How Percore scores each sense

The principle

Each session is five rounds. Each round shows or plays a target, takes it away, and you reproduce it. The score is a function of one thing only: the distance between what was there and what you made, in the natural unit of that sense. A perfect reproduction scores 100. A miss so large it carries no information scores 0. The free daily sessions use fixed target ranges; difficulty never adapts in the daily calibration, because a score that moved with the difficulty would not be comparable with yesterday's.

Colour

A single colour fills the screen for a few seconds and disappears. You rebuild it with three sliders — hue, saturation and brightness — so the three dimensions come apart and you find out which one you were actually tracking.

Each of the three errors is normalised to a fraction of its range. Hue is measured as the shortest way round the colour wheel and normalised over 180°, since that is the furthest two hues can be apart. Saturation and brightness are normalised over their own ranges. The three fractions are averaged, and the result is turned into an accuracy from 0 to 100. The colour demo on the home page shows the three errors separately, because the composite hides the lesson: the dimension you were wrong about is rarely the one you expect.

Pitch

One sine tone plays once, somewhere between roughly 140 and 760 Hz. The frequency is never shown. You find it again by dragging a waveform; the tone glides continuously under your finger rather than stepping between notes.

The error is the distance in cents — the logarithmic unit musicians use, a hundred to a semitone — between the target and your tone. Cents are used rather than Hertz so that the same miss costs the same wherever it happens: twenty Hertz out is more than a whole tone at 150 Hz and about a quarter-tone at 700 Hz, and the score reflects that. A near-exact match scores near 100, and being about an octave off scores zero. What is a cent in music? explains the unit.

Duration

A pulse with a low tone runs for an unannounced length between about one and three and a half seconds, then stops. You press and hold anywhere until it feels like the same length; releasing submits. There is no counter and no clock.

The error is relative: the difference between your hold and the target, divided by the target. A quarter-second late on a short interval therefore costs more than on a long one — which is how it feels, and which matches what is known about how timing error scales with the interval. How to improve your sense of time covers that.

Shape

A triangle sits on a grid at some position, rotation and size, held under a slight veil for a few seconds. It returns to the centre and you put it back: drag the body to move it, one handle to turn it, another to resize it.

Three errors — position, rotation and scale — are each normalised and then weighted 45, 35 and 20 per cent respectively. Position carries the most weight because it is the thing people misjudge most and the one that matters most in practice; scale carries the least because a small size error is the least visible of the three. The weighted sum becomes an accuracy from 0 to 100.

The other thirteen modules

Every practice module uses the same principle — the measured distance from the target in that sense's own unit — and every one is a real, scored exercise. Contrast and Brightness score the difference in the level you matched; Rhythm and Tempo score the timing of your taps against the pattern or pace; Direction scores the angle between where the sound was and where you placed it; Distance, Rotation and Motion score the gap between your estimate and the truth in their own units; Balance uses the phone's motion sensors. In Pro practice, difficulty adapts to how you are doing, so a practice score is a score at that difficulty and is kept apart from your daily series.

From rounds to readings

A session score is the average of its five rounds. From there the app builds the following. Everything up to the daily reading is free.

• Rolling average — each sense has a recent average built from its last sessions, which is the number the home screen and the radar show. It moves slowly on purpose.
• The radar — four axes, one per sense, drawn from the four rolling averages. There is deliberately no single composite number, because nobody can feel a blended score.
• Trend — how the rolling average has moved against your own recent weeks. A slip is shown in a muted tone, never in red. You are never compared with other people.
• Drift — leave a sense alone for a while and Percore lowers its confidence in that reading, showing it as an estimate with a confidence figure and the date it was last calibrated. The reading itself is never lowered. You did not get worse because you were busy.
• Which way you lean (Pro) — a score says how close you got, never which side you landed on. Once a sense has enough rounds to be sure, the app reports the sign: sharp or flat in pitch, long or short in duration, paler or more saturated and brighter or darker in colour, larger or smaller and over- or under-rotated in shape. It is the same deviation the scoring already computed, with the sign kept; nothing new is measured, and it is reported only for the four daily sessions, whose targets are comparable day to day.
• The daily reading — one plain sentence after each session, drawn from your own history. When there is nothing true to say yet, it says that rather than inventing something.

Pro adds the weekly reading, which compares two weeks of calibrations in words, and a week-over-week overlay on the radar — last week's shape drawn dashed behind this week's. Both need enough sessions to compare before they open.

Why everyone gets the same targets

The daily targets are the same for everyone on the same day. That is what makes a session comparable: with yesterday's, because the ranges are fixed; and with a friend's, because they faced the same colour, the same tone, the same interval, the same triangle. The Friends tab shows two readings side by side and nothing more — there is no leaderboard, by design.

What the numbers do and do not mean

A 78 on colour is not a rank, a percentile or a statement about anyone else. It is one measurement in your own series, and the series is the point. The scores are not a vision test, a hearing test or a cognitive assessment, and the app does not claim they transfer to anything beyond the thing measured. The scoring is published here, and reproduced exactly in the two demos on the home page, so that the number you see in the app is one you could have checked yourself.