In Peter Wason's 1960 experiment, participants were given the sequence 2-4-6 and asked to discover the hidden rule by proposing their own triples. Most guessed "increasing by 2" and only proposed confirming sequences — they reached 90%+ confidence while remaining completely wrong. The experiment exposed a near-universal cognitive flaw that existing research could describe but not make visceral.
Built a multi-agent Monte Carlo simulation where "confirmer" and "falsifier" agents reason about the same hidden rule using opposite strategies. Confirmers propose triples that confirm their current hypothesis; falsifiers deliberately probe what their hypothesis predicts is impossible. A shared hypothesis library of ~20 rules (from narrow to broad) provides the revision mechanism. Three live charts and a per-agent feed make the divergence visible in real-time across 30+ rounds.
Confirmers routinely reach 85–95% confidence while fewer than 10% hold the correct hypothesis. Falsifiers end at lower confidence but 75–85% accuracy. The scatter plot is the most damning chart: high-confidence wrong answers cluster perfectly by strategy. This isn't about intelligence — confirmers follow a locally rational procedure. The flaw is that a confirming test can never tell you your hypothesis is too narrow.