EXPERIMENT 1
A/A posterior sign probability
Does P(B > A) converge to zero when A and B are identical?
1/6
Ready to test the claim
Run the active simulation, or load the committed publication results.
REPRODUCIBLE STATISTICS
Do not trust the claim.
Run the simulation.
Six vendor-neutral experiments test what flat priors, repeated monitoring, expected-loss rules, and informative priors actually do. Every run happens locally in your browser.
Same seeded algorithms in Python and JavaScript. Open source. No data leaves your device.
WHAT IS BEING TESTED
Inference, monitoring, and decisions are separate choices.
The prior
A flat prior does not encode your historical effect distribution. An informative one can, and its misspecification is measurable.
The stopping rule
A posterior output does not automatically make repeated threshold crossing safe. The stopping rule is part of the procedure.
The loss function
Expected loss is a business rule, not an error guarantee. Its threshold determines how often noise becomes a decision.