Signal & Noise

Two hypotheses, one coin. Watch belief converge as evidence accumulates, flip by flip.

flips observed

The coin in play

Speed

Auto-run rate — 5 / sec

H₁ — fairmodels P(Heads) = 0.5

0.500

H₂ — biasedmodels P(Heads) = 0.7

0.500

Posterior probability over time

P(H₁ | data) P(H₂ | data)

Flip tape

No flips yet — start the coin to begin recording.

Each flip is drawn from the true coin above (independent of what the model believes). After every flip, both hypotheses' posteriors are recomputed with Bayes' theorem: P(H|D) = P(D|H)·P(H) / P(D), using the previous posterior as the new prior. The model always assumes P(Heads) is either exactly 0.5 or exactly 0.7 — it never gets to see which coin is actually being flipped.