wordle-ai-js

A JavaScript implementation of my strategy for finding the best first word to guess in Wordle. This is an O(n²) solution. It is intended to be compatible with "hard mode" gameplay where clues must be preserved from guess-to-guess.

There are two methods you can use:

• findWord will get you the best word as quickly as possible, taking about 8.5 seconds on my laptop with an AMD Ryzen 9 5900HX. Your mileage may vary in terms of speed.
• findWordSlowly takes longer but produces a sorted list of the dictionary words and their score. You can use that list to check best and worst performing words. It takes around 30 seconds to run for me.

Overview of Strategy

This strategy works by scoring each word and trying to optimize the choice of word based on that score.

To score a word, we pretend it is our guess and calculate the number of times each type of clue would occur for all possible secret answers. We are interested in worst-case performance, so we score the word based on its worst performing clue. Thus, the score of a word is equal to the number of words for the clue with the highest count.

To choose a word, we pick the word whose score is the lowest. Said another way, we are looking for the word whose worst performing clue is the best among all possible words.

Result

I pulled the word list from the wordle source code. Wordle keeps a list of all possible words, but answers are pulled from a smaller list, presumably curated so that answers are more guessable. For the purposes of this exercise, I combined both of those lists which can be found in dictionary.js.

Here are the 5 best and 5 worst words according to my methodology:

Best words:
    serai = 697
    reais = 769
    soare = 769
    paseo = 776
    aeros = 801
Worst words:
    fuzzy = 7875
    hyphy = 8035
    xylyl = 8087
    fuffy = 8158
    gyppy = 8189