Notes

What we learned building a word game, written down properly. Word lists, puzzle generators, retention, and the bugs that only show up once real people are playing.

What makes a word puzzle hard, measured

Five of our levels were losing players at two to four times the rate of their neighbours. Fixing them meant finding out which ingredient of a board actually costs time.

· 6 min read

The fairness rule that starved our content

We made one promise about our puzzles: if the letters spell an obvious word, it is on the board. Then we counted how many sets of letters can actually keep that promise.

· 5 min read

We shipped a barcode scanner in a word game

For fifteen builds our Android app carried a barcode reader, two unused font families and a GraphQL client. We had written the line that was meant to keep them out.

· 5 min read

Our daily puzzle left out words for 347 days

A player swipes a word that is plainly there on the wheel, the game says no, and something breaks between you. We had been doing that on almost every day of the year.

· 5 min read

There is no event for swiping an app away

We had a metric meaning "the app vanished during a video". It took three analyses to work out that it meant at least three different things.

· 5 min read

A new Xcode shipped an app that could not launch

The first build we made with the new Xcode died at launch on every device running the new iOS. Nothing of ours ran, so nothing of ours reported it, and it read as churn.

· 5 min read

We asked for notifications earlier and got fewer

Our permission ask sat behind a level about half of new players never reach. Moving it earlier is the obvious fix. We shipped it, and it halved the number who said yes.

· 6 min read

One animation loop costs the same as ten

Left on a desk, our game kept redrawing sixty times a second. We switched off one of the eight animations running on that screen and measured again. Nothing moved.

· 7 min read

Your crash-free rate is lying to you

One Monday our Android crash-free rate fell 8.8 points. The crash was real and the stack trace was real, but almost none of it was happening to anyone with a phone.

· 6 min read

Our retention magic number was a mirage

Players who cleared six levels in their first session came back at roughly double the rate. So we made more players clear six levels. Retention didn't move at all.

· 7 min read

Word search generators print wrong answer keys

We switched off one check in our puzzle maker and ran it twenty thousand times. Thirty-seven per cent of the grids hid a word somewhere the answer key never mentioned.

· 6 min read

The two rules that make a crossword a crossword

Take either rule out of our generator and the grids still look fine. They are not fine: 78% and 98% of them hide a run of letters that nobody wrote a clue for.

· 6 min read

Why your word game’s dictionary is lying to you

Our Scrabble helper once told a player that REQUIZ was their best move. Here is why every open word list will do that to you, and the cheap fix.

· 4 min read

What this is

No roadmap, no release notes, no announcements. Each of these is one specific thing that went wrong while we were building a word game, written up properly: what broke, how often, what we did about it, and what the fix still does not handle.

The word tools are free and the game is the thing all of this is in aid of.