Nine drum tabs put against their own recordings, with nobody listening to anything

Keep It Greasey, I’m the Slime and the six Shiner BelieveYouMe tabs, scored note by note against isolated stems on 2026-09-08. Detector frozen to the Watermelon cymbal pass: 1024-point Hann, 256 hop, band-limited positive spectral flux, 95th-percentile local peaks, 30 ms refractory, 25 ms one-to-one tolerance.

6 songs scored 26 of 30 lanes separate clock fitted on kick only 2 songs refused, and the reason is data

Why playback was never part of this

The session opened with a question about whether taking the audio-playback gate off would let a Songsterr tab be graded by ear. It would not. Sound leaving the speakers arrives in the room. Everything below reads waveforms as arrays and score files as structured data, and it returns a number for every written note.

Watching a tab play back is worse than useless here, because the playhead reports where the tab believes it sits. A reading taken from the tab, checked against a clock also taken from the tab, agrees with itself and settles nothing about the record.

Every scored lane, all 6 songs

"kick fit" is the share of written kicks landing within 25 ms of a kick-stem onset on the fitted clock. "wrong fit" is the same number after refitting the whole clock, with identical freedom, on a different recording. A song whose two columns are close has not been measured.

Keep It Greasey s604777

Drum track Drums (Vinnie Colaiuta), 248 bars, 4,948 note instances, 501.8 s stem. All five lanes separate from I’m the Slime as the wrong recording.

Two explanations I offered for the low confirm rates, both refuted

The refractory story was wrong. I first reported that 27.8 percent of note gaps fall under the detector’s 30 ms refractory. That counted gaps between events in different lanes, and each lane is detected on its own stem, so a kick and a hat at the same instant are never merged. The statistic that matters is gaps inside the kick lane alone, and it is 4.4 percent (57 of 1,310), with a median kick gap of 321 ms.

The tempo-boundary story was also wrong. Zero-confirm bars sit within two bars of one of the 19 tempo changes 12 percent of the time. Confirming bars do so 12 percent of the time. There is no clustering.

What settled it was rendering the bars and looking. Bar 42 confirms zero of nine kicks and bar 148 confirms eight of eight, and the two look comparable in density and subdivision. The reason most of Greasey’s written kicks find no onset is not established, and no edit to the tab follows from this pass.

I’m the Slime s642618 (Brandon edit)

80 bars, 1,280 note instances, 213.8 s stem. Its score carries 63 tempo automations across 80 bars, so the notation clock is already fitted almost bar by bar.

Shiner, BelieveYouMe

Six tabs, all authored on Songsterr by Brandon chavez, fetched live and scored against the album's split-kit stems. The wrong-recording control for each song is another song from the same album, which is the strictest control available: same band, same kit, same separator.

Asleep In The Trunk s5085156

So Far So s4938058

My Mirror Hates Me s5418315 · revision 7085816 · 114 bpm

134 bars, 2,321 note instances, notation 281.6 s against a 286.9 s song. Kick fit 23.0% against 19.3% on Not Too Much.

Not Too Much s5084025 · revision 6647986 · 85 bpm

122 bars, 1,676 note instances, notation 344.5 s against a 348.1 s song. Kick fit 92.3% against 19.4% on Broken Satellites.

Broken Satellites s5084035 · revision 6648000 · 116 bpm

171 bars, 1,410 note instances, notation 353.3 s against a 362.7 s song. Kick fit 82.0% against 10.2% on Jackie.

Jackie s5418285 · revision 7834599 · 144 bpm

102 bars, 1,052 note instances, notation 169.4 s against a 173.0 s song. Kick fit 41.9% against 2.2% on Asleep In The Trunk.

The one lane that should not be there

My Mirror Hates Me carries 1,196 tom notes in a 2,321 note track, which is 52 percent of everything written, at 4.17 tom strokes per second sustained across a five-minute song. No drummer plays a floor tom four times a second for five minutes. The lane scores 1.23 times chance and separates from the wrong recording at only 1.20 times.

The obvious guess is that a hi-hat part was written into the tom lane. That guess is refuted. Scored against every stem in the kit, the 1,146 notes at MIDI 41 reach 1.25 times chance on toms and 1.19 on the hat, with cymbals, snare and kick all at or below chance. No instrument in the recording supports them. The honest state is that these notes are unexplained, and the density argument stands on its own arithmetic regardless of any clock.

What the stems themselves turned out to be

14 of the 30 Shiner stem files carry a broken length header, and they decode short. The audio inside is real, and the total-samples field is unset, so the length reported by a reader is nonsense and the decoded file is missing time off the end. That is why two songs produced no verdict:

• Asleep In The Trunk has a kick stem holding 39.2 s of a 278.9 s song, which is 14 percent. Four written kicks fall inside it.
• So Far So has a complete stem set and an almost empty drum track: 79 note instances across 140 bars, nine of them kick. That tab was never written.

Nothing was deleted or overwritten. The originals sit untouched and the analysis ran on ffmpeg-decoded copies under data/shiner_wav/.

Back to the open Watermelon question

The Watermelon cymbal lane sits at 1.45 times separation, the weakest cymbal result on record here. Across this session cymbals reached 7.97 times on Keep It Greasey and cleared the bar on 4 of 6 songs. Cymbals are confirmable by this machinery, so the Watermelon result is a fact about that tab. One caveat holds: Watermelon used a hand-fitted snare-anchored clock and everything here used an automatic kick fit. Same detector and bands, different clock provenance.

What the notation says an instrument is, and what the stem says

A notehead shape does not name a drum. Reading bars 42 and 148 by eye I called the difference between their plain X and circled-X heads a cymbal articulation, and that was wrong on both counts. The score data says bar 42 is MIDI 42, Hi-Hat (closed) and bar 148 is MIDI 46, Hi-Hat (open), and the hat stem agrees with the score independently. Every articulation in Keep It Greasey with at least 25 instances, scored against all five stems:

Eleven of the twelve articulations name their own stem as the best match, several by a wide margin: Very Low Tom reaches 5.88 times chance on the tom stem against 0.09 on its runner-up. That is the tab's instrument assignment being confirmed by audio rather than assumed from a lane.

The exception is MIDI 44, Pedal Hi-Hat, at 195 notes. Its best stem is the hat at 1.49 times chance, under the 1.5 floor, with the snare close behind at 1.28. No stem in the kit confirms it. The 21 notes written as Hand Clap were below the 25-instance minimum and carry no verdict.

Shiner: what each articulation is, judged by the stems

The same test run on the four scoreable Shiner tabs. An identity counts as resolved only when the best stem beats the runner-up by 1.30 times, because several calls turned out to be near ties that the method cannot separate. Asleep In The Trunk and So Far So are absent for the reasons given above.

Kick and snare name their own stem on every scoreable Shiner song. Every failure sits in the cymbal family or the open hi-hat. Not Too Much is clean, with six of eight naming their own stem and the other two landing as ties.

One explanation tested and refuted

A crash lands with a kick or a snare in most rock drumming, so the obvious guess was that a 25 ms window cannot separate them. Measured, that guess fails. Cymbal onsets fall within 25 ms of a kick onset only 15 to 21 percent of the time on these recordings, and within 25 ms of a snare onset 5 to 7 percent:

Those rates are far too low to explain a crash lane scoring 3.95 times chance on the kick stem. What is left is a question about how cleanly the Shiner kit was separated, since that album came through a different separator from the Zappa material. That is a hypothesis and it has not been tested here.

Why the Shiner snare stems do not sound like snares

Brandon listened to the My Mirror Hates Me snare stem and said it is not a snare sound. He is right, and the reason turned out to cover the whole album rather than that one file. It also refuted the explanation this page carried in draft, which is recorded below rather than quietly removed.

The measurement that answers it

Splitting each snare stem into five bands and comparing against snare stems the same Moises separator produced on other material, percent of that file's total energy:

The crack is missing. Zappa snare stems from this separator carry 4.5 to 12.7 percent of their energy in the 1 to 4 kHz band, which is the region a snare wire lives in. Every Shiner snare stem on this album carries 1.4 to 3.3 percent, while packing 67 to 77 percent into the narrow 150 to 250 Hz shell band against 27 to 57 percent on the Zappa files. A snare with its shell and without its crack reads as a muffled thud, which is what the ear reported.

A claim this page made in draft, and withdrew

An earlier pass argued that the separator had moved Mirror's snare into its toms stem, resting on a spectral centroid of 4,885 Hz against an album median near 3,950 and on the tom stem running at 2.28 times the album median level. The band table above refutes that. Mirror's snare distributes its energy almost exactly as the other five snares do, and it holds the joint-highest shell fraction on the record. The centroid gap came from a 5.18 percent high-frequency tail pulling a mean that is very sensitive to one, so the centroid was the wrong statistic and the band profile is the right one. No note anywhere was edited on the strength of the withdrawn claim.

What is genuinely specific to Mirror

Level and density rather than identity. Its snare stem is the quietest and sparsest on the album:

Alongside that, Mirror's clock reached only 23.0 percent where Not Too Much reached 92.3, and its cymbals file decodes 26 seconds short. Read the Mirror row in the tables above as the least reliable in the set, and treat its 1,146 floor-tom notes as unexplained rather than as evidence of a moved lane.

Three songs from the newer Moises Pro separation

Brandon pointed at a second folder holding The Alligator, Endless Summer and Lazarus separated by a later Moises build. These ship seven stems apiece with the kit unsplit, so lanes here come from bands inside the one drums file rather than from separate stems. Bands were tightened for that: kick 30-180 Hz, snare 180 Hz-1 kHz below the cymbal region, cymbal 6-16 kHz above the snare crack. That is a new configuration and not the frozen five-stem spec.

Lazarus is a clean tab. Its clock locks at 67.3 percent where the wrong recording reaches 15.7, and all three lanes separate. The other two never got a working clock, and the reason is visible before any audio is touched.

Two of the three tabs do not span their own recording, and the cause is not tempo

The Alligator's notation runs to 0.798 times its recording and Endless Summer's to 1.318 times. An earlier version of this page read that as a wrong declared tempo. That reading is withdrawn. A duration ratio can equally come from omitted sections, written repeats, repeats expanded differently, an alternate recording, a count-in, a fade, a cut, or a repeat-handling bug in the timing parser, and a ratio alone cannot tell those apart.

The metronome stem settles it, because it is an independent musical source rather than a number fitted against the same events being tested:

The metronome agrees with every declared tempo and rejects both claimed values. The parser is not the cause either: the Songsterr measure objects on all three songs carry only rest, signature and voices, with no repeat or jump markers to mishandle. What remains is a coverage difference, with Endless Summer's score running longer than its recording and The Alligator's running shorter. Deciding which needs section landmarks matched against the audio, and that test has not been run.

The clock rule this run settled

The metronome stem, tested as a clock

Moises ships a metronome track, and Brandon asked whether the Zappa songs needed one. They already have one. Tested against the clock this page actually uses, on kick coverage:

The metronome wins on the one song whose clock was weakest, by 1.7 points, and loses badly on the song whose score already carries 63 tempo automations. Seeding the anchor fit with the metronome grid and refining made Keep It Greasey worse still, at 20.6 percent, so that combination was dropped. No verdict on this page changes either way, since Greasey's lanes separate at 3.71 to 7.97 times and a 1.7 point clock gain does not move them.