Finding the Odd Ball Without Knowing If It Is Heavy or Light

You have 12 identical-looking balls; one is either slightly heavier or slightly lighter than the rest (you don't know which, and don't know which ball). You have a balance scale and exactly 3 weighings. Find the odd ball and say whether it's heavy or light. This is strictly harder than the version where you're told the direction in advance, because every weighing outcome now has to be interpreted without knowing which way the "wrong" answer would tip.

Why not knowing the direction makes it harder

If you already know the odd ball is heavier, then a weighing that tips left tells you the culprit is on the left pan — done, no ambiguity. Without knowing the direction, a tip to the left means "either a heavier ball is on the left, or a lighter ball is on the right" — two very different conclusions bundled into one observable outcome. You have to design weighings so that this ambiguity gets resolved by the pattern across all three weighings, not by any single one.

Counting the possibilities first

There are 12 balls, each of which could be the odd one, and each could be heavy or light, giving $12 \times 2 = 24$ possibilities to distinguish (plus, in some versions, "all balls equal," but with exactly one odd ball guaranteed, 24 is the relevant count here). Three weighings, each with 3 outcomes, give $3^3 = 27$ distinguishable outcome sequences — just barely enough room for 24 possibilities, with 3 outcome sequences to spare. This tightness is why the puzzle needs a genuinely clever weighing plan rather than a naive halving strategy; a scheme that wastes even one weighing's worth of information will fail.

Working the first weighing and the branching logic

Weighing 1: put 4 balls on each pan, 4 aside. Two outcomes to handle:

Case A — balances. The odd ball is among the 4 set aside, and all 8 weighed balls are confirmed normal. This sub-case is now the known-direction problem in disguise: with 2 weighings left and 8 known-good balls available as a reference, weigh 3 of the suspect 4 against 3 known-good balls. If it balances, the 4th suspect is odd — weigh it against a known-good ball in the final weighing to learn heavy or light. If it tips, the odd ball is among those 3, and its direction is now known from which way the pan tipped, reducing to a simple 1-of-3 known-direction search in the last weighing.

Case B — tips, say left-heavy. The odd ball is among these 8, and you know a tentative direction fact for each: any of the 4 on the left is potentially heavy, any of the 4 on the right is potentially light (the 4 set aside are confirmed normal). Now weighing 2 is designed to split this specific set of 8 possibilities — for example, take 3 from the "potentially heavy" left group plus 1 from the "potentially light" right group and weigh them against 3 known-good balls plus 1 different "potentially light" ball, arranged so that each of the three weighing-2 outcomes (left, right, balance) narrows the 8 candidates down to a small enough set that weighing 3 can finish the job with certainty.

The exact ball assignments differ by source, but the shape is always the same: weighing 1 splits 24 possibilities roughly into a 16-possibility branch and an 8-possibility branch (via balance vs. tip), and each subsequent weighing is chosen to keep splitting whatever branch you're in as evenly as the $3^k$ budget allows.

What this means in practice

This puzzle is a favourite because it forces a candidate to actually count the possibility space ($12 \times 2 = 24$) and check it against the information budget ($3^3 = 27$) before touching a strategy — the same discipline used to sanity-check whether a proposed test, signal, or audit procedure has enough resolving power for what it's being asked to distinguish, before assuming a "reasonable-sounding" plan will work.