Seven million notes

MAESTRO is about 199 hours of competition piano, recorded on a Disklavier. The MIDI isn't someone's transcription of the score. It's a measurement of what the pianist's hands and feet actually did. So you can ask it questions.

performances	1,276
composers	60
hours of playing	198.7
notes	7,040,163
sustain-pedal movements	9,336,756

MAESTRO v3.0.0, parsed into Iceberg and queried with Trino on2026-09-20. Note that there are more pedal movements than notes.

Nobody plays a chord all at once

Four notes stacked on one line of the score are meant to sound together. They never quite do. Grouping onsets that fall within 50 ms of each other and measuring how wide each group is gives this:

notes	clusters	median spread	90th pct
2	840,928	10.9 ms	32.9 ms
3	403,725	20.7 ms	51.2 ms
4	267,613	26.2 ms	65.1 ms
5	124,377	32.1 ms	79.4 ms
6	61,168	34.1 ms	93.4 ms

Two notes played “together” are usually about 11 ms apart. Six-note chords stretch to 34 ms, and one in ten is wider than 93 ms. That's not untidy playing. It's voicing: you lean on the melody note and let the rest land underneath it.

Worth knowing about the method. Because groups are built by chaining onsets within 50 ms of each other, a group of n notes can't be wider than (n−1) × 50 ms no matter how the piece was played. So the widest values are set by the threshold rather than by the pianist, and aren't shown. The medians sit well below that ceiling, which is why those are the numbers here.

The pedal is where the century shows

The sustain pedal isn't a switch. It moves continuously, and it moves a lot: nine million movements across the corpus. Take the share of playing time the pedal spends past halfway down, average it per composer, and an order falls out that nobody put there:

composer	perfs	pedal down
Alexander Scriabin	35	85.2%
Sergei Rachmaninoff	59	81.1%
Claude Debussy	45	81.1%
Frédéric Chopin	201	77.7%
Franz Liszt	131	73.9%
Robert Schumann	49	72%
Franz Schubert	186	64.8%
Ludwig van Beethoven	146	49.3%
Domenico Scarlatti	31	33.4%
Wolfgang Amadeus Mozart	38	30.6%
Joseph Haydn	40	25%
Johann Sebastian Bach	138	21.7%

Composers with at least 20 performances in the corpus. Scriabin gets almost four times as much pedal as Bach.

That order is roughly chronological, which isn't a coincidence. Bach wrote for keyboards that had no sustain pedal on them at all, so pianists playing him today use it sparingly; the music doesn't expect it. By the time you get to Debussy and Scriabin the pedal is part of how the piece is written, and the sound they were after doesn't exist without it. Two centuries of instrument building, showing up in a GROUP BY.

How this was built

A short Python script turns the MIDI into three Parquet tables: performances, notes, and pedal. They stay separate because pedal is continuous and notes aren't, and squashing one into the other would mean deciding in advance what counts as “the pedal was down”. Spark writes them as Iceberg tables and Trino reads them through the same catalog, so there's one copy of the data and one definition of it. The code and the queries are in the repository.