Observation Deck / POGO Metrics
Research layer · the level cap

One friends list, two level caps

Everything else on this site is your story, built from your own export. This page zooms out: one real friends list, recorded twice — 390 trainers in February 2025, when the cap was 50, and 493 of its 497 friends in August 2026, under today's cap of 80 — every level, every catch, every kilometre walked, plotted to show what the level cap does to the number next to a trainer's name. Find out where you stand against them, and see why the tidy straight-line model this project started with falls apart exactly where most players live.

Today's game → Where you stand →
Two eras, never mixed. Chapters 01–05 describe the 2025 game, when the level cap was 50. Pokémon GO has since rebalanced XP and raised the cap to 80, so those levels are not comparable to the current game — chapter 06 starts the record over with a fresh snapshot of the same friends list under the new cap.
Where these numbers come from. Nothing here is anyone's data export. Every figure about a trainer is what their public profile already shows their friends — level, catches, battles won and distance walked. A profile also carries an XP figure: it is read only to check the game's published level costs against reality in chapter 08, and is never plotted or published per trainer. Chapter 08 adds one source that isn't a person at all — the game's own published tables, XP per level and the Level-Up Research tasks. Handles are replaced with placeholders before the data reaches this site, and any level with fewer than five trainers has its statistics withheld, because at that size an average is just one person's numbers. The mirror of that promise is chapter 03 on the landing page: your own export never leaves your device.
01 — The cohort

Who's in this dataset

Every trainer here came from one person's friends list, recorded as a single snapshot in February 2025. That makes it a real sample of real players — and also a biased one. Both things are worth knowing before you read anything else on this page. The 390 below is that 2025 snapshot in full; the same list's August 2026 re-record — 493 trainers, under the new cap — starts at chapter 06.

Trainers per level n = —
The shape here drives everything else: the sample is top-heavy. Nearly one trainer in five sits at level 50, and only a handful are below level 33. A model trained on this is a model of committed players, not of everyone.
Known limitations
Sampling
A friends list, not a random sample. It skews toward active, social players who add friends, so the least active are largely invisible here.
Level cap censoring
Level 50 was the ceiling in 2025. Trainers kept playing past it but stopped levelling, so for — of this sample, level says nothing at all about how much they've played.
Single snapshot
One snapshot, no timestamps per row. Totals are cumulative and lifetime, so a five-year casual and a one-year grinder look alike.
Assumed units
Distance is recorded as a bare number and assumed to be kilometres. The original snapshot didn't note the unit setting.
02 — The explorer

Level vs. everything else

Each dot is one trainer. Switch the metric, add fitted models, and then change how outliers get handled and watch the fit move. That last control is the interesting one.

Metric
Outliers
03 — The ceiling

The level-50 wall

Level 50 was the last level in 2025. Trainers who reached it carried on playing for years, but the number next to their name never moves again. That single fact puts a hard ceiling on how good any level-based model can ever be.

Spread within level 50 — trainers at the cap
04 — Model report card

How well does any of this actually predict?

This project began as a linear regression of level against each stat. Here is that model marked honestly against two alternatives, and against reality.

Goodness of fit — R², all 390 trainers of the 2025 era
R² is the share of variation the model explains — 1.00 is perfect, 0.00 is no better than guessing the average. Growth in this game compounds, so fitting a straight line to it leaves a lot on the table. A log-linear fit beats the straight line on every metric — though simply taking the median of each level beats both on catches and battles, and none of them can get past the ceiling described above.
The original predictions vs. reality
05 — Playstyle

Ratios don't care what level you are

Catches per kilometre and battles per catch describe how someone plays rather than how far along they are, so they sail straight past the level-cap problem. This is where the dataset is at its most useful.

Catches per km vs. battles per 1,000 catches colour = level
Reading this: far right = catches a lot per kilometre walked (lures, incense, dense city play). Far top = battles heavily relative to catching (raiders and PvP players). Bottom left = walkers who cover ground without farming it. There is no third axis here for a plain reason: catches, battles and distance are the only things a public profile says about how someone actually plays.
07 — Benchmark yourself

Where do you stand?

Enter your own numbers. You'll be ranked against the trainers in your level band — comparing a level 50 to a level 35 tells you nothing, so we don't. The toggle picks the cohort: today's cap-80 record, or the 2025 era for numbers from the old game. All four numbers are on your in-game profile; if you've already built your dashboard, your trainer card has your level and distance waiting.

Your stats vs. the cohort

09Your turn

That's one friends list, read twice under two versions of the same game — 390 trainers in 2025 and 493 in 2026 — and the clearest thing in all of it is how little a level tells you on its own. Your own export says far more — every catch, every raid, every kilometre, every friendship — and it never has to leave your device to do it.

Visualize my journey ▶ Watch the example story What's in your export