Wordle solver

The idea of this is to find the optimal solution to Wordle puzzles.

The current implementation, for each guess, works out the worst case number of possible solutions if we receive a particular result. We can use this to heuristically steer finding optimal solutions.

Solutions

A word can be uniquely determined in 5 guesses (that is, the fifth guess is guaranteed to be correct). There are probably lots of words that can be used as starters to guess successfully in 5 guesses, but I haven't brute-forced that list. I know "aesir" can be used as a first guess.

A word cannot be uniquely determind in 4 guesses - an exhaustive search shows this. This search took circa 11 hours on the 4 cores of my early 2015 MacBook Air, demonstrating the trade-off between ancient hardware, time optimising code, and waiting for a computation to complete.

It's much faster to show that some word can be used as a starting point for a 5-guess solution than to show no 4-guess solution exists, because the former allows a heuristic to find an actual answer quickly, while the latter requires an exhaustive search to show no solution exists.

Next steps

Potential follow-up projects:

• Find the best starting words, for some metric, like "fraction of words that can be solved by guess number 4".
• Build an actually useful solver, rather than something that just says "a solution exists". It would take your guesses and the replies, and list potential solutions and suggested guesses.
• Write up the algorithm in a bit more detail in this README.md.

Sources

The words in the words directory are sourced from the Wordle game itself, so that we're solving the actual real game. The scoring algorithm is reverse-engineered from its JS.