For researchers

Compare conflicting sources and see what they support

You have read five articles and they say five different things. You asked an AI and got a confident answer with no sources. You do not need another opinion. You need to see what your sources support and what they do not.

What Arbiter does with your sources

Each source becomes a named entry with a type, such as researcher, journalist or regulator, and a reliability that follows from the type. Each claim is tied to the source that made it. Arbiter then weighs the claims against each other and returns a verdict with its reasoning.

Every weight is visible. The starting reliability for each source type, and the reason for it, is published in the source-reliability table. No weight is hidden.
Agreement, disagreement and silence are separate. Where credible sources conflict, the verdict says so. Where no source has evidence, the verdict says that too.
Every conclusion traces back. The audit trail lists each claim with who made it, how strongly the evidence supports and opposes it, and how uncertain it remains. It is readable as JSON through the REST API on every plan.

A worked case: a study, the university press release about it, and a news story that quotes the press release. Enter all three, mark the press release and the news story as depending on the study, and their agreement counts once, not three times.

What it will not do

Arbiter does not decide what is true in the world. We measure the sources you give us against what they say. If the literature is wrong, Arbiter is wrong with it. What it gives you is a clear account of what your sources, taken together, support.

Start

Connect Arbiter to Claude in five minutes, with no code. Or open the app — free, no card required. Current plans are on the pricing page.