MATRIX WEB SUITE - TEST8 NAM TO SNAPTONE
GP-150 / GP-180 custom R18 firmware
==================================

Open index.html. All application code, the conversion worker and its WASM
resampler are embedded. No installation, account or cloud converter is
needed. Source NAM files remain unchanged. No firmware or APK is included.

NAM A2 FULL / LITE TO SNAPTONE
1. Open IR / SnapTone / NAM.
2. In NAM A2 -> SnapTone, select your .nam file.
3. Click Convert to SnapTone. Keep the page open; phones may take several
   minutes. Cancel conversion stops the worker without writing to a pedal.
4. Download .clo, or click Use for upload to stage the result above.
5. Connect BLE, Read slots, select a SnapTone slot and enter a 1-15 character
   ASCII name. Click Upload and review the specific overwrite confirmation.
6. Wait for completion and slot-name readback, then select that model for
   an audio test on the pedal. Start with a spare slot.

This is an approximate SnapTone fit, not native A2 Full processing on the
pedal. Supported: 48 kHz, mono, 23-layer A2 Full (8 channels) and Lite
(3 channels) with the supported A2 graph. A SlimmableContainer selects its
Full submodel for fitting. A1, LSTM, arbitrary WaveNet architectures and
different rates are rejected. Do not advertise this as an all-NAM converter.

The independent Local NAM converter and native NAM A2 Lite upload remain
available. They select the Lite submodel instead, keeping the neural model
in the device's native 7980-byte NAMB format. A2 Full exceeds that native
slot format, so use SnapTone fitting for Full.

COMPARISON WITH ORIGINAL SUITE
The complete path was compared against the installed official Suite DLL
using the same real A2 Full reference model and the same generated Matrix
70-second calibration signal. Suite output scaling (0.31), prewarming,
PCM quantization, nonlinear fitting, iterative FIR fitting and export
layout were reproduced. Calibration is generated locally; the official
Suite calibration recording and its DLL are not bundled.

Measured relative RMS differences for that reference:
  NAM calibration output: 0.00181%; maximum difference one PCM16 step.
  Nonlinear curve coefficients: 0.00000939%.
  Pre-filter coefficients: 0.1991%; post-filter coefficients: 0.0182%.
These are numerical reference differences, not measured listening errors
or proof of all-model equivalence. CLO files are not byte-identical.
The resampler version and floating-point FFT implementations differ.
Using Matrix's calibration instead of Suite's bundled recording can also
change the result compared with a normal stock Suite conversion.

VALIDATION AND CURRENT LIMITS
Actual browser conversion of A2 Full completed in about 94 seconds on this
computer. Cancel, restart, valid output, staging into SnapTone and blocked
upload while disconnected were verified. No browser JS errors were seen.
The download link was created and clicked; the in-app browser harness did
not observe a completed file download. Check file saving in your browser.
Bundled-worker tests cover a Full container and invalid inputs. Separate
original ARM64 component fixtures and existing protocol/editor tests pass.
tests.json contains the detailed results and SHA256 of index.html.

Physical GP-150/GP-180 conversion uploads, audio fidelity, iPhone/Bluefy and
Mac have NOT been tested in this change. Earlier web builds were confirmed
by the user on Android Chrome. This remains a community test build.

BLE / LOCAL FILES
Close other Suite apps before connecting. Use a browser that supports
Web Bluetooth (Bluefy on iOS is the intended route). If local-file BLE is
blocked by the browser, serve the same HTML over HTTPS or localhost.
Optional local server, if Python is installed:
  python -m http.server 8000 --bind 127.0.0.1
Then visit http://localhost:8000/index.html on that computer.
Conversion runs locally without Bluetooth or a network connection.

OTHER IMPORTS AND TRANSFER
IR: mono PCM24 WAV, 44.1 kHz, original direct-import limit 683 samples;
truncation is reported. No IR resampling or normalization is added here.
SnapTone: VTIS .clo/.snap/.snaptone type 0/version 0, valid CRC. Converted
to the 2696-byte device representation. Unsupported formats are rejected.
Only the reviewed slot is written. Target state is reread before sending.
Transfers retain the original protocol and ACK checks. Rejection stops
the transfer; there is no automatic retry. Cancellation cannot undo data
already written. Name readback confirms the slot/name, not every payload
byte or audio quality. User-model selection in the main effect dropdown
remains pending; use the pedal or Suite to select the imported slot.

EXISTING EDITOR CONTROLS
Block types, chain order, effect parameters, preset navigation, renaming,
export/save, global settings, Global EQ and EXP/CTRL remain available.
Save stock presets on the pedal before saving them through this editor.
MiniMoog has 47 controls; direct EXP assignment into packed synth cells
is excluded. Use the supported MIDI controls instead.

DEPENDENCIES / NOTICES
Runtime: browser JavaScript, Web Workers, WebAssembly; Web Bluetooth for
pedal communication. No separate runtime installation is required.
r8brain-free-src 7.5 is embedded as WASM (MIT, Aleksey Vaneev), with Ooura
FFT routines and WASI/Zig runtime components. License notices are in
THIRD_PARTY_NOTICES.txt and embedded in the HTML source.
https://github.com/avaneev/r8brain-free-src
https://github.com/WebAssembly/wasi-libc
https://github.com/ziglang/zig
Development used Python, Node.js, Zig, NumPy, Unicorn, Capstone and
pyelftools. These are not required to use the page.
Reference testing used the official NAM Core A2 example (not distributed):
https://github.com/sdatkinson/NeuralAmpModelerCore
