Atlas of Doors is built as a measured game. Every door you open, every card you play and every run you abandon is a logged decision, so I never have to guess how the game is doing. Three weeks in, here is the whole scoreboard: 6,622 players, 6,638 runs started, 4,894 runs played to an end, and the two leaks that are quietly costing us most. The good parts and the embarrassing parts, same page.
This is the dashboard I look at every morning. It reads straight from the live event stream, not from a spreadsheet I keep by hand.
A run is not a tap. It is a full descent: pick an element, open doors, fight, draft cards, get to the boss or die trying. 4,894 of them were played to a real ending:
| Ending | Runs | Share of finished runs |
|---|---|---|
| Won | 3,136 | 64% |
| Lost | 1,308 | 27% |
| Abandoned | 450 | 9% |
| Total finished | 4,894 | 100% |
4,680 of those 4,894 carry a complete decision trace, meaning a full record of every choice from the first door to the last. That is the north-star metric: not "views", not "installs", but fully labelled play. It is also the unit of the data product behind the game.
Of the 1,025 players who finished a run, 486 came back and finished another one. 222 players have five or more full runs behind them. That is the group the whole game is really being built for.
"Committed player" is a deliberately harsh definition, because a metric you can flatter is a metric you will flatter.
A committed player is someone who:
Active time is measured from the gaps between a player's own events, and any single gap over 3 minutes is capped, so an idle or backgrounded tab never counts as playing.
That makes 157 a deliberate lower bound. The real number of engaged people is higher, but I would rather under-count than lie to myself. Every metric on the dashboard carries its own written definition like this one, so that it cannot quietly drift between versions.
157 people cleared that bar. On a base of 6,622 that is 2.4%, and it is growing: the committed bar has gone up nearly every day of the window, peaking at 61 committed players active in a single day. A game with a real core has a shape like this. A game people bounce off does not.
Raw totals go up simply because more people arrive. The honest question is whether an average player does more today than two weeks ago. They do, on every ratio I track.
| Per active player | Window rate | Daily range | Direction |
|---|---|---|---|
| Sessions | 1.50 | 1.13 to 1.61 | Up |
| Runs started | 0.99 | 0.10 to 1.43 | Up |
| Runs finished | 0.74 | 0.00 to 1.24 | Up |
| Committed share | 2.4% | 0% to 13.9% | Up |
Read the second row again: the average person who shows up now starts roughly one full run. Early in the window it was one run per ten visitors. The updates in this period (a clearer front door, a rebuilt combat screen, a rebalanced card roster) did not just pull more people in, they made the people already here go deeper.
Activation is the share of engaged visitors who actually start a run. It is the first thing a game has to earn.
A third of everyone who genuinely opens the game now starts a descent. At the low point it was a quarter.
Ten points of activation on this traffic is roughly 660 extra people who actually played the game instead of glancing at it and leaving. It is still the cheapest number on the board to improve, because the people are already there. They are standing at the door.
1.2 minutes on Jun 28. 4.4 minutes now. The median web game sits at 5 to 6 minutes, so we are nearly at the genre floor and climbing.
This is the update working. A session used to be "open it, look at it, close it". A session is now "play a run". The next jump has to come from the reason to start a second run in the same sitting.
Here is the number that decides everything, and here is us being bad at it.
Of the players old enough to count, the share who came back the next day. Mobile-web sits around a 10 to 15% floor. We are under it.
The coarser "seen on 2 or more distinct days" proxy. This one is climbing, which is the encouraging half of the story.
A 4% day-1 means 96 out of 100 people who tried Atlas of Doors did not come back the next day. The newest cohort is at 7%, so it is moving the right way, but it is nowhere near the 15% a game needs to compound. Everything on the roadmap below exists to move this one number.
The leakage funnel walks the whole player journey and marks the steps where people leave. Two steps do almost all the damage.
Percentages are share of everyone who opened the game. The right column is the drop-off at that step.
Six out of ten people who see the home screen never even reach the screen where you pick a path. They have not judged the game. They have judged a title screen. That is the single most expensive rectangle in the product, and right now it is asking people to make a decision before it has given them a reason to care.
Then, of the people who start a run and open a door, nearly four in ten never reach floor 2. This is the first fight. Whatever happens in that first combat, some of it is landing as confusion rather than as a challenge. Notice how gentle every floor after that is: 11%, 11%, 9%, 9%, 7%, 5%. Once someone survives floor 1, they mostly go all the way to the boss. The game is not losing people because it is boring. It is losing them at the door.
Both big leaks are in the first 90 seconds. Neither is in the game itself. The content is holding people fine, which means the fixes are cheap and the upside compounds through every number above.
The Descent is the difficulty ladder. Every rung permanently stacks one more handicap, and you only unlock the next rung by winning the current one. Reaching the last rung means you have won the game at every difficulty below it.
| Tier | Runs started | Finished | Won | Win rate |
|---|---|---|---|---|
| Base (Threshold) | 3,295 | 1,774 | 1,086 | 61% |
| Descent I | 779 | 583 | 432 | 74% |
| Descent II | 438 | 382 | 313 | 82% |
| Descent III | 298 | 278 | 231 | 83% |
| Descent IV | 250 | 231 | 189 | 82% |
| Descent V | 215 | 201 | 162 | 81% |
| Descent VI | 185 | 176 | 144 | 82% |
| Descent VII | 176 | 164 | 125 | 76% |
| Descent VIII | 153 | 143 | 109 | 76% |
| Descent IX | 138 | 131 | 101 | 77% |
| Descent X (the final rung) | 257 | 229 | 163 | 71% |
257 runs were started on the last rung, and 163 of them were won. These are people who have beaten the entire core game, then beaten it again nine more times with a new permanent handicap stacked on each time, and then went back for more. Look at the shape of the ladder: it narrows to about 140 runs a rung, and then the final rung jumps back up to 257. People do not climb to the top and stop. They climb to the top and live there.
The win rate rises from 61% at Base to the low 80s in the middle of the ladder. That is not the handicaps being weak. It is survivorship: the only people up there are already excellent, and they are good enough to eat a stacked handicap without noticing. Two conclusions, and I like exactly one of them:
Everything above collapses into one sentence: the game is good, the front door is not, and people do not come back tomorrow. So the next three weeks look like this.
The home screen has to stop being a title card and start being a trailer. Show the game, not the name of the game. Get people into a fight before they get to make a decision. Every point of activation here multiplies into every other number on this page.
Floor 1 is where four in ten new players quit, while every later floor holds fine. That gap is a teaching problem, not a difficulty problem. The first combat has to explain itself while it is being played, without a tutorial wall in front of it.
Retention is not a popup that begs. It is a game you are in the middle of. The daily seeded run, the Descent ladder and the meta-progression all exist for this, and they are clearly not yet loud enough to pull someone back on day 2. That is the hardest and most important problem on the board.
An 80% win rate on Descent VI is not an endgame. The 157 committed players deserve a mountain that fights back.
Because the numbers are the game. Atlas of Doors logs every decision so the design can be argued with evidence instead of taste, and it seems dishonest to use that data to make the game better and then only show you the parts that flatter it. If you have played it, you are in these charts somewhere. Tell me what made you close the tab. That is the most valuable thing anyone can give me right now.