Deterministic generic instrument¶

examples/synth-plugin exports RAVET Evidence Synth (org.ravet.evidence-synth) through export_plugin!. Its ordinary EvidenceSynth processor has no format or device code. The musical subtractive synth remains in examples/tutorials/src/synth.rs; this example instead makes exact host timing and routing observable. Both are RAVET-owned implementations.

A positive note-on retriggers one 64-sample bipolar click: the first sample is velocity / 4, then its sign alternates. Notes overlap by retriggering one voice; note-off does not truncate the one-shot. It is a test instrument, with no editor or audio input and stereo f32 output. The fixed sample counts do not change with sample rate. Every output sample is initialized.

Notes and validated short MIDI are echoed unchanged at their original offsets, including CC and note-off. MIDI echo is not audio-delay compensated. Prepared output capacity is bounded, processing/reset allocate nothing, and malformed known parameter values or insufficient supplied output capacity reject before DSP/control mutation. An absent output-event sink disables echo. Reset clears the click/delay and restores CC 1 while retaining both parameter values. State uses the generic parameter envelope. The conservative declared tail is 64 plus latency samples (96 by default), with exact silence after the one-shot drains.

Build and check¶

cargo test -p ravet-synth-plugin --locked
python3 scripts/bundle-plugin.py --package ravet-synth-plugin \
  --formats clap,vst3,au,aax,lv2 --validate

Use the SDK settings for formats needing external SDKs; AU/AAX require their supported build platforms. The default output is target/plugins/debug/ravet-synth-plugin.<format>. The native-table test checks exact samples, MIDI bytes/offsets, CC, finite tail, reset and latency-change acknowledgement. The CLAP test checks real ABI rendering, ports, latency/tail and MIDI output, including allocation counts. VST3, AU, AAX and LV2 product fixtures also check this instrument's actual format entry points. The VST3 fixture loads the real product library through the adapter classes and checks hidden CC mapping, preserved note IDs/output offsets, exact audio/tail and latency notification; the official validator separately loads the final module. Its independent adapter mock remains a regression check.

REAPER fixture¶

python3 tools/reaper-validation/instrument.py --format clap \
  --bundle target/plugins/debug/ravet-synth-plugin.clap --reaper /path/to/reaper
# Use --format vst3 or lv2 and that format's bundle for the other cases.
python3 -m unittest discover -s tools/reaper-validation -p 'test*.py'

The sibling of the Delay harness reuses its new work folder, private configuration, scan paths, process timeout and normal-resource snapshot guard. It starts REAPER with that work folder as its current directory. Only native macOS/Linux CLAP, VST3 and LV2 runs are provided by this sibling; its --windows-root route is refused. It never copies a licence fixture or normal REAPER configuration. On macOS its private app uses the distinct bundle identifier org.ravet.reaper-instrument-harness.

The generated MIDI item contains notes at 0.5 and 1 seconds (velocity 127), with CC 1 = 64 at 0.9 seconds. Four two-second stereo 48 kHz renders check:

• pdc: 32 reported samples, with REAPER PDC aligning both click onsets to samples 24,000 and 48,000.
• zero: latency changed to 0, with the same aligned onsets.
• cc_off: latency 32, CC response disabled; both amplitudes are 0.25.
• echo: first-instance audio pins disconnected and its latency 0; a second instance receives MIDI and produces the expected CC-scaled clicks. Its latency is 32. No first-instance audio contributes to this render.

Every rendered sample is checked (tolerance 1e-5 for host PCM rounding), including polarity, amplitude and silence after exactly 64 samples. The second click is 0.25 * 64 / 127 except in cc_off. The script also checks REAPER's reported first-instance PDC field. Two serial instruments with both lookaheads enabled would cause REAPER to compensate two audio delays despite undelayed MIDI echo; the fixture deliberately avoids that ambiguous timing setup. Format fixture checks independently verify generated MIDI bytes and offsets; the REAPER render alone does not inspect MIDI packet bytes or exclude a host MIDI-through route.

See the dated receipt and support matrix. Linux REAPER 7.82 passes CLAP/VST3/LV2. macOS host attempts are recorded as startup failures, with no instrument render evidence. Validators and offline tests do not establish DAW/device support, audible quality, polyphony or real-time deadlines. AAX SDK 2.8 exposes no tail report in this adapter, and LV2 has no exported tail port; the shared processor's finite tail is measured in their audio tests.