Play and solve wordle!

WordList is just a list of words.

By default, it loads the dictionary from /usr/share/dict/words, but var wordler.Loader can be changed to load any wordlist.

wordlist.OptimalGuess() contains the exciting heuristic to choose the best next guess.

Solver will solve your wordle for you!

The interactive solver can be used to solve any wordle; args allow changing word length. It always plays using the "hard" rules.

Puzzler will run a wordle for you; solve it yourself.

Args allow changing word length, number of guesses, and more.

wordler/main connects a Solver to a Puzzler and runs simulated wordle interactions; it's helpful for gathering statistics on solution success rate.

Simulator is for testing. It confirms that Solver and Puzzler score guesses and use them for solving with a reciprocal approach.

All stats are based on 1000 6-guess iterations on 5-letter wordles.

• Using /usr/share/dict:
  • Random guesses:
    • 91.5% success rate
    • Average guesses to win: 4.55 guesses
  • Optimize guess based only on heaviest weighted-averge letter frequency:
    • 79.0% success rate
    • Average guesses to win: 4.91 guesses
    • This is worse than random because words with repeated common letters become optimal -- the first guess is always 'arara', which only includes 2 letters! (The second guess was often 'neese' which is also awful.)
  • Optimize based on most-new-letters-in-guess followed by heaviest weighted-average letter frequency:
    • 96.0% success rate
    • Average guesses to win: 4.21 guesses
• Using Wordler dictionary:
  • Random guesses:
    • 84.3% success rate
    • Average guesses to win: 4.58
  • Optimize based on most-new-letters-in-guess followed by heaviest weighted-average letter frequency:
    • 87.0% success rate
    • Average guesses to win: 4.36 guesses
  • Weight letters based on how many words they appear in instead of total number of times they appear (which is above "letter frequency"):
    • 87.9% success rate
    • Average guesses to win: 4.36 guesses
  • Use weighting, but use wordler's hard mode, which is actually easier than my initial implementation.
    • 89.0% success rate
    • Average guesses to win: 4.34 guesses

• [ ] Optimizations.
  • [ ] egrep ^.ater$ /usr/share/dict/words reveals a pathological case where bater can take 8 guesses. Optimized solver should seek to avoid these pathological cases by avoiding guesses that use a pattern of letters repeated in other words. Another test case: baker is solved ~34% of the time with the current algorithm due to all the ^.aker$ words. NOTE: this TODO is based on dict, not the wordle dictionary.
  • [ ] Running iterator with --iterations=1000 --solution=glass only wins 48.1% of games played.
• [x] Easier Hard Mode. It turns out that my Hard Mode is significantly harder than Wordler's, and following theirs allows additional optimizations. For example, if guess arose results in a and o identified as in the puzzle but in the wrong place, Wordler alows guess atoms to follow. I don't. The optimization would be for solver to maintain a list of possible solutions (my Hard Mode) separate from a list of permissible guesses (Wordler's Hard Mode) and then identify the optimal guess based on the possible solution set. Hard Mode requires merely that: 1. Letters known to be in the right place must stay there. 2. Letters known to be in the puzzle must be included in your guess, but may continue to be in the place known to be incorrect.
• [ ] Fix resource leak in the platform dictionary loader -- iterator slows as it iterates when running with platform dictionary.
• [ ] load dictionary cross-platform