10% Chance to Detect Poison in N Kegs

This brain teaser is about extracting as much information as possible from a very small number of binary outcomes. It belongs to the family of combinatorial identification puzzles where each observed pattern must be interpreted as a code that points back to the hidden choice. For quant prep, it's a classic example of turning a story problem into an abstract information problem.

It trains your ability to reason about information content, discrete probability, and how many states you can distinguish with limited tests. You practice encoding and decoding possibilities, thinking in terms of outcome patterns, and trading off certainty versus success probability. It forces you to quantify how much "signal" each extra test provides.

This matters in quant interviews because trading and risk problems often boil down to optimal information extraction under constraints. Strong candidates show they can structure messy narratives into clean probabilistic frameworks, reason about efficiency, and justify why a particular design is close to optimal. These are exactly the skills top firms look for in quantitative interviews on platforms like MyQuantPartner.

What it tests

The core structure of this problem class is combinatorial encoding using binary representations to maximize information extraction from limited tests. When each tester's outcome (alive or dead) is observed, their collective results can be interpreted as a binary string, allowing you to uniquely identify up to $2^k$ possibilities with $k$ testers. The principle generalizes to any scenario where a set of binary outcomes can be mapped to a set of possibilities, and the challenge is to maximize the number of distinguishable cases given the constraints. The reason this works is that each tester's participation doubles the number of distinguishable outcomes, as each can be in one of two states, and all combinations are possible. When the identification does not need to be certain (i.e., probability less than 1), you can further increase the number of cases by allowing ambiguity, partitioning the space so that each outcome corresponds to a group rather than a single item, and accepting a lower probability of correct identification.