Heads Needed to Suspect a Trick Coin

This question is about deciding whether an apparently lucky run should make you suspect a biased source rather than normal randomness. It sits at the intersection of conditional probability and Bayesian reasoning, a core theme in quant prep and technical interviews. You interpret observed outcomes to reassess which hidden mechanism is most plausible.

It trains your ability to translate an informal story about uncertainty into precise probabilistic statements, especially posteriors over hidden states. You must be comfortable with priors, likelihoods, and how evidence accumulates, as well as doing clean, error-free probability algebra under time pressure.

This matters for quant interviews because model selection, parameter inference, and signal detection all rely on this logic. Successful quants constantly re-estimate beliefs as new market data arrives, exactly as in this problem.

What it tests

This problem class is governed by Bayesian inference, where we update our beliefs about which of several possible sources generated observed data, based on how likely each source is to produce that data. The core structure is: there are multiple types of objects (here, coins with different biases), and you observe a sequence of outcomes (here, heads). The probability you want is the posterior probability: given the observed data, what is the chance you had a particular type of object? This is computed by weighing the likelihood of the data under each hypothesis by the prior probability of each hypothesis, then normalizing. The pattern holds because the observed data is much more likely under some hypotheses than others, so enough evidence can overwhelm even a small prior probability.