IFSB-12 Question Answering Corpus: span verified question and answer pairs for Islamic finance liquidity regulation

This dataset contains 799 question and answer pairs derived from IFSB-12, Guiding Principles on Liquidity Risk Management for Institutions Offering Islamic Financial Services (Islamic Financial Services Board, March 2012). Every answer is a contiguous verbatim span of the standard, verified character for character against the source; 797 of 799 required no typographic normalisation. The pairs cover 291 of 299 eligible text units across four question types (Basic, Scenario, Comparative, Exception) and nine categories specific to the scope of IFSB-12. A further 395 declinations are recorded, each with an explicit reason: where a text unit does not support a question type, the pipeline records why rather than generating a question the text cannot answer. Each item carries its citable reference, page, section, the guiding principle it elaborates, the answer span offsets into the source unit, automatic warning flags, and the full generation configuration (model, prompt version, system prompt hash, reasoning effort, output mode), so that any pair can be traced both to its place in the standard and to the settings that produced it. Answers were produced by having a language model identify the answering passage, which the pipeline then copied from the source after verifying the match; the model did not write the answer text. Questions are model generated and have not yet been validated by domain experts. Warning flags are automatic detector output and have not been adjudicated. Users should treat both as provisional.