Disagreement is data

When your sources disagree, you should see both

Most tools resolve conflicting sources into one confident sentence, and the conflict was the interesting part. Agent Bayes gives each position its own node and its own citation chain, so a live debate in your field still looks like one on the page.

Create your account

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

The reason you are here

You asked about a contested question and got a shrug

The answer came back balanced and useless. "Findings are mixed." "Results vary by context." "More research is needed." Every sentence technically true, none of it telling you who found what, under which conditions, or which camp your own data would land in.

That flattening is not a wording problem. A model asked to summarise will reach for the average, and the average of two incompatible findings is a sentence no researcher in your field would sign.

The structure of a contested literature is the finding. A tool that collapses it has removed the thing you were looking for.

How it holds the line

Two positions, two nodes, two citation chains

The split survives the write step, which is where other tools lose it.

Ask about something contested
Point the agent at a question your field has not settled. Retrieval runs around ten queries in parallel across your corpus and re-ranks the results together.
Opposing evidence stays opposed
Where sources conflict, the map gains one node per position rather than one node per topic. Each carries the citations that support it and nothing else.
You decide what it means
Edit either node, ask for a prose discussion across both, or expand one branch further. The interpretation is the part of the work that stays yours.

In the product

The map shows the split at a glance

Branches sit side by side on the canvas. Nothing is hidden behind a summary you have to unpack.

The Agent Bayes mindmap canvas with branching citation-backed claims — Claims branch on the canvas, so competing positions stay visible next to each other.

What you can rely on

Why the split survives

Sibling nodes, not a blended sentence

Two findings that contradict each other become two nodes positioned alongside one another. Neither one is softened to make room for the other.

The rule is architectural

The stage that writes into your mindmap is explicitly forbidden from averaging opposing views. It is not a prompt preference that drifts between runs.

A cited prose discussion on request

Ask for it and the agent writes a structured passage that moves through each school of thought and foregrounds where the evidence genuinely diverges.

Confidence reflects the split

Each claim is scored 0 to 100 on source strength and agreement. A contested claim shows a lower score, so the disagreement is visible in the number too.

What Agent Bayes does not do

It does not adjudicate. It will not tell you which side of a debate is correct, because that judgement is the contribution you are being paid to make.
It does not invent controversy. If your corpus agrees, the map agrees, and the system reports thin evidence as a knowledge-base limitation.
It cannot see papers you did not add. A debate your library only covers from one side will look one-sided.

The third one is worth sitting with. This tool is only as balanced as the library you point it at, which is why curation stays your job.

Before you ask

Questions researchers ask first

What if the sources do not actually disagree?

Then you get one node. Sibling nodes are how a real conflict is represented, not a shape the agent imposes on agreement.

Can I see the reasoning behind each side?

Every node keeps its citations with author, year, printed page range, and the passage the claim was drawn from. Click a citation and the reader opens at that page.

How is this different from asking a chatbot for both sides?

A chatbot produces both sides from whatever it absorbed in training. Agent Bayes retrieves from the papers you indexed, and each position carries the specific passages behind it, so you can check whether the split is real in your own literature.

Does it work on a literature that is mostly settled?

Yes, and it will tell you so. When your corpus does not support a distinction, the system reports that as a limitation of the knowledge base instead of manufacturing a controversy.

What does it cost to try?

The free plan costs nothing and gives you 200 one-time credits once you verify your email, enough for roughly five indexed papers. A typical research query costs 10 to 20 credits.