An identifier is a precise request
When someone searches for invoice INV-7048, they usually need one particular record. A document about invoicing in general may be topically relevant and still be useless. The research on BM25 explains how lexical matching uses observed terms and their distribution to rank documents. That makes it a valuable starting point for identifiers, although an ordinary tokenizer may split punctuation or discard characters that matter to the business meaning of the reference.
Preserve the complete reference
A proposed ABOLTUS search design should retain both the original identifier and a normalized searchable representation. For example, INV-7048 and INV 7048 could be considered related forms while INV-7049 remains a different candidate. Search results should show the actual matched value, document version and surrounding label. A dedicated identifier field would help distinguish an invoice reference from an identical number inside a telephone field, without silently changing the source document or inventing a relationship between records.
Test the difficult neighbors
The BEIR research demonstrates why retrieval needs evaluation across different tasks and domains. For identifier search, a useful local test set would include near-identical numbers, leading zeroes, mixed letters and digits, and references embedded in scanned pages. Reviewers should check whether the correct record appears first and whether wrong neighbors are clearly distinguishable. This is an intended design direction, not a claim that the current ABOLTUS website provides invoice search. Semantic retrieval can assist discovery, but the exact-reference task deserves its own measurable success criterion.