Purpose-Built Embedding Models for Rule-Bearing Text: Why General-Purpose Embeddings Score F1 = 0 on Regulatory Conflict and What a Compliance-Grade Model Must Preserve

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Typeset edition of Sam Sammane’s previously unpublished manuscript dated 21 April 2026. The original substantive text, numerical claims and references are retained. This paper presents six requirements for embeddings of rule-bearing text and the original Q-Prime position. A separate edition note clarifies publication status and points to two companion manuscripts: Embeddings, Geometry and Entanglement; and Q-Prime: Structured Embeddings for Rule-Bearing Knowledge. Preliminary preprint; not peer reviewed. Historical performance claims are reproduced as reported in the original manuscript, not as independently replicated results. CC BY-NC 4.0 applies to the paper only. QGI website: https://www.qgi.dev