Before Meridian's data-preparation team declares the charter-desk corpus 'ready' for indexing, what should they verify about document-type coverage, beyond simply confirming that files were successfully extracted and chunked?

Verifying that the corpus's coverage matches the actual scope of expected questions is correct because assessing and improving source-data coverage for a retrieval corpus is an explicit data-preparation responsibility distinct from mechanical success metrics like extraction or chunk counts - a corpus can have every file successfully processed and still be missing entire contract types, clause categories, or scenario classes the assistant will actually be asked about, exactly as the HS-code jurisdiction gap illustrates elsewhere in this domain. Confirming only that every extracted file produced at least one chunk is insufficient because it verifies the mechanics of the pipeline worked, not that the pipeline's input was complete in the first place - a perfectly executed pipeline over an incomplete input set still yields an incomplete, under-covered index. Confirming that chunk count exceeds an arbitrary round number like ten thousand is a meaningless proxy for coverage - a corpus could easily clear that number while still being missing an entire important document type, or could legitimately need far fewer chunks and still be fully adequate; raw volume does not measure whether the right content is present. Confirming that every chunk is exactly the same size is not a coverage check at all, and it is arguably the wrong goal - as this domain's other items establish, different source types legitimately need different chunk sizes, so uniform chunk size is neither achievable nor desirable, let alone evidence of adequate coverage. The exam-relevant skill is distinguishing pipeline-mechanics success from actual data-quality and coverage sufficiency.