The fairness rule that starved our content

The rule is one sentence. If the letters we hand you spell an obvious word, that word is on the board.

It is the right rule. It is the difference between a puzzle and a puzzle that keeps saying no to words you can see. We have written about what happens without it, which is that you leave words out on three hundred and forty-seven days of a year and never notice.

Then we tried to build a thousand levels under it, and found out what it costs.

How many racks pass

Our levels are built from a rack of letters — six, seven or eight of them — and the board is made of the words that rack spells. Under the rule, you do not choose which words go on the board. The rack chooses, and you either accept all of its obvious words or you reject the rack.

So a rack is usable only if the set it produces is one you can actually lay out: not too many words, not too few, enough long ones to be interesting, with a sensible spread of lengths.

We ran every candidate rack through those bars.

Twenty-one seven-letter racks passed. Four eight-letter ones.

Out of the whole English language. I had expected the number to be smallish and I had not expected it to be countable on your fingers.

Why it collapses so fast

The arithmetic is unkind in a specific way. Adding a letter to the rack multiplies the words it can spell, and the rule says every obvious one of them has to go on the board. So the better your rack — the more pronounceable, the more common its letters — the more words it demands, and the faster it blows past any sane board size.

Racks with common letters spell too much. Racks with awkward letters spell too little, or spell things nobody would call obvious. The window where a rack spells nine to twelve good words and nothing else is narrow, and it gets narrower with every letter you add, which is exactly backwards from what you want, because longer racks make the better puzzles.

There is a second squeeze. Variety. Even if a rack passes, you cannot use it at level 12 and again at level 60, because people notice. So the usable set has to be spread thin across the whole run, and the ones that pass most comfortably are exactly the ones that feel most natural, so they are also the ones a player is most likely to remember seeing.

What we actually did about it

Twenty-five racks is not a thousand levels, so something had to give. The interesting part is what we were willing to give up, because we decided quite early that it would not be the rule.

We relaxed the bars instead, in a specific order, cheapest first.

Plurals of words already on the board came off. If CAT is on the board, CATS being accepted as a bonus word rather than occupying a row of its own is not something anybody experiences as unfair — it is the same word and the player has already had the insight.

Proper nouns went the same way. Names are a grey area in every word game and nobody expects a crossword row for a surname.

Then words below our commonness floor. A word in the dictionary that most people would not produce on demand is not an obvious word, which means the rule never covered it in the first place. That one is less a relaxation than a clarification of what "obvious" meant.

And we widened what counts as an acceptable board shape, allowing a slightly wider range of word counts and lengths than the first pass permitted.

That got us from eighty-five levels to a thousand. Every step is a small retreat from a strict reading, and every one of them was checked against the same question: would a player swipe this word, see it rejected, and feel cheated? If yes, it stays on the board, whatever it costs us.

The ordering is the whole design

What I would carry to any puzzle game is that a rule like ours is not a constraint on the generator, it is a constraint on the content budget, and the two feel completely different in a planning meeting.

"Every obvious word is on the board" sounds like a quality bar. It is actually a statement about how many levels you can ship, and nobody realises that until somebody counts. Had we found this at level nine hundred instead of in a feasibility script, the pressure to quietly start trimming words would have been enormous and the reasons would have sounded excellent.

So count early. Take your fairness rule, apply it with no exceptions, and ask how many levels survive. If the answer is twenty-five, you have not discovered that the rule is wrong — you have discovered the real shape of the work, which is that your content pipeline is a search problem over racks, not a writing problem over words.

Where this leaves us

Honestly: with a supply that still runs out. The relaxations bought a thousand levels, and good long racks thin out well before the end of what we would like to ship. Past a certain depth the choice is between repeating racks further apart than anyone will notice, accepting shorter boards, or finding a different puzzle shape for the later game.

We have not solved that. We have made it visible and bounded, which is a smaller claim than I would like to be making but is the true one.

If you want to see what a rack produces before you commit to it, our word unscrambler does the same job the first stage of our generator does — give it a set of letters and it returns every word they spell, grouped by length. That list is the board you would owe the player.