dashersw/coyopedal/editing-factory-models-and-presets

Factory models and presets

Sources and packing

The build's prebuild step, scripts/pack-factory-assets.mjs, runs tools/models_to_bin.py and tools/presets_to_json.py into build/factory/. The results are flashed to their partitions and also embedded in the app, so an OTA replaces the factory list. Run the script by hand to check packing: node scripts/pack-factory-assets.mjs.

The presets are not packed into anything: presets_to_json.py checks the file and copies it, so the partition holds assets/presets.json byte for byte and esptool read_flash on it gives back something you can read. The board parses it with src/native/storage/preset_json.c, which is also what reads a preset file off a card and what writes one — so a list saved on the pedal, a list in this repo and a list in a folder in the browser are the same document.

Invariants

• Never renumber `coyopedal_fx_block_t` (src/audio/effects.h). Stored presets use its order:

gate, compressor, overdrive, reverb, modulation, delay. Chain order is set in coyopedal_fx_process_*, not by the enum.

• A preset is a name, a profile id, six amp values in tenths of a dB, whether the amp is engaged,

and six blocks of an enabled flag and up to five int16 params. The document is described at the top of src/native/storage/preset_json.h, and tests/test_preset_document.py pins the field widths, the block names and their order across the checker, the parser and the enum. Changing the shape means bumping COYOPEDAL_PRESET_JSON_VERSION and updating the reader and writer in src/native/storage/preset_json.c.

• A preset's profile must name a model the packer knows about. Categories must be known ones.
• Up to 256 catalogue entries and 32 saved presets. Saved user presets live in NVS, which

gea flash never overwrites, so test on a board with existing presets too.

• One pedal per block, so presets carry no algorithm choice.
• Users load original `.nam` files from SD. Parsing, validation and cache preparation happen

on the device (src/native/storage/, src/audio/nam/). Never add a host-side conversion step the user would need. Keep the source file, and check .s3cache contents against it.

• Only 48 kHz, eight-channel A2-Full WaveNet models are supported, including a matching member

of a SlimmableContainer.

Verify

Run npm test (the parser round trip, the prepared-model test against the factory .namb, and the container tests), then npm run build:firmware. Clean-sounding presets can only be confirmed by the user listening on the device.