No description
Find a file
monkey-w1n5t0n dbe0f5d8ba fix(ml): one named example capacity; train() and trainAsync() no longer diverge
Phase 2, S35. Two real defects from one root cause, both confirmed by trace
rather than taken from the audit:

1. Divergence. WasmIML built its TS Dataset mirror with a cap of 100 while
   every addExample() ALSO pushed into the C++ FIFO ring, capped at 128. Since
   train() reads the C++ ring and trainAsync() reads the TS mirror, past 100
   examples the two trained on different datasets — silently.
2. Latent OOB read. nisps_ml_train sizes its sample-weight span by the C++
   side's example_count() (up to 128), but wasm-iml.ts allocates that heap
   buffer from the TS dataset's size (<=100). Once the ring exceeds the mirror,
   the span reads past the end of the caller's allocation.

Fix: name the capacity ONCE as nisps::ml::kDefaultMaxExamples = 128, used by
FixedStorage's default template arg, DynamicStorage's default ctor arg, and the
MLP<> alias (which is the only real FixedStorage instantiation path and carried
its own independent 128 literal — the last copy of this dual truth). Expose it
through nisps_ml_describe and have the TS side read it instead of hardcoding.
Dataset's constructor default is removed entirely: a default was what invited
this bug class, and the sole call site now always supplies the describe() value.

ABI NOTE: this extends nisps_ml_describe from a 6-int to a 7-int descriptor.
nisps_ml_describe always writes 7 ints regardless of the caller's buffer, so
every call site had to grow in the same change or it would overflow the WASM
heap by 4 bytes per call. All five sites updated: three in wasm-iml.ts (init
defaults, init per-instance, reshape re-describe — the finding said there were
two), one in wasm-worker.ts, one in tests/cpp/parity_wasm.mjs. The parity
harness's expected-dims check now also pins the new max_examples slot.

Regression test: tests/cpp/test_mlp_storage_defaults.cpp — pins the two storage
policies to one constant, and drives MLPCore<DynamicStorage> exactly as
bindings.cpp does past the old TS cap, asserting it saturates at 128 and not at
100. Fail-before/pass-after confirmed by temporarily setting the constant to
100: 2 failures, named. Reverted: green.

Audit correction: the cited dataset.ts:81 is the FIFO eviction check; the
hardcoded default was at dataset.ts:45.

Gates: run-all-tests.sh ALL GREEN, parity PASS.
2026-07-21 13:22:38 +02:00
.github/workflows chore: delete retired-playground artefacts and root relics 2026-07-21 12:47:58 +02:00
.vscode memlnautmodes, channelstrip 2025-11-19 15:17:37 +00:00
assets/media docs: restructure design docs into docs/specs (adr/plans/recon), update path references 2026-07-13 23:15:46 +03:00
codegen feat(codegen): P5 — TS emission targets manifold/src/modes/generated/ 2026-07-18 12:28:16 +02:00
docs docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep 2026-07-21 12:49:39 +02:00
firmware refactor(firmware): delete vendored daisysp and the input_router layer 2026-07-21 12:49:25 +02:00
manifold fix(ml): one named example capacity; train() and trainAsync() no longer diverge 2026-07-21 13:22:38 +02:00
nisps fix(ml): one named example capacity; train() and trainAsync() no longer diverge 2026-07-21 13:22:38 +02:00
schemas feat(slp-workshop): new MEMLCelium-based mode + port Jolt & OU-noise RL learning 2026-06-28 22:15:36 +02:00
scripts refactor(firmware): delete vendored daisysp and the input_router layer 2026-07-21 12:49:25 +02:00
src refactor(firmware): delete vendored daisysp and the input_router layer 2026-07-21 12:49:25 +02:00
tests/cpp fix(ml): one named example capacity; train() and trainAsync() no longer diverge 2026-07-21 13:22:38 +02:00
vcv chore(vcv): delete the dead test rig 2026-07-21 12:49:25 +02:00
.envrc chore: add .envrc + playground package-lock 2026-06-28 04:14:30 +02:00
.gitignore feat(manifold)!: P1 — retire playground/, manifold is the sole browser app 2026-07-13 23:27:56 +02:00
.gitmodules ci: restore verification — reachable submodule pin, codegen + WASM freshness gates 2026-07-21 11:57:32 +02:00
AGENTS.md docs+ci: P5 doc sync; codegen golden wired into run-all-tests stage 5 2026-07-18 12:45:45 +02:00
ALIGNMENT.md docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep 2026-07-21 12:49:39 +02:00
CLAUDE.md chore(agents): converge project instructions 2026-07-13 22:58:26 +02:00
LICENSE Initial commit 2025-04-10 14:17:18 +01:00
MAP.md docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep 2026-07-21 12:49:39 +02:00
README.md feat(manifold)!: P1 — retire playground/, manifold is the sole browser app 2026-07-13 23:27:56 +02:00
synth-midi-cc.json feat(midi-devices): canonical external-synth CC templates + dual codegen 2026-06-28 20:03:17 +02:00

Neural Interactive Shaping of Parameter Spaces

https://musicallyembodiedml.github.io/memlnaut/approaches/nisps

Firmware

The hardware firmware targets the MEMLNaut RP2350 build and uses repo-local helper scripts for the known-good build configuration:

git submodule update --init --recursive

scripts/build-firmware.sh
scripts/flash-firmware.sh
scripts/build-and-flash-firmware.sh

Notes:

  • The scripts build for rp2040:rp2040:solderparty_rp2350_stamp_xl with Optimize3.
  • The build forces C++20 because the firmware uses std::span and concepts.
  • build-firmware.sh accepts an optional variant name such as MEMLCelium or BreakOr. Matching remains case-insensitive, so memlcelium still works. If you omit it in an interactive shell, the script parses MEMLNaut-NISPS.ino, prompts for a variant, and rewrites the active MEMLNAUT_MODE_TYPE before building.
  • flash-firmware.sh accepts an optional mountpoint argument, or auto-detects common UF2 bootloader mounts such as /run/media/$USER/RP2350 and /run/media/$USER/RPI-RP2.

Manifold (browser app)

Try NISPS in your browser — no hardware required. Manifold is the React front-end running the same C++ engines + ML as the firmware, compiled to WASM:

cd manifold
bun install
bun run dev

Staging deployment: https://meml.lnfinitemonkeys.org/next/

Train a neural network to map input gestures to synth parameters through interactive machine learning: place examples, or use verdict-based feedback (explore-and-place, geometric dislike).

(The former SolidJS playground was retired in July 2026 — archived on branch archive/playground-solidjs, tag playground-solidjs-final.)