RFQ-30: A Synthetic Test Suite for Missing and Conflicting Procurement Inputs

Draft technical note and reproducibility package prepared for Arcforth. RFQ-30 contains 30 AI-assisted synthetic procurement intake cases: ten complete, ten incomplete and ten conflicting. Two deterministic extraction baselines were executed. The first-value baseline matches 230/240 annotated fields but releases ten conflicting requests; the duplicate-aware baseline matches 236/240 fields and still releases four prose-conflict cases. These are descriptive results on a constructed fixture, not operational error estimates. The package includes data, reference labels, code, actual per-case outputs, a frozen local protocol, a data card and limitations. No LLM model was evaluated, no human review time was measured, and no client data or outcomes are included. AI assistance was used in authoring and analysis. Independent practitioner review has not occurred. Corporate authorship and the proposed CC BY 4.0 license require owner confirmation before publication. This draft is not peer reviewed.