Expected Flips Until Consecutive Heads

This problem is about expected value in a stochastic process where some outcomes accumulate progress, some wipe it out, and one special outcome ends the game. It forces you to recognize that the raw description hides a clean probabilistic structure, typical of quant interviews, and to translate a narrative random process into a compact mathematical model.

It trains your ability to formalize Markov-style dynamics, identify first-hitting events, and work with conditional expectations. You practice grouping outcomes by their functional effect on the state, recognizing effective success probabilities, and using memoryless structure to simplify the process into something tractable for quant prep, like a geometric-type model.

This matters for quant interviews because pricing, risk, and algorithmic trading often involve path-dependent events, regime switches, and resets. Interviewers want to see you abstract messy rules into a state process, compute expectations efficiently, and explain your reasoning clearly under time pressure.

What it tests

Whenever a process involves repeatedly sampling from a set with certain outcomes that reset progress, and others that advance or terminate the process, the key is to model the system as a Markov process and focus on the order in which the critical outcomes occur. The probability that a particular outcome occurs before any of several others can often be found by symmetry or by solving a system of equations for first-hitting probabilities. The expected value of a quantity conditioned on a particular order of events can often be reduced to a simpler process by grouping outcomes that have the same effect on the system's state. This reduction frequently turns the problem into a geometric distribution, where the expected number of steps until a 'success' is simply $1/p$, with $p$ the effective probability of the terminating event. The principle holds because the memoryless property ensures that the process 'restarts' after each non-terminal, non-resetting event, so only the relative likelihoods and effects of the key outcomes matter.