I now have enough puzzle solutions to make some meaningful observations about my Worbot program that solves the daily Wordle puzzle. First, here are my Wordle stats (left) and Worbot’s (right).
Obviously, I started tracking Worbot stats a month or so after I started working the puzzle myself, hence the different totals. The stats on the right also don’t show the first few attempts where I tested the program, including the one where it failed to solve PROXY.
As you can see, Worbot has a better record than I do. It has three times as many solves in three attempts than in five. My ratio is about 2 to 1. It plays in hard mode, too; I don’t. In case you don’t know what hard mode is, it means that you must use the letters that have been identified in prior guess each time you guess. Worbot also takes it further and cannot guess a word that uses a letter than has previously been shown NOT to appear in the target word, i.e. the gray letters become off limits. I think the key to its success is the word lists I have programmed it to use. The list it uses for its initial guess is limited to 64 words that: 1) have all or almost all high-frequency letters; and 2) have been verified to appear in the actual Wordle program target word list (2315 words). The program chooses randomly from those 64. It at least has a chance of getting the word on the first try, although it hasn’t succeeded at that yet. Neither have I.
The main list that Worbot uses for all subsequent guesses is my normal master 5-letter word list, but ordered by frequency in English. That master list (over 4,000 words) does NOT use any knowledge of the official Wordle list. So later guesses may be of words that have no chance of being the correct one. Nevertheless, the fact that it searches the list in order of frequency makes it highly likely it will find the most common word that fits all the previous clues from the colored squares. I believe this increases the chance that it will guess the right word in fewer guesses than me, since I make guesses based on instinct and general vocabulary with no knowledge of whether other words may be more common. That’s because I have made the assumption from the relatively low number of words in the official Wordle list that all of the target words are quite common.
By the way, I had fun programming the statistics page to look much like the Wordle one. That was actually the hardest part of the programming.