Lacunari

Use case · Scientific & medical literature

Which findings rest on a single paper?

In a literature corpus, the most important property of a claim is often how many independent sources support it — and that is precisely the property a relevance-ranked search will never show you. A single striking paper and a finding replicated eleven times look identical in a result list.

Corroboration as a structural property

Lacunari keys documents by content hash and counts independent contributions, so "how many separate sources does this rest on" is a property of the data rather than an interpretation. It reports claims supported by exactly one document as a distinct category of finding — not because single-source claims are wrong, but because knowing which ones they are is the first step of any serious review.

$ lac web gaps --kind uncorroborated ◻ 14 subjects appear in exactly one document

Coverage holes across a question

The same timeline logic that finds a missing year in an archive finds the period, population or dosage range nobody has published on. It reports the hole and stops there — it will not speculate about why the gap exists, and it will not tell you the absence of evidence means anything in particular.

Never read is a category

The most common finding in any real corpus is material you hold and have never extracted. Those documents are invisible to every question above, so Lacunari tracks held-but-unread explicitly rather than letting it hide inside a total count.

What it will not do

  • It does not assert conclusions. An edge means two entities were named in the same document and nothing more. No causal claim, no semantic inference.
  • It does not merge similar names silently. Candidates are proposed for a human to rule on.
  • It does not run a model. Every result above is set-based SQL over data you hold, which means it is reproducible and does not change because a vendor shipped a new checkpoint.

Reproducibility is the point. Results can be attested with a content hash and re-derived by a colleague on another machine — so "I ran the analysis and got this" becomes something a reviewer can check rather than something they have to take on trust.