Evidence Weighing Trainer — Bayesian Critical Thinking

Claim Under Evaluation

Loading scenario…

Evidence

—

Supports —

Source reliability—

Your belief in the claim 50%

Show Bayes-optimal target (ghost needle)

Needle response speed 1.0×

Live Readouts

Your belief

50%

Critical Thinking Score

—

Confirmation bias index

—

Round

1 / 8

How it works

Each round presents one piece of evidence with a direction (supports / contradicts the claim) and a reliability r ∈ [0.08, 0.95] representing source quality — a peer-reviewed RCT is reliable, an anonymous anecdote is not. The evidence contributes a log-odds shift:

ΔL = direction × reliability × k     (k = 2.2)
logit(p) = ln( p / (1 − p) )
p*_optimal = sigmoid( logit(p_prior) + ΔL )

This is the standard Bayesian log-odds update rule, with reliability standing in for the evidence's diagnosticity (its log-likelihood ratio). After you move the slider and submit, the sim compares your chosen belief to p*_optimal computed from the same starting point:

Critical Thinking Score — 100 minus your deviation from the Bayes-optimal update, averaged across rounds. Reacting exactly proportionally to reliability scores near 100.
Confirmation bias index — positive when you shift further than justified on evidence that already agrees with your current lean, or shift less than justified on evidence that contradicts it (classic motivated reasoning). Near zero means you weigh evidence by its quality, not by whether you already believed it.
A separate "objective" trajectory updates purely by the formula above, ignoring your choices — the gap between your final belief and that trajectory is your overall calibration error for the scenario.

This is the same normative model used in medical diagnosis, forecasting and intelligence analysis: strong, independent, low-conflict-of-interest evidence should move belief a lot; weak or biased evidence should barely move it at all, regardless of which direction you'd like it to point.

Your belief needle Bayes-optimal target

Scenario complete

Final belief: —

Bayes-ideal belief: —

Average Critical Thinking Score: —

Click "New Scenario" to try another claim.

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.