A short history of the hexle project

The first code for hexle was written on the 25th of May 2024. I wrote that first prototype because I wanted to explore some of the techniques that another word puzzle by the name of squaredle was using. I knew that squaredle had a much more human touch, and frankly I did not expect that a purely algorithmic approach would ever compete with squaredle in terms of puzzle quality, but that was fine becayse at the time there was no intention of letting anyone actually play it.

Within a few weeks I thought I was done with the project. A basic word generation algorithm existed and I had plonked it up at hexle.xyz to show off to a few friends that I knew played squardle. This iteration of the puzzle was nearly unplayable for a couple of reasons:

1. There was no concept of bonus / normal words
2. There were no letter usage markers appearing partway through the game.

Even with these constraints people did collaborate to finally complete one of the puzzles. The chief strategy at this time was to try absolutely everything that had any remote chance of being a word, since the puzzle was not ‘complete’ until all esoteric bonus words had also been found. The main effort came from Anothermouse and Beetrot.

Because people were actually requesting that a little bit more effort be put in to take the game from mind-bendingly frustrating to play-able, I decided to add the additional parameters that would make the game easier to play.

…It could have also had something to do with the fact that I was completely infatuated with beetrot and as such was trying to impress her.

Chapter 2: Bonus Words and Normal Words

The first iteration of bonus/normal words DID make the puzzles much easier to complete. The basic approach entailed taking the top 10,000 most-used words as counted by google in a study it released some years ago, and then using that as a cutoff under which every word is considered bonus.

Simple, or so I thought.

There were certainly some quirks, for example many forms of a single lexeme would be classified as ‘bonus’ even if their particular form was not intuitively distant from their lemma. People definitely noticed this, especially when confronted with cases in which the singular form was ’normal’ while its plural was ‘bonus’.

This would not have been such a large sin, except that it coincided with another feature that I had decided to release. I had hoped this feature might induce some friendly competition among my friends.

Chapter 3: ‘Friendly Competition’

The appearance of a public-facing leaderboard that ranked each player by how many errors they made had the immediate effect of transforming everyone who played into an enthusiastic amateur linguist. Every day I heard a new theory about how some or other word deserved to be classed as bonus or as normal.

Everyone seemed to be under the impression that if it were not for these errors in the corpus, the scoreboard would be more ‘accurate’ (I.E, they would be higher up)

I would routinely get emails and messages helpfully pointing out these mistakes

sillage is definitely a word. So is regas from a few days ago

~ an email I recieved from a disgruntled player

Chapter 4: What’s Next?

There are nearly 5 million hexle word submissions since I started the project, which is much more than I could have ever anticipated. My parents remain very dilligent players and have made a habit of finishing the daily puzzle before I even wake up, and if for no other reason than that I intend to keep the game up to date as much as possible.

The next major update to the game may focus on board generation and player quality of life. I’ve had plenty of requests for the ability to rotate the board in 45 degree intervals, but my own personal gripe with the puzzles is that they are very random. I would love to find a way to make the puzzles feel more cohesive and less arbitrary - though I know some players actually find this element interesting.