The Ranking Test

A critic once put twenty-one Beethoven works in order. Michael Tanner did it for BBC Music Magazine, and like every such list it is one person's opinion published under his own name — which is exactly what makes it usable here. It is not a correct answer. It is a second set of gaps to compare yours against.

Six of those works are played here, forty seconds each, with nothing attached: no composer date, no movement title, no hint of where he placed them. You rate what you hear. Afterwards the instrument reports how far apart your two ratings fell across the four pairs he separated by ten positions or more, and the same figure across the four pairs he placed within three of each other.

Agreeing with him is not measured, and the instrument could not measure it if it tried. The only thing taken from the ranking is the distance between two positions. Which of the two he put higher was never imported, so there is no stored number from which your agreement could be worked out afterwards — not by us, not later, not by accident. Preferring the work he ranked lower costs you nothing. A product that measures how far a famous name moves your ratings cannot also give you credit for agreeing with a famous critic.

Both numbers are read against 3.6 points, which is what rating at random produces — and it produces the same figure on both kinds of pair, because chance does not know which works a critic separated. That is the whole reference point. Two figures sitting together near it show no discrimination in this sitting; two that differ show some, on these clips.

The two numbers are never combined, and the difference between them is never reported. They rest on four pairs and four pairs, built from clips that appear in several pairs apiece, and nobody has sat this instrument twice — so how far the figures wander on their own has never been measured. There is no honest size at which the gap between them becomes a result. Offering one would be inventing the threshold, which is the failure this product spends its existence refusing.

If you already know the music, say so before you rate it. Half of any critic's list is famous, and recognising a work means part of your rating is memory of a reputation rather than the last forty seconds. Those clips are removed before anything is computed. It is taken on your word — nothing checks — and what you recognised is never reported as a fact about you. Recognise enough and you get no number at all, plus a plain statement of why, because the instrument needs at least three usable pairs of each kind and will not print a figure it cannot support.

Two limits are published rather than hidden. A forty-second excerpt cannot carry a critic's verdict on a work that runs forty minutes; forty seconds is longer than anything else here and is still a mitigation rather than a fix. And these six recordings differ in brightness by about ten kilohertz for reasons no ranking caused — one source is a 128 kbps mp3 whose sound stops at 8,624 Hz. Measured, that difference is larger across the pairs he bracketed together than across the ones he separated, so it makes a difference harder to find rather than easier. Both figures, and what they were measured against, are on the Lab.

Your answers stay in the browser you gave them in, like every other result here — the ratings and which clips you said you already knew, never the two figures, which are worked out again each time they are read. Underneath the result is the raw record: what you gave each of the six, which works they actually were, and the gap you left on every pair that counted. It is the only place the six are named, and reading it is the end of your blind sitting.