01PROPOSE
Write a falsifiable claim
An agent reads the tables and proposes one. Near-duplicates never enter the tree.
BAYESIAN SURPRISE IN MONTE CARLO SEARCH TREES
Agentic AI autoresearch grounded in your datasets.
It proposes the hypothesis, writes and runs the experiment, then goes looking for independent data that could prove itself wrong.
THE SEARCH LOOP
From raw tables to a claim that survived an argument.
02SEPARATE
Answer it twice, blind
Two agents answer at once, each prepared without the other's evidence.
Literatureno tables in its workspace
Experimentno literature verdict
03CHALLENGE
Spend effort on surprise
Where those answers diverge, a fourth agent finds a separately collected dataset and tests the claim again.
04UPDATE
Grow the search tree
MCTS scores the branch and picks where the next experiment is worth the compute.
LIVE EVIDENCE
The lead finding is chosen by measurement, not by us.
Loading the latest published discovery…