Preparing your practice rounds…

Why splitting the pack works

Each case file shows every word in the site's working answer pool that matches a real opening line. Those are your suspects. Your job is to choose one accepted guess that separates them into the smallest feedback groups. A one-word group identifies its answer immediately; a large danger group leaves several suspects looking identical. Casual mode supplies four measured choices, while Expert mode keeps the original free-entry challenge. Underneath the cards, the arithmetic is the same partition math as the site's solver, and the sibling mode Beat the Solver grades the move you would play. Missing Word covers endgame deduction. Outside word games, How Sure? scores how well you know what you know. Each practice set has five rounds, ordered from easiest to hardest.

How scoring works

Every probe gets a value from one measure, expected elimination: group the candidates by the feedback pattern your probe would receive, sum (group size)² divided by the candidate count to get the expected remainder, and subtract that from the count. It is the average number of candidates your probe rules out. Your round score is that elimination as a percentage of the best available. "best" meaning the lowest expected remainder among all 14,855 accepted guesses for that candidate list, with ties broken by the smallest worst-case group. When several probes tie on value, the reveal says so.

Daily total bands with interpretations
Daily totalBandWhat it means
95–100% Optimal You matched the best available split in nearly every round.
80–94% Strong Most of your probes were close to the best available.
60–79% Mixed Most probes split the candidates well, but one round left a large group.
40–59% Weak Several probes left large groups. Check which letters they tested.
Below 40% Very weak Your probes ruled out few candidates. Check for letters you have not tested.

The practice total is the plain average of the five round scores, so 100% means five best-available probes. A share bar fills at 80%, the level where a probe stops costing you turns against the best line.

Rounds are built offline from the site's 2,374-word working answer pool and its reconstructed accepted-guess list; candidate counts describe this site's pool, not what the NYT will choose. The objective is always explicit: minimize the average number of possible answers after one guess. A possible answer may be best; sometimes a different word separates the remaining possibilities more effectively.