Binghamton's entropy-ranked guesses solved 99% of simulated Wordle puzzles

Shannon entropy is a measure of uncertainty, used in information theory to study how efficiently information can be transmitted, stored or processed ^[13]. Applied to a guess, it scores the feedback that guess would produce. Binghamton's team computed it over the options still standing after earlier guesses had eliminated some, and preferred the word that would remove the greatest number of remaining possibilities ^[7][16]. Since the game colors each letter separately ^[6], a word built from letters that split the survivors into smaller groups can be worth more than a word that might actually be right ^[8].

"Let's say you're at a certain guess. The previous guesses will eliminate a whole bunch of options, and based on the remaining options, guessing some words will send you into a trajectory where information gain is speedier," said Congyu "Peter" Wu, the assistant professor who led the work at the School of Systems Science and Industrial Engineering ^[9][3].

Donald Stephens, a doctoral student on the team, described the swap in objective ^[4]. "By applying Shannon entropy, the objective shifts to maximizing the expected reduction in uncertainty rather than the probability of being right," he said ^[11]. Some of what the method recommends can look strange or even random to a human player, according to the university ^[17].

The 99 percent is a solve rate in simulation ^[1]. Binghamton did not report how many games were simulated, an average number of guesses, or a score for a likelihood-first strategy run on the same puzzles ^[15]. At 99 percent, roughly one puzzle in a hundred still ran past the limit of six guesses ^[14]. A solve rate tells you the six-guess budget held ^[5]; it leaves unmeasured the gain over a player who always guesses the likeliest word left, and Stephens's claim that the approach "can lead to solving the puzzle in fewer guesses" is the part a guess-count distribution would settle ^[11].

The entropy score requires an enumerated set of candidates to compute over ^[16]. Wordle supplies one, because the opening guess can be any valid word and the feedback rule is fixed ^[18][6]. Running the strategy during a real game means launching a separate script or program alongside the puzzle ^[12].