A cut can change the meaning
Splitting every document into equal-sized blocks is convenient for an index, but it can separate a rule from its exception or a question from its answer. TextTiling and related text segmentation research study boundaries between topics rather than assuming that every fixed number of characters forms a coherent passage. This provides a useful starting point for organizational documents, where headings, lists and local context often carry meaning that a flat text stream obscures.
Preserve the document hierarchy
For a proposed ABOLTUS ingestion pipeline, a passage about access approval should retain its parent heading and the relevant neighboring condition. If the section spans several pages, page boundaries should not automatically become semantic boundaries. Conversely, two unrelated sections on one page should not be fused merely because they fit inside a convenient block size. A result can then identify the precise passage while showing its place within the larger policy, helping users recognize whether they are reading a general rule or a special case.
Compare segmentation against real questions
A useful experiment would retrieve evidence for questions whose answers cross paragraphs, tables or section breaks. Reviewers could compare fixed-size splitting with structure-aware alternatives using the same source snapshot. More elaborate segmentation is not automatically better: noisy headings and extraction errors can create false boundaries. The intended ABOLTUS direction is to preserve structure where it is reliable and measure whether it improves evidence recovery. The cited research supports testing passage design, not claiming that one segmentation algorithm understands every document correctly.