Original Wordle dictionary: 2,315 answers and 12,972 accepted guesses. Answers are sampled uniformly. This is not the current NYT editorial pool.

Each decision is evaluated using only earlier clues. The objective is expected additional guesses to solve, including the move being evaluated—not information alone and not six-guess win probability.

With eight or fewer candidates, dynamic programming searches all legal guesses for the exact minimum. Larger positions screen the full dictionary, then evaluate eight finalists plus your move with a greedy continuation strategy and exact endgames. These are estimates, not proofs of global optimality.

Decision cost is the gap from the best evaluated move. Information is Shannon entropy. Luck compares your resulting candidate count with all possible feedback outcomes; ties use midranks. Analysis runs locally in a separate worker.

Ranked games keep the answer on the server and count losses or skips as seven guesses for leaderboard efficiency. Instant practice is private and does not affect the leaderboard. Signing in syncs your practice history to D1; only your GitHub handle and aggregate ranked stats are public. Server validation prevents direct score forgery, but cannot prevent someone using an external solver.