The project’s namesake and spiritual commander.
A Wordle Codebreaking Operation
Named After a Comedian
“At one point, host Kelly Ripa asked the actress who was better at the game, and Burnett stated that she had solved Wordle in one guess seven times.”
— Daily Mail, March 2024
On a Sunday afternoon, Papa dropped a Wordle result into the family group chat. Wordle #988. Three guesses. A tidy green grid.
Serena responded the same day. Then Alex. Then Shannon. Within a week, the entire family was playing. Results were shared daily. Scores were compared. A quiet but unmistakable competition had begun, the kind that starts with “oh I got it in two” and ends with someone not speaking to someone else for a week.
It was wholesome. Competitive, but wholesome.
For about six weeks.
The thing about Wordle is that when you share your results, you share a grid of colored squares. Green means right letter, right spot. Yellow means right letter, wrong spot. Grey means wrong letter. You do not share the actual letters you guessed. Just the colors. The pattern.
The thing about patterns is that they contain information.
Consider: your father shares a result. His first guess produced the pattern green-green-green-yellow-grey. You don’t know what he guessed. But you know that whatever the answer is, there exists a five-letter English word whose first three letters match it exactly, whose fourth letter appears somewhere else in it, and whose fifth letter doesn’t appear at all. Many words have no such partner. Those words cannot be the answer.
Now your brother shares his result. Different pattern. More eliminations. Your sister shares hers. By the time three family members have shared six rows of colored squares, the number of words that could possibly be the answer has often collapsed from 2,300 to fewer than five.
Often to one.
The question was whether this could be automated. The answer, it turned out, was yes.
The first prototype was built in Excel. Not because Excel was the right tool but because Excel was the available tool. The core operation—generate all guess-solution pairs, compute their patterns, filter by observed constraints—is fundamentally a series of joins and lookups. Excel can do joins and lookups. It just can’t do 45 million of them.
The Excel version worked, barely, once you were already down to about a hundred candidate solutions. Getting from 2,300 to 100 required manual narrowing. It was the equivalent of building a nuclear reactor out of graphite bricks in a squash court: functional, ugly, historically important.
The move to Python changed everything. A Jupyter notebook was created. It was named Carol-B, after Carol Burnett, who had recently claimed on national television to have solved Wordle in one guess seven times. The name was aspirational.
Carol-B worked as follows:
Step 5 is the key insight. Each row of a family member’s grid doesn’t just eliminate solutions—it eliminates entire paths of guesses. If someone plays on hard mode (and most people do), their second guess must be consistent with the information from their first guess. This constrains not just the solution but the guesser’s entire trajectory, which further constrains the solution.
With two or three family members’ results, the solution set typically collapses to single digits. With four or five, it’s often exactly one word.
The project’s namesake and spiritual commander.
The data speaks for itself.
| Date | Puzzle | Player | Guesses |
|---|---|---|---|
| Apr 13 | #1029 | Nick | 1 |
| Apr 19 | #1034 | Nick | 1 |
| Apr 24 | #1040 | Nick | 1 |
| Apr 26 | #1041 | Nick | 1 |
| May 2 | #1047 | Nick | 1 |
| May 4 | #1050 | Nick | 1 |
| May 23 | #1068 | Nick | 1 |
| Jun 6 | #1083 | Nick | 1 |
Eight hole-in-ones in eight weeks. For context, the probability of guessing Wordle correctly on the first try is approximately 1 in 2,300. The probability of doing it eight times in eight weeks is approximately 1 in a number with 27 digits.
The Carol-B notebook’s Inputs.xlsx file contains manually entered pattern data from family members’ Wordle results. The dates in that file span April 24 to May 5, 2024—sitting squarely in the middle of the streak.
The family chat, predictably, descended into chaos.
The accusations began almost immediately. How does someone solve Wordle in one guess? Once, maybe—a lucky stab. Twice, suspicious. Three times, implausible. Eight times in eight weeks was not a streak. It was an indictment.
But the beautiful thing about paranoia is that it is not selective. It spreads. On July 23, Sade posted a hole-in-one. On August 13, Frances did too. Were these legitimate? Unclear. The family was divided. Everyone suspected everyone. Alliances formed and dissolved. The group chat became a theater of suspicion in which every green row was a potential act of fraud and every “nice!” was laced with distrust.
Papa, the man who started it all with Wordle #988, posted a hole-in-one on November 27, 2025—a full year and a half later. Was it real? Frances and Sade were skeptical. Papa maintained his innocence. The tribunal remains in session.
The solver was quietly archived. In its place, a leaderboard was built—a public-facing website that tracked the family’s Wordle and Connections scores week by week. Dr. Strehlke’s World Famous Puzzling Leaderboards. The hole-in-ones stopped. Normalcy returned. The weapon was retired; a monument was erected.
But buried in the leaderboard project’s ROADMAP, under “Features,” there remained a single line:
Incorporate solver functionality including Carol machinery
A Cold War reference hidden in plain sight. The warhead, disguised as a weather satellite.
The notebook has been declassified. The original Excel prototype is displayed under glass. The data is presented without apology.
Carol Burnett, dressed as Grand Moff Tarkin, watches over the proceedings.