Faust DSP sources, compiled WASM + JSON descriptors, AudioWorklet processors,
and EOCModule subclasses for all three effects.
- meml-4b4: 4-band parametric EQ (fi.low_shelf, fi.peak_eq × 2, fi.high_shelf)
10 params: freq/gain per band, Q for the two mid bell bands
- meml-cpe: stereo feed-forward compressor (co.compressor_mono × 2)
7 params: threshold, ratio, attack, release, knee, makeup, mix (parallel)
- meml-wwc: zita reverb (re.zita_rev1_stereo) with pre-delay and M/S width
8 active params + mod_rate placeholder (zita internal mod not yet exposed)
All three DSPs compile cleanly under Faust 2.83.1. moduleFactory in
eoc-chain-ui.js now returns real instances for eq/compressor/reverb.
- Add AdditiveEngine (48 params) and FMEngine (55 params) wrapping FaustEngineBase
- Remove comingSoon flags from ENGINES descriptors
- Replace stub onSwitch handler with real engine construction + init + setActiveEngine
- Fix setActiveEngine EOC rewire to work with both C15 (_bridge.audioContext) and Faust (_audioCtx)
- Fix _startEocChain audioCtx resolution to support Faust engines
- Fix synth param send loop: use N_OUTPUTS instead of hardcoded SYNTH_PARAM_MAP.length
- Add rebuildHeatmap(paramMeta): resets rawParamValues/_lastSentParams and rebuilds
#heatmap-cells from engine paramMeta; colors from SYNTH_PARAM_COLORS for C15,
_colorFromGroup() hash for other engines
- Replace static SYNTH_PARAM_NAMES/COLORS lookups in buildHeatmap, setHeatmapValue,
and showParamPopup with activeEngine.paramMeta[index] lookups
- Scope MIDI CC localStorage key per engine via midiCCStorageKey()
- Add reloadMidiCCMap() helper; called in setActiveEngine() after engine switch
- loadCCMap/saveCCMap in midi-cc-map.js accept optional key parameter
- Wire EOCChain into audio graph after c15.start() (both start-btn and
quick-play paths); guarded by _eocInited flag so init happens once per
AudioContext lifetime
- getOutputNode() on C15Adapter now returns limiter (last node before
destination) so EOC inserts correctly between limiter and destination
- Expose limiterNode getter on C15Bridge
- Add getCurrentParamValue(i) to EOCModule base class, backed by _paramValues
array that setParam() writes to
- Add _buildParamSliders() to EOCChainUI: collapsible ▶ params toggle per
module row, one range slider per paramMeta entry, calls module.setParam()
on input; labelled "manual" when nispsMode=bypass
- eoc:change listener in a-app.js logs paramCount (hook point for future modes)
- saveState()/loadState() persist eocModules (id, enabled, params) and
eocNispsMode; restore re-adds modules via moduleFactory and re-applies values
- CSS: eoc-params-section, eoc-params-toggle, eoc-param-row/label/slider;
eoc-module-row gains flex-wrap to accommodate params section below controls
Adds EngineSwitcher module with clickable engine cards (active highlight,
coming-soon greyed state, loading bar animation, confirm-on-switch when
training data exists). Wired into the Engine drawer; C15 active,
Additive/FM marked coming-soon with toast. Dock title reflects active
engine; engineId persisted to/restored from localStorage.
- Add WASM ML Engine section documenting the dual-instance architecture
(main thread for inference, worker for training), all C bindings, and
the Dataset/computeWeights JS layer
- Add Debug Probe section documenting window.__nisps API
- Add Testing section with Playwright e2e test infrastructure
- Update output modes (now 4: visual, synth, MIDI CC, audio canvas)
- Document known issue: double loss scaling (meml-ues)
- Note batch inference support in input-heatmap.js
- input-pipeline.js: guard circular clamp against div-by-zero when
input is exactly at center (0.5, 0.5) — dist=0 produced NaN
- snapshot-stack.js: jumpTo() used slice(0, index) which excluded the
target snapshot; fixed to slice(0, index + 1)
- a-app.js: _lastSentParams not resized when MLP output count changes
during mode switch, causing stale throttle state and potential param
flood on first frame after switch
Adds the Faust DSP toolchain infrastructure: placeholder additive and FM DSP
files, build.sh (faust -lang wasm per .dsp), faustJsonToParamMeta() to convert
Faust JSON UI trees into the standard paramMeta format, FaustEngineBase
(SynthEngine subclass wiring init/setParam/noteOn/noteOff through AudioWorklet
messages), and FaustWorkletProcessor base class for concrete engine processors.
Add WasmIML.extractWeights() and WasmIML.createWithWarmStart() to preserve
learned joystick mappings across output-count changes; resizeMLP() now
transfers hidden-layer weights and shared output nodes instead of cold-starting.
InputHeatmap.update() now accepts options.inferBatchFn to evaluate
all grid points (plus the divergence center point) in a single WASM
call instead of 256 separate round-trips. The per-point inferFn path
is preserved as a fallback when inferBatchFn is not provided.
Add evalLoss, inferBatch, and getLayerStats to the window.__nisps
debug probe so Playwright tests and dev console can access the new
WasmIML capabilities. InputHeatmap (Phase 3) is not yet wired into
a-app.js, so batch inference heatmap integration is deferred.
Expose inferBatch, trainEx, moveWeightsEx, evalLoss, and getLayerStats
from the WASM binary into the JavaScript layer:
- inferBatch: batch inference for heatmap sampling
- trainEx: replaces train() with per-iteration loss history capture
- moveWeightsEx: native output pin mask support (removes save/restore hack)
- evalLoss: compute loss without updating weights
- getLayerStats: per-layer weight statistics (meanAbs, maxAbs, dead/sat fracs)
Worker also upgraded to trainEx, returning full lossHistory in payload.
Five new C functions for the WASM module:
- nisps_mlp_infer_batch: N-point batch inference in a single call
- nisps_mlp_train_ex: training with per-iteration loss history output
- nisps_mlp_move_weights_ex: moveWeights with output pin mask to skip pinned nodes
- nisps_mlp_eval_loss: compute MSE loss without updating weights
- nisps_mlp_get_layer_stats: per-layer weight magnitude, dead, and saturation stats
- Add window.__nisps debug probe (gated on ?debug=1) exposing iml state,
getOutputs/getLoss/getWeights/getExampleCount, and action triggers
(thumbsUp/thumbsDown/train/randomise/clearExamples/saveState)
- Fix WasmIML bug: this.dataset was a plain object; import Dataset and
use new Dataset(100) so computeWeights() is available for training
- Fix WasmIML.addExample/clearDataset to use Dataset API methods
- 44 Playwright e2e tests across 4 spec files:
- ml-engine.spec.js: WASM inference bounds, training loss, thumbs
up/down behavior, async training, example capture semantics
- ui-interactions.spec.js: drawer open/close, mode switching,
heatmap bar counts, preset chips, keyboard shortcuts (1/2/Z)
- input-pipeline.spec.js: input→output variation, clamping, joystick
drag, post-training output bounds across the full input space
- persistence.spec.js: URL params (?preset, ?spread), localStorage
round-trip, saveState probe
Add MIDI CC preset system for bundled device configurations:
- midi-cc-presets.js: listPresets/loadPreset/loadPresetFromFile API
- presets/polybrute.json: Arturia PolyBrute CC map
- Preset selector in quick controls bar and MIDI CC drawer
- File import button for loading JSON presets from disk
Bug fixes in a-app.js:
- Destroy IML instances before resizeMLP() to free WASM memory
- Fix randomiseWeights() call (was drawWeights())
- Fix thumbs-up to store rawParamValues instead of post-pipeline outputs
Add Dataset.computeWeights() with three modes:
- global: exponential recency decay (newest examples weighted higher)
- local: spatial suppression of older examples near the current input
- combined: both applied together
IML and WasmIML now compute weights on every train() call using the
active mode. Exposes recencyBias, weightingMode, localRadius properties.
WASM worker path passes sampleWeights through to C++ via the new binding.
Add optional sample_weights parameter to MLP::Train() and the WASM
nisps_mlp_train binding. When provided, weights replace the uniform
1/N scaling per sample — enabling recency, spatial, or any custom
importance weighting without changing the training interface.
Fixes from Opus 4.6 review (C1-C6, I1, I3, I4, I8):
C1: static lastOutputs → per-instance lastOutputsForDelta member
C2: add_example() now on audio thread only (owns iml); worker reads
from mutex-protected staging area (stagedFeatures/stagedLabels)
C3: Worker reads stagedWeightsForWorker (not iml.get_weights()),
eliminating concurrent read/write on iml's MLP
C4: Worker spins on weightsPending before writing pendingWeights,
preventing double-write race
C5: RAND and CLEAR now enqueue Randomize/Clear jobs through worker
instead of directly mutating iml on the audio thread
C6: OSC callbacks stage JSON into oscStagedJson + atomic flag;
audio thread applies in process() (no recv-thread mutation)
Also fixed:
- I4: Separate pendingJob field (enqueueJob no longer overwrites currentJob)
- I8: Removed redundant swapReady atomic
- noiseLevel, cachedNovelty, cachedConfidence now std::atomic<float>
- Worker syncs examples back to iml after training via load_examples()
Replace direct-mutation threading with shadow IML:
- Background thread clones weights from main → shadow IML
- Training and perturbation operate only on shadow instance
- New weights staged in pendingWeights, swapped atomically by audio thread
- Thumbs-down now enqueues Perturb job instead of calling move_weights directly
- Audio thread applies new weights via iml.set_weights() at safe point
- Examples copied to shadow for training, results copied back as weights only
Threading invariant now fully enforced: background thread never writes to
the inference IML. Audio thread applies staged weights between inference calls.
Standalone test exercising IML inference pipeline without VCV runtime:
- Basic inference: all 12 outputs valid in [0,1]
- Input responsiveness: 12/12 outputs change between corners
- Randomize: weights change produces different output mapping
- Spread parameter: measurably different distributions at 0 vs 1
- Expressiveness: 6/6 corner pairs distinct, full range utilized
- Smoothness: 100% of sweep steps are smooth (no binary jumps)
Finding: spread sigmoid saturation effect is architecture-dependent
with small [16,24,16] network (filed as meml-l5a for investigation).
Full spec for MEMLNaut VCV Rack module based on interview:
- 2-8 configurable CV inputs, 12 raw + 5 derived CV outputs
- RL feedback via panel buttons + CV triggers with learn gate guard
- nisps-core C++ engine with background thread training
- User-configurable inference rate (block to audio rate)
- Bidirectional state transfer with companion webapp (file + OSC)
- 10-phase development plan from skeleton to distribution
- OSC pill: add status dot indicator (gray=off, green=connected,
pulsing yellow=connecting), use flexbox for reliable alignment
- OSC pill: add hover state for better affordance
- Synth panel: add separator line above OSC row so it doesn't
blend into the arpeggiator controls
- Bottom sheet: increase expanded height from 55vh to 60vh so
OSC row isn't cut off on mobile
- Help modal: update step 3 to mention the OSC pill in the
bottom bar as the primary connection method
Browser-side OSCOutput module sends parameter values over WebSocket to
a companion Deno bridge script that converts them to OSC/UDP messages.
Enables controlling SuperCollider, Max/MSP, Pure Data, TouchDesigner,
or any OSC-capable software from the NISPS playground.
- Browser module (js/synth/osc-output.js): WebSocket client with
auto-reconnect, throttle (~20fps), and dead-zone filtering
- Deno bridge (osc-bridge/bridge.ts): zero-dependency, compiles to
standalone binaries via deno compile for Linux/macOS/Windows
- Test receiver (osc-bridge/test-receive.ts): terminal dashboard
showing live OSC parameter values with bar charts
- OSC pill button in floating bar for quick connect/disconnect
- Help modal section with platform-aware download, setup guide,
and examples for SuperCollider/PD/Max
- GitHub Actions workflow for cross-platform binary builds
- OSC sends in both visual and synth modes
Design spec for input zoom, compound control axes, pinning system,
exploration noise controls, and visualization enhancements. Covers
the full parameter inventory with implementation phases and open
design questions for experimentation.
Re-apply fixes from ddb6b77 that were lost in the arpeggiator restore:
- play-btn: align-self: flex-start (pin to top)
- Play drawer: trigger on .play-btn:hover instead of parent :hover
(prevents preset select from opening the drawer)
- preset-select: align-self: flex-start (pin to top)
Restore the original arpeggiator as the default sequencer. ShapeSeq
code is preserved but gated behind a URL feature flag (?shapeseq=1):
- Arpeggiator import, state, and all wiring fully restored
- ShapeSeq modules loaded via dynamic import() only when flag is on
- ensureShapeSeqInit() creates engine/viz/UI lazily on first audio start
- animate loop routes inputs to ShapeSeq only when flag + playing
- setOutputMode shows/hides ShapeSeq container only when flag is on
- HTML: ShapeSeq container added but hidden by default
- CSS: ShapeSeq styles added (inert when container is hidden)
All 13 ShapeSeq modules in playground/js/shapeseq/ are untouched.
To test ShapeSeq: add ?shapeseq=1 to the URL.
