions are grouped into families of 150,886 versions so that the version in force on the date of the case is used. A ranking of provisions by legal force, derived from the text of the provisions themselves, gave no measurable gain; I report this negative result. The machine collects facts and shows all variants; a human interprets and decides. Part of a series; overview: https://doi.org/10.5281/zenodo.23949487. Files: English version (authoritative) and German version (Deutsche Fassung). Not legal or tax advice.