7 Rhythm, Accuracy, and Offset
Rhythm sense
Staying locked to the beat. It can be built outside osu! too, through an instrument or music theory, and it feeds both accuracy and sightreading.
Accuracy
- Means hitting on time (300s) and keeping unstable rate (UR) low. UR is the standard deviation of your hit errors in milliseconds, multiplied by 10.
- Rough guide on simple-rhythm maps: under 100 is good, and strong accuracy players get under 80. Complex rhythms push these numbers up.
- Turn the hit error bar on, under Options, Score meter type, Hit error. It appears at the bottom of the screen and shows every hit as it lands: inside the blue band is a 300, green is a 100, orange is a 50. It is the only feedback in the game that arrives while you can still act on it. 72
- Hover over the performance graph on the results screen to see UR and hit error. It's only available right after playing or watching the replay.
- Audio helps: use loud, responsive hitsounds and set effect volume above music volume so you can hear yourself drifting off the beat.
- The map's timing decides what counts as on time, not the song. Mappers interpret music differently, and older maps often place rhythms the music doesn't have, which is why older maps build your library of patterns.
- If one pattern in an otherwise comfortable map always comes out late, memorize it and deliberately tap a little earlier there.
Fact check: Timing
- Confirmed. UR is the standard deviation of your hit errors in milliseconds, times 10. It measures consistency, not correctness: hitting every note equally late can still produce a low UR. 51
- Lazer. Stable measures errors in map time, so a Double Time UR must be divided by 1.5 (Half Time: multiplied by about 1.33) to compare with nomod. Lazer has measured UR in real time since version 2023.1130.0, so no conversion is needed there. 51
- Nuance. "Under 100 is good" is a community rule of thumb, not an official value. What a given UR means for your 300s depends on OD; see the worked example in Ch. 35.
- Research. People tapping along to a beat naturally land ahead of it by tens of milliseconds (around 30 ms is typical), and a smaller early lean tends to go with less variable tapping. An early-leaning hit error is normal and trainable. 90,91,92
- Research. Musical training measurably improves rhythm perception and synchronization, and percussionists tend to show the sharpest timing of all musicians. 93,94
- Research. Hearing your own taps helps timing only slightly, but delayed feedback clearly disrupts it, and people shown the size and direction of their errors cut their early lean sharply. Keep hitsounds crisp and low-latency, and check your hit error regularly. 96,97,98
Korilak’s take
- At OD 10 you have 19.5 ms either side for a 300. Confirmed. The formula gives ±20 ms, but stable only counts a hit inside a window if the error is below the rounded value, so the effective window is half a millisecond narrower. 31,46
- Once AR 9 is comfortable, play to the song rather than the approach circle. Supported. People synchronize more precisely to sound than to flashes; moving visual cues narrow the gap but sound still leads. 95
- Never practice timing with Autopilot. Supported. Autopilot moves the cursor for you, so you rehearse tapping without the aim it has to coordinate with. A meta-analysis of whole versus part practice backs the textbook rule that tightly linked components are best practiced together. 62,143
- Drumming made his timing innate before osu!. Supported. It matches the musical-training research above. 94
Tonehh’s take
- Don't chase UR; it drops by itself as you improve. Supported. UR measures spread, harder maps with higher OD force tighter timing, and practice curves keep improving while slowing down. He played patterns because he liked them; lower UR came as a side effect. 12,46,124
- Comparing UR across maps rarely makes sense. Supported. Length, rhythm complexity, and comfort all move it. Short maps also give noisy numbers: estimated from 50 hits, a UR is typically about 10% off its true value either way; from 500 hits, about 3%.
- Use the hit error graph to catch over- or under-streaming, then consciously adjust. Supported. Error feedback reduces timing bias in lab studies. 7,97
ScarletStory’s take
- Time to the song, not to the hitsounds. Supported, and it sharpens the usual advice. Every source in this book says to make hitsounds loud, and this one explains what they're for: hitsounds are a correction signal that tells you whether your timing drifted, not the cue you play to. Play to the music, then use the hitsounds and the hit error bar to adjust. That matches both Korilak's rule about playing to the beat rather than the approach circle, and the research showing people synchronize more precisely to sound than to visuals. 31,72,95,97
- A 300, 100, 300, 100, 50 pattern means you're guessing, not keeping the beat. Agrees with the chart. Scattered errors are a rhythm problem, not an offset problem (Ch. 7).
Take: jesse1412's thread: Accuracy
- Practice accuracy on maps you can already full combo, and raise a minimum accuracy floor from 97% toward 99%. Supported. thelewa, one of the most accurate players of that era, argues that chasing an SS over a thousand retries teaches nothing, while holding a floor on maps within reach teaches timing. It is the consistency half of Ch. 28, written in 2012. 66,103
- Use loud hitsounds, effects at 100 and music at 70. Agrees with Tonehh and khz. The same two numbers appear in a 2012 forum reply and in a 2020s video, arrived at separately. Hearing your own taps helps timing, and delayed or buried feedback hurts it. 8,66,96,98
- Practice hard sections slowed down in the editor until you read them perfectly, then speed up. Supported. Same segmentation the whole-versus-part meta-analysis supports, and the same method Tonehh reaches for with practice difficulties (Ch. 14). 8,66,143
- Play some taiko to build accuracy. Experience-based. Plausible through rhythm training rather than transfer of aim; musical and rhythmic training does improve synchronization. 93,94
Drill: Raise the accuracy floor
- Pick three maps you can already full combo comfortably.
- Set a floor: 97%. Play each until you clear the floor, then raise it by half a point and repeat. Stop at 99%; chasing an SS over a hundred retries teaches nothing.
- Check the hit error numbers after each attempt. If your early average is clearly larger than your late average and your UR is low, that's offset, not accuracy. 51,66
Offset
A low UR only means your hits were tightly grouped. You can be consistently early with a great UR and still get 100s; that's the problem offset solves.
| Your hits are... | Local offset (per map) | Universal offset (options) |
|---|---|---|
| Consistently early | Press − before the map starts | Move the slider toward + |
| Consistently late | Press + before the map starts | Move the slider toward − |
Local offset moves in steps of 5 ms, or 1 ms while holding Alt. On the results screen, the left hit-error number is your average early error and the right one your average late error. If early is clearly larger and your UR is low, try a small negative local offset (around −5 to −10) and test from there. Offset only helps once your hits are tightly grouped; if they're scattered, that's an accuracy problem, not an offset problem.
Fact check
- Confirmed. Local offset changes by 5 ms per press, or 1 ms with Alt. Negative values move hit objects earlier, which is why minus fixes early hits; universal offset works the other way. 57,58
- Nuance. The wiki's guidance matches the caution many top players give: on a correctly timed map, leave local offset at 0 unless your hardware adds delay, and read the needed value from hit errors that are consistently off in one direction. 57
Train this with your own data
This chapter trains accuracy. Numbers come from your real top plays (see how on your profile).
Finished this chapter?
Log in to track progressMark it complete for 🍪 100 and track your way to Codex Graduate.
Community discussion
Discussion is not evidence: it never changes the chapter directly. Corrections go through Codex suggestions, where they're checked against sources.
Forum threads about this chapterStart one
None yet — longer questions and write-ups belong in the forums, where they stay findable.