Why Does Note Repeat Drop or Double a Hit When You Switch Subdivisions? A Four-Beat MIDI Boundary Test

The manuals support rate changes, but not one universal re-phase rule

Ableton’s Note manual says a held pad can repeat at a selected beat division and that the performer can switch divisions to make rhythmic variations. Ableton’s Push manual separately documents holding a pad to play or record an even stream at a selected Repeat interval, with finger pressure affecting the repeated notes’ volume. M-Audio lists 1/8, 1/8T, and 1/16 among the editable Oxygen Pro Note Repeat divisions. Apple describes Logic Pro Note Repeat as transforming a held incoming MIDI note and exposes rate, velocity, and gate as distinct controls.

Those are useful facts, but none is a cross-product timing contract for this exact gesture. In particular, the cited documentation does not establish a universal answer to: “When the rate changes on beat 2 while the pad remains held, must the next repeated onset be newly anchored to beat 2?” A system might re-anchor, preserve an earlier phase, defer the change, or behave differently with internal versus external clock. Firmware, app version, transport state, latch state, and the input method can also define separate conditions. Therefore T is a declared target to test, not a specification that every device is obliged to meet.

Berklee’s rhythm lesson distinguishes eighth-note, triplet, and sixteenth-note pulses, encourages moving between rhythmic units, and stresses knowing where beat 1 is. That musical framing explains why the bar-line anchor matters. It does not specify how any MIDI repeat engine must schedule a boundary.

Program target T before touching the rate control

Create T manually in a MIDI editor with Note Repeat, arpeggiation, swing, randomization, and quantization disabled. Use one MIDI note and one short, easy-to-distinguish sound. Place the onsets shown below, then add the right-boundary anchor on the next beat 1. T is the listening and counting reference; it is not a processed copy of C.

Why half-open? A boundary onset has exactly one owner. Beat 2 belongs to [2, 3), never to beat 1’s window; the following beat 1 belongs to the next window, never to beat 4. This prevents one marker from being counted twice and makes “drop” or “double” testable.

• Pitch invariant: every test onset uses the same MIDI note and channel.
• Clock invariant: clock source, transport state, meter, and tempo stay unchanged through T/B0/C/B1.
• Generation invariant: no arpeggiator, second repeat engine, echo, MIDI feedback path, or probability process adds events.
• Reference invariant: T remains untouched after programming; the raw captures remain unedited.

B0 → C → B1 checks fixed-rate repeatability around the switching take

B0 checks each fixed rate in a sustained block before the switching take. C asks the boundary question during one uninterrupted hold. B1 repeats the fixed-rate blocks afterward. This sequence can reveal a setup that does not repeat before and after C, but it cannot rule out a one-off clock, route, control, or performance disturbance inside C. If the system cannot record the rate-change control events, the conclusion covers the complete gesture rather than the repeat engine alone.

1. Record the condition: identify the external product and version, clock source, transport running or stopped, rate-control method, and recording destination. Use momentary or non-latched operation so the physical hold defines C; if the system cannot do that, this exact test does not fit it. Record each rate-change control event and its musical position if the system exposes it; otherwise label the result as a test of the complete gesture.
2. Choose an operable tempo: use a tempo at which you can change the control on the numbered beat without anticipation or panic. Keep that tempo for every artifact. The number is a session condition, not a recommended BPM for everyone.
3. Capture B0: make one separate block for 1/8, 1/16, 1/8T, and 1/8. In every block, start the hold before an unscored full beat, score the next complete beat, and keep holding through one further complete beat. Use a fourth, unscored beat to release away from the scored boundary; require no new onset events only in the following full observation beat. Confirm the four middle-window counts before proceeding.
4. Capture C: begin at 1/8, maintain one physical hold, and change only the rate at beats 2, 3, and 4. Do not release and re-strike to make the target pass. Continue recording past the following beat 1 for at least one full current-rate interval, then release; never use the appearance of a next-1 event as the release cue.
5. Capture B1: repeat every sustained B0 block without correcting C or changing settings. Compare B0 and B1 before interpreting C.

If the product cannot select one of these rates, cannot change rate during a hold, or cannot expose its generated MIDI output, this exact test does not fit that system. Record the limitation instead of silently substituting another division or using audio alone.

Preserve raw MIDI and assign every Note On to one window

Keep T, B0, C, and B1 on separately named clips or lanes. Preserve the original MIDI capture: do not quantize, nudge, trim, consolidate, or replace notes before the audit. If you later need a corrected musical clip, duplicate the raw take and label the edit. This boundary test needs event identity more than cosmetic alignment.

In the unedited capture, count the repeated voice’s onset events as parsed by the recorder. In a MIDI 1.0 view, do not count a velocity-zero Note On used as a release encoding as another onset. Use the ruler and beat lines to assign ownership. You do not need to convert offsets into ticks or milliseconds, and this guide supplies no universal “close enough” threshold. First answer the categorical questions: How many onsets are in the window? Do their beat phases match T? Does the capture continue past the following beat 1, and—only if the target counts and phase already pass—is the right-boundary anchor present and excluded?

Audit C from left to right and stop at the first repeatable mismatch. A later extra onset can be a consequence of an earlier phase decision; diagnosing everything at once obscures the first boundary that changed the stream.

A MIDI onset is not a gate length, velocity, or sample tail

Apple’s Logic Pro documentation exposes repetition rate, velocity, and gate length separately. Ableton’s Push manual says finger pressure can change the volume of repeated notes. Ableton Note distinguishes Trigger playback, which can continue regardless of note end, from Gate playback, which stops with note release. These product-specific examples point to one general diagnostic rule: do not use one layer’s symptom as proof about another layer.

Start with the repeated MIDI voice soloed and keep downstream effects bypassed or unchanged if possible; the event list remains the counting evidence. After the 11-onset checkpoint passes, compare gate lengths, then velocities, then audio. Change only one of those variables on a duplicate. A “double” heard in the speakers with one captured Note On belongs downstream of the repeat-onset question.

The first failed check determines one result code

Read this table from top to bottom and stop at the first matching row. That precedence makes the codes mutually exclusive: U outranks B, baseline checks outrank C, C counts outrank phase, and phase outranks the right-boundary anchor. For D or X, report the first differing window and its signed count difference; do not assign both codes to later consequences.

Use “drop” and “extra” as descriptions of the captured event stream, not as claims about engineering intent. A C result that differs from T may be a documented or undocumented phase policy rather than a fault. The evidence can still be useful: it tells you whether this exact mid-hold gesture is dependable in the recorded condition.

Repeat the full arc before changing the musical phrase

Run a fresh B0 → C → B1 capture once more with the same condition sheet. The boundary target passes only when both sustained fixed-rate baselines show 2, 4, 3, 2 in their scored middle windows and no new onset in the full observation beat after the release-sacrifice beat, C contains the same 11 onsets at the intended beat phases, the capture extends past next beat 1 with its target anchor visible and excluded, and no raw event was manually repaired. If the two runs disagree, report both rather than averaging them into a clean story.

Only after the test is repeatable should you transfer the gesture to a beat. Change one dimension—sound, tempo, clock source, latch state, controller, or rate sequence—and treat that as a new condition with a new B0/C/B1 arc. If the external system does not meet T, workable musical alternatives may include programming the target, using a product-supported automation path, or releasing and re-triggering at the boundary. Test the chosen alternative; do not assume it preserves phase.

Does a C failure prove that the device is broken?

No. It proves only that the captured mid-hold stream did not match the declared target under the recorded conditions. Compare product-specific documentation, version notes, and support guidance before calling the behavior a defect.

Should I quantize C before counting?

No. Quantization can move boundary-adjacent events into another window or hide the phase behavior you meant to observe. Preserve raw MIDI; edit a labeled duplicate only after the audit.

Why is the following beat 1 not onset 12?

The test total covers exactly one bar: [beat 1, next beat 1). The following downbeat is the right-boundary anchor and belongs to the next half-open window. When the target phase otherwise matches, showing it proves that beat 4 was not judged by a cropped ending; it does not test what happens after release.

Can I run the test from audio if MIDI output is unavailable?

You can make an audio observation, but overlapping tails, velocity, choke, and gate behavior can hide or imitate triggers. Label the result audio-only; do not report an 11-onset MIDI pass or fail without MIDI event evidence.