What 283 tracked osu! sessions look like

Figures recalculated 2026-08-13.

TrackMyOsu has recorded 283 osu! play sessions from start to finish across 161 public profiles. Because every session has a deliberate beginning and end — a player clicks Start before playing and Stop afterwards — the resulting dataset answers a question the official osu! profile page cannot: not what a player's totals are, but what actually happens inside one sitting. This page is an analysis of that dataset. The median tracked session runs 101 minutes, and 48% of sessions produce at least one new top play.

What this dataset is, and what it is not

Every figure below comes from sessions belonging to players who have logged into TrackMyOsu and left their profile public. That is 161 players — a small, self-selected sample, not a representative slice of osu!'s player base. People who install a tracker are, almost by definition, more deliberate about their practice than the average player. They are also more likely to be actively trying to improve, since that is what a tracker is for.

So read what follows as a description of how tracked players play, not as a claim about osu! players generally. Where a number is small enough that it could move substantially with a handful of new sessions, that is said outright rather than dressed up. Nothing here is extrapolated, modelled, or smoothed: every figure is a direct count or median over the stored session records, recalculated hourly.

One further limitation matters more than the sample size. A tracked session only exists if the player remembered to start it. Sessions that were played but never tracked are invisible, and there is no way to know how many of those there are or whether they differ systematically from the tracked ones. It is reasonable to assume the tracked set skews toward sessions a player expected to be worth recording.

What a typical session looks like

The median tracked session runs 101 minutes. That number is more interesting than it first appears, because it is a measure almost nobody collects. Total play time on an osu! profile tells you how many hours a player has accumulated over a career; it says nothing about how those hours were distributed. A player with 2,000 hours spread over hundreds of short sittings has had a very different relationship with the game than one who accumulated the same total in long weekend marathons.

Across the tracked set, players average 1.8 sessions each. That average is doing a lot of work and hides a wide spread: tracking is the kind of habit people either keep or abandon quickly, so the distribution tends toward a large group with a handful of sessions and a much smaller group with a great many. Treat the average as a rough centre of mass rather than a description of a typical player.

How pp gain distributes across sessions

The median session moves a player's pp by 0.01. Cumulatively, tracked players have gained 3,415 pp across all sessions that ended higher than they started.

The gap between those two numbers is the most instructive thing in this dataset. pp gain is not evenly spread across sessions — it is concentrated in a minority of them. Most sessions move pp very little or not at all, and a small number move it a great deal. This follows directly from how the rating system works: profile pp is a weighted sum of a player's best 100 plays, so a session only registers if it produced a score good enough to displace one of those. An hour of solid, enjoyable play that sets no personal best is worth exactly zero pp, however much it improved the player.

48% of tracked sessions produced at least one new top play. The inverse of that figure is the more useful one: the majority of sessions end with a player's top 100 completely unchanged. That is normal, expected, and not a sign that anything went wrong. It is simply what improvement looks like when measured by an instrument that only registers new records.

Across all sessions, 805 new top plays have been captured at the moment they were set, along with the play they displaced.

Session length and what it does not predict

The intuitive assumption is that longer sessions produce more pp, and in the crudest sense that holds — more attempts mean more chances to set a record. But the relationship is far weaker than most players expect, and it is not linear. A long session is not reliably a productive one.

The mechanism behind this is worth stating plainly. Setting a new top play requires beating a score that is already near the top of a player's ability. That is far more sensitive to how sharp a player is in a given hour than to how many hours they put in. Fatigue works directly against it: accuracy degrades, timing drifts, and the probability of a personal best in hour four is often lower than in hour one, despite hour four containing just as many attempts.

This is the argument for tracking sessions at all rather than looking at daily or weekly totals. A weekly summary showing twenty hours played and 40 pp gained tells a player almost nothing actionable. Twenty individually recorded sessions, some of which produced records and most of which did not, at least allows the question what was different about the ones that worked to be asked.

When people play

Saturday is the most common day for a tracked session to begin. Weekday distribution is one of the more stable patterns in this data and one of the least surprising: play concentrates where free time concentrates.

It is worth noting what this figure does not establish. Because sessions are timestamped when they start, a session that begins late at night and runs past midnight is attributed entirely to the day it began. Players are also spread across time zones, and the underlying timestamps are stored in UTC, which shifts some sessions relative to the player's local calendar. The pattern is real but its edges are soft.

What this data cannot tell us

Several things people reasonably want from a dataset like this are simply not in it.

• Whether a player improved. pp measures records, not skill. A player can get materially better over a month — steadier timing, cleaner accuracy, more consistent reads — without their top 100 changing at all.
• Why a session went well. Nothing here captures sleep, warm-up, mood, input device, or what the player was working on. Those are almost certainly the variables that matter most, and none of them are recorded.
• What any individual should do. These are aggregates over a self-selected group. A player's own session history is a far better guide to their own play than any median computed across strangers.
• Anything about untracked play. The single largest blind spot, and an unmeasurable one.

How these figures are produced

Every number on this page is computed directly from stored session records: paired snapshots taken when a session starts and ends, plus the intermediate refreshes captured while it runs. There is no sampling and no estimation. Medians are used rather than means wherever a small number of extreme sessions would otherwise dominate the result, which for pp gain in particular is the difference between a figure that describes typical play and one that describes a handful of exceptional nights.

Only public profiles are included. Players who set their profile to private are excluded from every figure here, and no individual player is identified anywhere on this page. The figures refresh hourly; the date at the top reflects the most recent recalculation.

If you want to see what this looks like for your own play rather than in aggregate, the same records are kept per player — start a session before you next play and stop it when you are done. See the guides for how the tracking works, or the glossary for what each stat means.