One: the source is attached to the sentence, not the page
A page-level bibliography cannot be checked. When five sources sit under twenty conclusions, no reader can tell which supports which — and in practice, some support none of them. The unit of provenance must be the claim.
Two: the record is kept as it stood
Public records change, and pages disappear. A citation to a URL that no longer says what it said is worse than no citation, because it looks like diligence. The record must be retained, dated and fingerprinted at the moment it was read, so a reader six months later sees what the analyst saw.
Three: inference is labelled as inference
Most useful conclusions in commercial real estate are inferred from incomplete record. That is legitimate — but only when it is declared. Mixing an established fact and a reasonable inference in one confident sentence is the single most common failure of automated research.
Four: the gaps are stated as plainly as the findings
What the record does not reach is part of the answer. A paper that lists only what it found invites the reader to assume nothing else exists. Naming what was sought, from whom, and why it could not be established is not weakness; it is the part a professional reader checks first.
The test
- Can you click a single conclusion and see the records beneath it?
- Was each record dated and kept as it stood?
- Does the paper distinguish what is established from what is inferred?
- Does it say where it could not reach, and what that limits?
If a conclusion cannot be opened, it is an opinion with a footnote.