Research
What if your theory had to survive AIs that can't mark their own work?
Bring the theory. Independent AIs design the test, run it and grade it, then hand you the evidence so you can decide what comes next.
Independence
No AI grades its own homework.
One designs the experiment without seeing data. Another runs it without seeing your paper. A third grades the results, and plain code with no AI assembles what you get back.
Discipline
The answer can't move after the fact.
Every design has to state its predictions in numbers, and the result that would prove it wrong. If a design is edited after it locks, the run flags it. Each prediction ends confirmed, refuted or not judgeable.
And everything around it.
Works from your AI
Start a round from your AI through MCP, using a paper and spec kept in Paddock.
Knows what needs you
Broken locks, stalled work and missing inputs come first, worst at the top.
Honest numbers
Costs include failed attempts, estimates are marked, and a missing cost never shows as zero.
Recovers on its own
Silent or failed runs are retried, then marked stalled if they stay stuck.
Stop anything
Stop one experiment or a whole request at any point.
Keeps private text private
Reports likely to quote or restate your private notes are held back.
Straight answers.
Does it come up with ideas?
No. You still form the theory and decide on the next round. Research tests ideas.
What do the checks actually judge?
They check that a design can be tested and that private text stays private. They never judge the science.
Are the privacy checks a guarantee?
No. They catch likely restatements, but a reworded note can get past them.
What does a run cost?
It depends on the experiment. Every run shows its cost as it goes. There are no spending limits yet.
Can my team use it?
Not yet. There are no per-person workspaces. Research also needs Paddock and a Chaio account.
What if your next theory had to prove it?
Research is invite only for now. One Chaio account signs you in.