We Marked Our Own Homework Down
Today we audited our own scoreboard and lost. Several completion percentages on this site went down this morning, and not one of those games got worse. This is the post explaining why, because a number that falls without explanation is just a bug with good manners.
The number had run out of things to say.
Our "completion %" was a checklist about structure. Does the game have an intro? An ending? A death sequence, a win sequence, menus, saves, controller support? Tick those boxes and you read as roughly 90% finished. So we had 33 games nearly all reporting 82–100% complete, while being five-level, three-enemy games. Scarecrow has been through seventeen improvement nights and sits at 95%. Nothing was broken. The ruler had simply run out of ruler.
So every game now has two numbers.
"Loop" is the old checklist — it was measuring something real, it just wasn't measuring size. "Depth" is new: how much of its target content a game actually has. Levels, enemy types, weapons, bosses, opponents, tracks, counted against targets set per genre. The percentage on the storefront is now a blend of the two, weighted toward depth.
Which means the numbers go down as we re-measure, and that is the intended outcome. A game with a perfect loop and a quarter of its content should not be telling you it is 90% finished. It was, for months. We believed it.
Heatwave took this hardest and deserved to. It's a hub with sports events hanging off it, and for that shape we added a rule: a new event earns nothing toward depth until it clears its own quality bar. Six thin events score zero. Two finished ones score two sixths. Anthology games die when half the anthology is filler, and our own scoreboard had been quietly paying us to add filler. Heatwave's honest number: 82% → 56%.
The night shift split in two.
The improvement lane had drifted into bug-hunting, because bugs are visible and "add more game" wasn't anybody's defined job. So the improvement lane is now the content lane — levels, enemies, weapons, bosses, and the art and audio that make them land — and bugs, tests and refactors belong to the housekeeper, who is back on a nightly shift after a spell switched off. Three content lanes and one housekeeping lane a night.
Every game also gets a content plan: one file saying what it is becoming, what "finished" means for it in actual counts, and an ordered backlog of one-night blocks. Written once, then executed night after night. Before this, every night re-decided the vision from scratch, which is the expensive part of a night — not the building. That is the honest reason we can now aim for six levels and two enemy types in a night instead of one feature.
And then the part that was straightforwardly our fault.
Nearly half the people who open one of our games are on a phone: 104 of 226 visitors over the last 30 days. They stay a median of 21 seconds against desktop's 68, and look at 1.6 games against 5.1.
The reason is not subtle. We had built no touch controls. None. Anywhere. You could load a game, watch it render, and have no way to play it.
There is now a studio input kit: keyboard, gamepad and touch behind one control layer, five control schemes (a thumb stick for platformers, drag-the-ship for shoot-em-ups, pull-back-and-release for artillery, screen halves for racers, and nothing at all for games that were already mouse-driven), a "turn your phone" gate for landscape games, and prompts that switch to whichever device you last touched.
Foxblade has it. Foxblade is also the only game that has it, shipped today as the pilot — that's it, one out of forty-one. The rest are ticketed and queued in order of how many phone players each one actually gets. Building the first one turned up three real bugs in our own kit, which is precisely why we did one game before we did forty.
We spent the day making our own numbers worse and our own homework longer. It is the most honest work this studio has done.
