Overall · all time
Per run · all time
How runs end
What the orb actually hit, and how far into the run. Ceiling and floor mean the player did not get it turned around in time, which is a control problem. Gate means they were in control and still could not fit through, which is a spacing problem. The two want opposite fixes, so read the Under 3s column first; the line under the table says what share of all runs end that fast. unknown is every run recorded before this shipped, and nothing is back-filled into it.
Where runs end
Every finished run, bucketed by score. From 25 up, each band edge is a score-orb unlock: EMBER 25, TOXIC 60, NOVA 90, PRISM 140, SIRIUS 200. The 200+ bar is everything past the last one.
Orb funnel
Distinct players who have unlocked each orb, in the order the collection lists them. The starter orb reads zero because nobody ever unlocks it — every player begins with it.
Orbs actually equipped
Runs played with each orb. Unlocking one and choosing to wear it are different things — a legendary nobody equips is a reward that did not land.
Arenas flown
Runs played in each arena, from 1.11. DEEP FIELD is where everyone starts.
Daily Run streaks
Players reaching each streak length. PULSAR unlocks at 3 days, SOLAR at 7.
New players a day
First time each device was ever seen. This is acquisition — the only line that says whether anyone new is arriving.
Active players a day
Distinct devices sending any event, by UTC day. Plotted separately from new players rather than on a shared axis, because one is usually a fraction of the other and a single scale would flatten it.
Runs a day
Finished runs. Rising while active players is flat means people are playing longer, not that there are more of them.
Average score a day
Mean score of runs finished that day. Watch the direction, not the value — a rising line means players are getting better, or that only the good ones came back.
When people play
Finished runs by weekday and hour, shaded from quiet to busiest. This is the measured answer to what hour a daily reminder should fire at, and where a live event or a push would land. Counted in UTC by the worker and shifted into your timezone here, so a window spanning a clock change is an hour out on the far side of it.
Platforms and versions
Where installs are coming from.
Versions in the wild
Where players came from
First-touch campaign label, taken from ?src= on the first visit and never overwritten afterwards - so someone who arrived from one link and later opened another still counts once, against the link that actually found them. unknown is everyone who arrived with no label: direct visits, Play installs (Google does not pass the link through to the app), and everybody who played before this shipped. Retention columns show their own denominator and count only cohorts whose return day has fully elapsed; a percentage over a handful of people is not a result.
Cohort retention
Each row is everyone who first played on that UTC day; each column is how many of them were still sending events N days later. Newest cohorts are at the top, so compare rows upwards to see whether retention is improving as you ship changes. Day 0 is normally 100% — anything less means an install record exists with no events on its own day. A dashed cell is a day that has not happened yet, which is not the same as 0%.
Players
Click a row to open everything held about that install.
Batch cleanup
Tick installs in the Players list above, or preview the whole database with the rules below. An install matches if any checked rule applies, including flagged dev and tester installs. Review the exact list before deleting; this permanently removes events, the install record, promo redemptions and internal flags.
Mint codes
A code is twelve characters and gives whatever you tick below. Nothing is decided in the app, so you never need a release to hand one out.
Codes
Loading…
Did the codes do anything?
Loading…
A redemption is not the point — coming back is. "Played after" counts people who finished a run on a later day than they redeemed, so it cannot count the run they were already in the middle of. Anyone with Share Stats off is invisible to this and is reported separately rather than counted as a failure.
Is anyone poking at it?
Failed attempts carry no id of any kind — not an install, not an address. They are here so a spike is visible, not so anyone can be traced.