Gap finding

Find the gap before your committee does

A gap is a claim about absence, and absence is the hardest thing to evidence. Agent Bayes maps what your corpus actually covers, reports where the evidence runs thin instead of writing over it, and cites everything, so the gap you name is one you can defend.

Create your account

Free to start, no card. 200 credits when you verify your email.

The reason you are here

You think there is a gap. You cannot prove it yet

The feeling arrives early: nobody seems to have looked at this combination. The evidence for it arrives much later, if at all, because proving an absence means showing you looked properly and found nothing.

Meanwhile the tools work against you. Ask a model what is missing and it will generate something plausible, which is the one answer you cannot take into a proposal.

What you need is a system that says "your sources do not address this" and means it, because it only ever read your sources.

How to evidence an absence

Map the present, then interrogate the edges

The gap argument is built from what the corpus does contain, which is why the first step is not optional.

Map what your corpus does cover
Run a review pass so the map holds the established claims with their citations. You cannot argue something is missing until the present is on the page.
Ask what is not there
Point the agent at the boundaries: which populations, methods, periods or conditions the literature has left alone. Where the corpus is silent, it reports the silence.
Turn a hunch into a defensible claim
A gap you can cite is a gap you can defend. Each supporting claim links to the passage behind it, so the case for the gap is traceable in a viva or a proposal.

In the product

Slice the corpus and compare coverage

Tags scope both your own searches and the agent's retrieval, so coverage questions get tested rather than guessed.

Labelled collections of passages grouped across documents in a project — Curate collections of evidence across documents, then intersect them.

Semantic search over a corpus with filters for tags, author and year — Scope a search by tag, author or year to test where coverage stops.

What you can rely on

What makes an absence checkable

Thin evidence is reported, not padded

When your corpus does not support an answer, the system says so as a knowledge-base limitation. That report is the raw material of a gap claim, and most tools suppress it.

Conflicts stay visible

Unresolved disagreement is a different kind of gap from silence. Opposing findings stay as separate nodes, so nobody-has-looked reads differently from nobody-agrees.

Slice the corpus to test coverage

Tag by method, population or period, then scope a run to one slice. Coverage questions get answered by comparing slices rather than by intuition.

Confidence scores show the soft spots

Each claim is scored 0 to 100 on source strength and agreement. A cluster of low scores is a map of where your field is least sure of itself.

What Agent Bayes does not do

It cannot see beyond your library. Every gap it reports is a gap in the corpus you indexed, not a proven gap in the field, and the distance between those two is your responsibility.
It does not search the open literature, so it cannot rule out a paper you have never downloaded.
It does not decide whether a gap is worth filling. Plenty of gaps exist because the question was not interesting.

The first limit is the one to say out loud in a proposal. Naming the boundary of your search is what makes the rest of the argument credible.

Before you ask

Questions researchers ask first

Can it tell me a gap exists in the whole field?

No, and no honest tool can. It reads the library you gave it, so what it reports is a gap in your corpus. Narrowing that distance is the point of curating a thorough library, and the page says so rather than pretending otherwise.

How is this different from asking for a summary?

A summary rewards coverage and smooth prose, so absences disappear. Gap finding needs the opposite. The system has to keep reporting that your sources do not address something, which is a first-class output here rather than a failure.

Does it distinguish silence from disagreement?

Yes, and the difference matters for a proposal. Silence shows up as a reported knowledge-base limitation. Disagreement shows up as sibling nodes holding opposing findings, each with its own citations.

Will my supervisor accept this as evidence?

They will accept a cited argument. Every claim behind your gap traces to an author, year, printed page range and passage, and you can open the source at that page in one click.

What does it cost to try?

The free plan is enough to index about five papers and run a few queries, with 200 one-time credits once you verify your email. A typical research query costs 10 to 20 credits.