Why Did My MIDI Drum Loop Overwrite the First Pass? An Overdub vs Replace Test

“Overdub” and “Replace” are intentions; scope determines the result

Official products agree on the broad contrast but not on every boundary. Ableton documents a looping MIDI overdub that adds played notes to the clip and lets recording be paused while playback continues. Apple describes cycle overdubbing as merging the performances from multiple passes into one MIDI region. Steinberg separates a main MIDI Record Mode—New Parts, Merge, or Replace—from a second MIDI Cycle Record Mode—Mix, Overwrite, Keep Last, or stacked variants. Native Instruments documents Overdub as adding events, while its Replace operation removes events for selected Sounds as the playhead reaches them and then disengages under several state changes.

The practical consequence is simple: “Replace is on” does not yet predict what will be deleted. Apple documents four replacement behaviors that distinguish regions from region contents and continuous erase from input-dependent punch. Steinberg shows that a main record mode and a cycle-lap mode can combine. Treat the visible label as one field in the state sheet, never as the conclusion.

Create a four-event control and freeze every recording boundary

Start in a disposable project or duplicate the real clip, region, part, and project before testing. Use one bar of 4/4 with a 16-step reference. Count a stored drum onset or positive-velocity Note On as one event; do not count its release message as another hit. The timing may remain human, but every intended event needs an unambiguous reference step.

Control totals: C contains K=2 and S=2, so four onsets. A correct additive H pass contributes four new onsets, for eight. K′ means two newly performed onsets on the same Kick voice or Sound as K—not a new instrument—and is used only in R. X is a single, intentionally wrong rim or percussion hit used only for the recovery test.

1. Freeze the range: write the loop start, loop end, bar count, time signature, and whether the right boundary is inclusive in the visible editor. Do not resize it between copies.
2. Freeze the target: record track, clip/region/part/pattern name, selected pad or Sound, input port and channel, armed tracks, monitoring state, and every visible lane.
3. Freeze both mode layers: record the main track record mode and the cycle/lap mode separately when the system exposes both. Also note whether the mode changes automatically at Stop, loop wrap, or focus change.
4. Freeze event transforms: leave Note Repeat, arpeggiation, chance, destructive input processing, and automatic take selection off. Either turn input quantization off or hold one unchanged setting through every copy; it must not become an unlogged variable.
5. Make a recovery point: save the project, then duplicate C into O and R before any experimental recording. Do not create F yet; make it only from the verified O. A history stack is not a substitute for a named copy.

Run four named copies; R must start fresh from C

Do not turn O into R by switching one button after recording. R needs a fresh duplicate of the untouched control so its survivor list is attributable to the replacement operation, not to a previous overdub. Use a slow, comfortable tempo and stop recording before the next lap starts unless the documented behavior explicitly requires a completed lap.

For O, the expected invariant is preservation: if the product’s documented additive behavior is active, the four C events should remain while four H events are added. For R there is deliberately no universal expected total. Because K′ uses the same voice as the existing K, selected-Sound replacement can now reveal itself by removing old K while S survives; wider replacement may remove S too, while input-dependent behavior may preserve old events outside the new-input boundary. Logic’s replacement options, Cubase’s combined record/cycle modes, and Maschine’s selected-Sound behavior show why. Write the observed scope rather than forcing one expected count.

Use counts to catch loss and duplication before listening

A beat can sound plausible while hiding one lost kick or two coincident hats. Inspect the MIDI editor or event list before judging feel. Record onsets by voice, not merely the total: a total of eight could still conceal one missing snare and one duplicate hat.

Also inspect note starts at steps 1, 5, 9, 13, and the loop seam. A note drawn just outside the loop may be hidden rather than deleted. If the editor shows the right eight events but playback sounds doubled, expand overlapping parts and take lanes. Steinberg documents that overlapping MIDI parts can play together; stacking is preservation, not proof that only one take is audible.

The first failed invariant tells you where to look

Read R on two independent axes: voice scope asks whether either snare disappears, while time scope records the exact K1/K9 survivor pair. Choose exactly one row on each axis; a selected-Sound operation can also use continuous deletion through time, so the voice and time labels are allowed to combine.

Match the recovery tool to the size of the mistake

Use Undo when the last recorded operation is known and nothing valuable happened afterward. Use a targeted erase or direct MIDI-editor deletion when one pad or one onset is wrong and the product exposes that precision. Use Replace or Punch only when a bounded span genuinely needs a new performance and the C/O/R test has proven its scope. Use a new part or stacked take when you want alternatives rather than immediate destruction.

Apple explicitly distinguishes region-level and content-level replacement, plus continuous erase and event-dependent punch. Maschine documents a more selective model tied to chosen Sounds and playhead position. Cubase documents both track record and cycle record modes. These are useful examples of why screenshots of one application cannot define a universal workflow. If a device manual is unclear, make a second R variant with no K′ input in the first half and same-voice K′ only in the second; compare old K1, old K9, and both snares to reveal whether deletion follows the record range, the selected voice, or incoming events.

MIDI loop overdub and replace questions

Why did my hi-hat get louder on every loop?

The most direct candidate is that additive recording stayed active and you played the same hat again on later laps, creating coincident events. Count stored H onsets and inspect overlapping parts before changing velocity or the sample. If the editor contains only four H events, continue with the MIDI-path double-trigger test.

Can I replace only the snare without erasing kick and hat?

Only after the product’s documented and observed scope says so. Selected-Sound replacement, content punch, note-range erase, and whole-region replacement are different operations. Test with C and R; if the exact scope cannot be stated in one sentence, edit the snare events directly or record a separate take.

Is an empty replacement pass harmless?

Not universally. Apple documents replacement options that erase across the recording range even when nothing is played, alongside punch options tied to incoming events. Steinberg’s modes also distinguish lap completion and the moment input begins. Silence is therefore part of the test, not proof that nothing changed.

Is MIDI overdub the same as audio loop recording?

No. MIDI stores editable performance events; audio stores a waveform. Akai’s MPC support documentation explicitly limits its Overdub button to MIDI tracks and describes audio layers as separate tracks. A DAW may preserve audio loop passes or take lanes, but that is a different data and recovery model from merging drum-note events.

Should I quantize while running this test?

Either state can work if it is fixed and documented, because the primary evidence is event survival. Turning input quantization off keeps the record operation easier to isolate. If it remains on, do not change its value mid-test and check the raw event count as well as the apparent step positions.

Does FingerDrum use these exact Overdub and Replace labels?

This article makes no such claim. It is a general MIDI recording audit for compatible software and hardware. Consult the current product’s own interface and manual, then use the same C/O/R/F evidence structure to document what it actually does.