28 Measure, Drill, Retest
Diagnosis tells you what to work on. This chapter is the loop that tells you whether the work did anything, which is the part most practice skips.
Where this chapter comes from. The small-goal method and the baseline that makes it work are ferret_irl's. The skill-cap and consistency split is the community's, stated plainly in a forum reply years earlier. The rest of the loop is this book's synthesis of both with the measurement habits in Ch. 33. An earlier draft credited the whole five-question framework to ferret_irl on the strength of a second-hand summary; his post doesn't contain it, and this one says so. 38,69
Five questions a session can answer
Sessions go wrong when they try to answer all of these at once. Pick one.
| Question | What you are training | Where the chart puts it |
|---|---|---|
| Skill cap | Patterns just past what you can currently play, roughly where accuracy falls to the high 80s or low 90s | The challenge point: difficulty high enough to inform, low enough to execute 69,108 |
| Consistency | Reliable execution of what you can already do | Consistency, built from the three control skills (Ch. 6) |
| Technique | The physical method: grip, tapping, movement economy | Technique efficiency (Ch. 13) |
| Reading | Understanding the pattern rather than recognizing it | Pattern processing and rhythm sense (Ch. 8) |
| Performance | Executing when the score matters | Nerve control (Ch. 14) |
A map that fails you on three of these at once teaches you very little, because you can't tell which one moved. Pick maps where your target question is the hard part and the rest is comfortable.
The loop
- Question. One sentence, answerable: "can I hold aim through 180 BPM spaced streams?"
- Baseline. Three maps that ask that question, played warmed up, scores recorded. This is the step people skip, and without it nothing later means anything.
- Drill. 15 to 30 minutes on the specific failure, at the hardest level where your technique stays controlled.
- Retest. The same three maps, same conditions, same warm-up.
- Record. Misses, accuracy, UR, combo, and where in the map the misses happened. The last one matters most.
- Adjust. Better numbers: raise one difficulty variable. Unchanged numbers after a fair trial: change the drill, not the map. 38,69,148
- Research. Goals work when you can tell whether you met them: specific, difficult goals outperform "do your best" by a clear margin, and feedback on progress is one of the conditions that makes them work. A baseline is what turns "get better at streams" into a goal at all. 148
- Nuance. "Small goal" doesn't mean easy goal. The research favors goals that are demanding as well as specific; performance only levels off when a goal passes the limits of current ability. Small here means measurable, not comfortable. 148
- Research. Expect the retest to look worse before it looks better when you have just raised difficulty; that dip is what the challenge point framework predicts, not evidence the drill failed. 108,110
Goals small enough to check
| Instead of | Use |
|---|---|
| Get better at aim | Cut misses in this jump section from 5 to 2 |
| Become a speed player | Hold 190 BPM bursts for 10 notes instead of 6 |
| Reach 5,000 pp | Raise this map from 96% to 97.5% |
| Get more consistent | Keep UR under 100 on three attempts in a row |
| Fix my fingerlock | Play this burst pattern twice without locking |
- Train skill cap and consistency on purpose, and mix them. Supported. Skill cap means playing above your comfort zone, around 87 to 95% accuracy for the player who wrote it down; consistency means high accuracy and low miss counts inside it. Only skill cap and you skip fundamentals; only consistency and you stay at the same star rating. Both halves of that warning are what this book has been arguing from the chart. 69,108
- Both extremes are wrong for beginners too. Agrees with Korilak. Advice given to a new player in another thread: SS-ing everything is too safe and pass-hunting is too loose; full combo with good accuracy and move up. One reply adds a caveat worth keeping: don't force 98% while your tapping is still forming. 31,71
- If you think your goal is small enough, it isn't; if you think it's focused enough, it isn't either. Supported, with a caveat. His test: a goal should name a map and a number you can compare against a score you already have, like cutting a play from 26 hundreds to 20. Goal-setting research agrees that specific goals with feedback beat vague ones, and adds the caveat that goals should stay demanding: small means measurable, not easy. 38,148
- Big milestone goals mindblock you through the hype you attach to them. Supported. A first 500 pp play is an outcome goal with an audience attached, and attention on the outcome is exactly what the choking research identifies as the mechanism behind falling apart under pressure (Ch. 14). 106,107
- Without a comparison score, play once for a baseline and build the goal from there. Agrees with the top players. Same habit as beating your own locals, stated as a rule. 40
Every goal on the right has a number you can read off the results screen, which means training either moved it or didn't. That's also what makes the collections system in Ch. 33 work: the collection is the record.
Change one thing at a time
The same discipline applies to your setup. A player who changes area, grip, tapping style, skin, mods, and practice routine in one week has learned nothing about any of them, because there's no way to attribute the result. Change one variable, give it enough sessions to adapt, compare against your baseline, then keep it or revert it.
Two things get confused here, and it's worth separating them. Thrashing is changing your area, grip, skin, and practice routine at once because nothing is working; it destroys attribution and it's what every osu! source in this book warns against. Deliberate variation is changing one parameter slightly, on purpose, between practice sessions, and the evidence favours it: a second session with subtly altered force requirements nearly doubled learning compared with repeating the identical task, while skipping the extra session was about 25% worse. The same study is clear that the alteration has to be small. Vary your practice; don't thrash your setup. 35,159
- Keep your normal setup for anything you want to score on. This is the performance condition and it doesn't change.
- In training only, alter one parameter slightly: a few millimetres of area, a small BPM change with osu! trainer, one step of CS or AR, or a map you know played at a different speed.
- Run the drill at the altered setting, then return to normal for the retest in the loop above. The point is the adaptation, not the altered score.
- Keep the change subtle. Doubling your area or jumping 40 BPM is a different task, not a variation, and the research says the benefit disappears at that point. 35,83,159
- A setup can't be evaluated while it's changing. Agrees with Tonehh. One of them copied a top player's area years ago and has moved about 5 mm since; the failure they describe is never letting any configuration run long enough to judge. It's the same rule BTMC applies to stream variables and the same rule this loop applies to drills. 19,34,40
Speed should expose weaknesses, not hide them
If you can't play a lower BPM cleanly, jumping to a much higher one rarely fixes what's wrong; it usually buries it under noise. When you're diagnosing tapping rather than training it:
- Drop the BPM until every tap is deliberate and you can feel which finger is failing.
- Take off anything that smooths over bad inputs while you diagnose. On a Hall effect keyboard that means raising the actuation point or switching rapid trigger off for the session: a key that fires at 0.1 mm will register a mash that a normal switch would have swallowed, which is convenient in play and unhelpful when you're trying to see your own technique (Ch. 17).
- Rebuild control at the lower speed, then climb again in steps.
This isn't an argument against rapid trigger or against playing fast. It's the chart's control-before-speed rule applied to the one piece of hardware that can disguise whether you have the control yet.