# One Apple MIDI synth remap check — collected 2026-10-17

This is an original artificial MIDI / instrument control, not a transcription,
editor round trip, default Apple piano or general DAW / hardware certification.
The actual receiver was **Apple: AUMIDISynth**, aumu/msyn/appl,
reported component version 65536 on macOS 26.6.2 (Build 25G83).
The input instrument is the included original CC0 triangle sample and minimal
SF2 bank0/preset0. No system soundbank is included or was loaded for these renders.

## Three controls

| File | Note raw channel (display) | CC64 raw channel (display) |
|---|---:|---:|
| A-matched-ch1.mid | 0 (1) | 0 (1) |
| B-note-only-ch2.mid | 1 (2) | 0 (1) |
| C-matched-ch2.mid | 1 (2) | 1 (2) |

All three are 65-byte Type1 files with two tracks, PPQ480, tempo500000
microseconds per quarter. Note60/velocity80 starts at tick0 and ends at
960 (1 second). CC64=127 at0 and CC64=0 at1920 (2 seconds). Each EOT is
3840 (4 seconds). A→B changes only the note status channel bytes at offsets
47 and52; B→C changes only CC64 status channel bytes at43 and57.
The independent raw-SMF decoder cross-checks all events with Mido, without
importing the generator as an oracle.

The original 48kHz mono input sample contains44 triangle cycles /4800 frames
(root69,440Hz). The SF2 loops it and sets releaseVolEnv to -2400 timecents
(nominal0.25s input setting). The native synth, not Python, processes the
messages, envelope and returned audio. The SF2 setting is not a claim that
all receivers produce an exactly0.25s acoustic decay.

## Measured result and limits

Three fresh native instances used the same bank/preset and setup on both
channels. Each accepted every required setup/event/render call and produced
240000 stereo Float32 frames at48kHz (5 seconds). EOT4s is not an audio-stop
message; the fifth second was captured for release. AudioUnitRender returned
the samples written to the included WAVs, with no Python output oscillator,
post gain or envelope. MIDI was scheduled at samples0,48000 and96000 by
splitting render buffers at event boundaries.

The complete A and C WAVs are byte-identical. All three share identical
first-second PCM. Between1.25 and1.75s, A/C RMS is0.03617873371485977;
B is exact zero. The last sample above absolute1e-7 is2.2185s for A/C and
1.2181458333333333s for B. These figures describe this component, patch,
preset and host setup. They do not verify real key release, pitch recognition,
Omni/multiport routing, a DAW, half-pedal behavior or every MIDI receiver.

Three optional calls were rejected: Disable internal reverb property(-10879),
Set global volume0dB(-10878), Set reverb volume-120dB(-10878). Their failure
statuses remain in the portable measurement JSON. Identical accepted CC91/93
zero messages are not proof that every effect was disabled. No such claim is made.
Subjective listening is unverified. Detailed actual MIDI, native calls, clock,
component and numerical results are in apple-receiver-measurements.json.

## Read-only checks (default)

Python3.12.14 was used. The authoring checks used Mido1.3.3; WAV checks used
NumPy2.5.3 and SoundFile0.14.0. These are existing tools, not bundled dependencies.

```
python verify.py       # requires Mido; reads the three MIDIs, writes nothing
python verify-sf2.py   # standard library; reads the original bank/sample
python measure.py     # requires NumPy and SoundFile; reads native WAVs
```

These commands do not instantiate a synth, send MIDI or change your files.
To author replacement input fixtures in a separate scratch copy, explicitly run
`python generate.py` (standard library; writes its five input files there).
The generator creates the input sample/bank, not the output WAVs.

## Optional explicit native rerender on compatible macOS

The original Swift source is included; no executable, Apple component, SDK,
system bank, account data or model is distributed. Apple Swift6.4 and macOS SDK27.0
were used. Exact native output on other OS/component versions is not guaranteed.
The following commands are explicitly optional and create files in your copy:

```
python verify.py --write-schedules schedules
swiftc render-apple.swift -o render-apple
./render-apple msyn cc0-triangle.sf2 schedules/A-matched-ch1.mid.schedule.json A-new.wav A-new.json
```

Check the Swift usage before running. A native call can fail in a sandbox or on
another OS; preserve the returned status instead of substituting a synthetic
receiver. The generated WAV must come from actual AudioUnitRender samples.

## Source and license

Original MIDI, sample, SF2, render/measurement source, measurements and WAV
artifacts in this folder are dedicated under CC0 1.0 (LICENSE-CC0.txt).
That dedication does not relicense Apple's software or any third-party Python
package. Those components are not included.

Protocol background: https://midi.org/about-midi-part-3midi-messages
and https://midi.org/midi-1-0-control-change-messages . MIDI channel-message
scope explains the control; the reported receiver result comes from the actual
included native-render evidence, not from the protocol page alone.
