Oh, wait, actually the best Wordle opener is not “crane”…

In this correction video, the creator reveals a subtle bug in his Wordle simulation—duplicate letters in guesses were colored incorrectly, which slightly skewed earlier results and invalidated his original claim that "crane" was the optimal opening word. After fixing the logic, he shows that the "best" opener depends on the method: "soar" leads for one-step information gain, "slain" for an exhaustive two-step search, and "salet" wins a full simulation of all official Wordle games—though "trace" and "crate" are nearly as good and more human-friendly. Crucially, he argues that the technically optimal word is overfit to the official answer list and impractical for real players, since it requires memorizing second guesses for every possible pattern. The video's real purpose, he stresses, is not to dictate how to play Wordle but to teach information theory, the limits of greedy algorithms, and the value of deeper search—skills meant for writing meaningful algorithms beyond casual games.

• ▶ 0:00 The creator confesses to a subtle bug in Wordle simulation code: incorrect coloring for guesses with repeated letters, which affected only a small percentage of cases and slightly skewed results.

• ▶ 0:37 Correct Wordle logic is explained: with a guess like "speed", duplicate letters should be colored based on how many matching letters actually exist in the answer, not all duplicates flagged indiscriminately.

• ▶ 1:21 Root cause: a "really dumb" optimization trick for speed was added mid-project without being fully thought through—and ironically, the final solution precomputes all patterns, making that optimization unnecessary.

• ▶ 2:28 The original conclusion about the optimal opening word was wrong: “crane” only won because the algorithm was playing a slightly different game; after fixing the issue, a different word is theoretically optimal for the official Wordle list.
• ▶ 3:18 Though the video’s real purpose was teaching information theory, the internet took away “best Wordle opener” as the main claim, so the creator feels obligated to issue a correction.
• ▶ 3:43 The original analysis intentionally avoided the official answer list to prevent overfitting, instead using English word frequencies to estimate likely Wordle answers—a meaningful approach worth revisiting as its own lesson.
• ▶ 5:14 Using the official Wordle answer list and one-step expected information, "soar" ranks as the top opener, though it only looks one step ahead.
• ▶ 7:00 Extending to an exhaustive two-step search shifts the ranking: "slain" becomes the best opener, while "soar" drops to 14th.
• ▶ 7:51 In full simulation of all 2,315 Wordle games, "salet" wins by a hair; "trace" and "crate" perform almost as well and are more recognizable real words.
• ▶ 9:11 The technically optimal opener is not best for a human, because it requires memorizing second guesses for every pattern and is overfit to the official Wordle answer list, so any list change invalidates it.
• ▶ 9:46 The real takeaways are to forget the "best" guess and instead remember how to quantify information and recognize when a greedy algorithm fails versus a deeper search.
• ▶ 10:04 The purpose of these algorithmic exercises is not to change how you play casual games, but to build skills for writing algorithms in more meaningful contexts.

• ▶ 0:00 Bug Confession and Root Cause (0:00 - 1:58) - Admits a bug in the repeated-letter coloring rule and explains its root cause and irony.
• ▶ 1:58 Impact, Corrected Conclusion, and Revisit (1:58 - 4:13) - Details the mostly cosmetic effect, the changed optimal opening guess, and why the final analysis deserves revisiting.
• ▶ 4:13 Corrected Analysis and Revised Rankings (4:13 - 8:51) - Recaps the overfitting method and presents corrected expected-information, two-step, simulation, and average-score rankings.
• ▶ 8:51 Optimal Starters and Takeaways for Humans (8:51 - 10:21) - Shares optimal starting words, explains why they aren't best for human play, and gives final takeaways.