A tight group can still be off-centre
Unstable rate measures the spread of hit errors, not whether their average is centred on the intended time. It is ten times the standard deviation of hit errors measured in milliseconds. Consistently late hits can therefore produce a low UR and poor accuracy. Check average error and the distribution alongside UR; do not interpret a single smaller number as proof that every part of timing improved.
Compare compatible measurements
Use the same client, mods, rate and object selection. Stable's rate-mod UR convention and modern lazer's real-time UR convention differ; do not apply a conversion twice. Also check OD: identical input errors can produce different judgement counts under tighter windows. A partial failed play and a full completion may cover very different sections even when they share a map title.
Practice: contrast simple and complex rhythms
Suggested exercise: choose a familiar map with simple rhythms and a second map with rhythm changes at a manageable speed. Keep settings unchanged and record several attempts. If the simple rhythm stays centred but the complex pattern spreads out, practise recognising and executing those transitions. If both shift consistently to one side across multiple maps, investigate the setup and your timing reference before deciding that a harder map is the solution.
Do not calibrate from one bad play
Avoid repeatedly adjusting offset after individual misses. Write down the original setting before any controlled change, then compare multiple familiar maps under otherwise identical conditions. Keep the change only when it improves the repeatable bias without introducing another problem. If the evidence is mixed, restore the baseline. A wide distribution is not fixed merely by moving its centre.
Keep exploring