Not every source gets an equal say
Source-weighted fusion · reliability scoring
Every source carries a reliability score. A filing outweighs an anonymous tweet; conflicts of interest are discounted automatically.
Housecarl Arbiter · Confidence scoring for your AI
Decisions you can defend.
Three credible sources. Three different conclusions. Your AI picks one and sounds certain. Connect Arbiter and it weighs each source by how reliable and how recent it is, lays out the disagreements instead of smoothing them over, and hands back an answer that says how confident you should be — and why.
No gut calls. The reasoning is on the record.
If you'd lean on AI more if you could check its work, this is that check: Arbiter scores source reliability, detects contradictions between sources, and shows you where a confident-sounding answer has nothing behind it.
$0
To start — no card
2 min
To connect your AI
11
Source types, each weighted
What it does
Disagreements, laid out — not averaged away
Contradiction detection
See where your sources actually disagree — every reading that holds together, who backs it, and how strong it is.
Old news counts less
Temporal decay · recency weighting
Claims lose weight as they age. Stale rumors don't pollute a live thesis.
Every answer shows its work
Full audit trail
Every claim has a source, a score, and a timestamp. Retrace the reasoning after the decision.
Works inside Claude, ChatGPT, and Gemini — one connection, no code. Developers: REST API and MCP.
Sound familiar?
Your AI sounds sure — and it's wrong.
Fluent answers, invented citations, contradictions smoothed over. What to do when you can't trust the confidence →
Your sources don't agree.
Twelve tabs, three conclusions, one decision that won't wait. How to settle it with the evidence weighed →