In a previouis post I discussed my approach to “solving” the game of WORDLE. In the meantime I have discovered quite a few excellent works about the subject, the game has been sold to NYT and even more clones can be found, some quite interesting.

These are a few closing observations.

1. The best word to start a game of WORDLE with is SALET. This result is identical to those presented by Jonathan Olson and in the beautifully videos by 3Blue1Brown:

Jonathan Olson has an approach based on the full game tree, similar to mine. I only have a different method of exploring the tree. After each guess my algorithm orders the possible next guesses in the decreasing order of the number of different color configurations that can be created. This number is a proxy of how much information the guess can provide. The algorithm recursively follows only a limited number of high information guesses. This speeds up the exploration of the game tree. At around 50 (compare with 13000) the results do not change any more, indicating that we have attained the optimum.

2. Notwithstanding some complaints the new version of the game hosted by NYT is not more difficult. The algorithm for choosing the solution of the day is based on the same number of days since some date last year. Also, consistent with the removal of some six possible solutions out of 2315, the best strategy (again starting with SALET) is slightly faster (3.41 average steps compared with 3.42).

3. The romanian version of WORDLE has as optimal starting word CARIT. For the romanian speakers that have no idea what this word means very close behind (at 2.34 average number of steps) is CERTA. The smaller number of steps needed to solve the problem is due to the reduced solution space (1673 compared to the 2315 solutions for the original).