2026-04-15 17:32:32 +02:00
# MAP
2026-07-13 23:27:56 +02:00
MEMLNaut-NISPS — Neural Interactive Shaping of Parameter Spaces. One C++20 codebase (`nisps/`) compiles to two targets: (1) Arduino/RP2350 firmware for the MEMLNaut hardware, (2) WASM in the Manifold React browser app that runs the same engines + ML through an AudioWorklet. (The former SolidJS playground was retired 2026-07-13 — branch `archive/playground-solidjs` , tag `playground-solidjs-final` ; the browser-only C15 engine lives only there for now.) See `CLAUDE.md` for the long-form architecture narrative and `ALIGNMENT.md` for current strategic gaps.
2026-04-15 17:32:32 +02:00
## Layout
2026-04-29 18:57:27 +02:00
### `nisps/` — platform-agnostic C++20 library (the only ML/DSP/engine code)
2026-07-21 14:03:07 +02:00
- `nisps/core/` — `perf.hpp` (hot-path/inlining attrs), `types.hpp` , `concepts.hpp` (`MLEngine`, `AudioEngine` , `Mode` ), `ring_buffer.hpp` (SPSC lock-free cross-core channel, replaces pico/util/queue), `event_queue.hpp` (single-threaded in-engine event FIFO — deliberately NOT RingBuffer, which is an atomics-based cross-thread channel), `rng.hpp` (xoshiro256+ deterministic), `math.hpp` (fast_sigmoid, `Curve` enum + `apply_curve` ).
refactor(ml): one runtime-configurable training default (S26)
The operator's call: "there should be one default learning rate and one
default max iterations and they should both be configurable at runtime."
There were SIX copies, not the four the audit described, and they did not
agree:
nisps/ml/mlp.hpp no-arg train() hardcoding 1.f / 1000u / 0.001f —
and firmware's ONLY training path calls exactly
this, so firmware had no runtime knob at all
wasm-iml.ts train() and trainAsync() TS default params (x2)
engine-api.ts learningRate ?? 1.0, with no maxIterations knob
vcv/src/iml.hpp 200 / 0.1 / 0.00001 — silently divergent
external_synth_midi.hpp its own kDefaultLearningRate/kDefaultMaxIterations
schemas/modes/*.json x9, identical, read by nobody at runtime
Now: schemas/ml_defaults.json is the single declaration (validated against a
sibling meta-schema, matching the midi_device.schema.json convention), codegen
emits it to C++ and TS in the same run, and MLPCore carries a TrainConfig whose
default member initialisers read the generated constant.
set_train_config()/nisps_ml_set_train_config() make it runtime-overridable on
every target; the explicit-argument train() overload is untouched. min_error
joins the tuple — it was duplicated identically and belongs with the other two.
The per-mode ml block loses default_learning_rate/default_max_iterations.
default_spread stays (genuinely wired on both targets) and input_channels stays
(codegen-time validated, real information for sound_analysis_midi).
VCV BEHAVIOUR CHANGE, deliberate: MEMLNaut.cpp constructs IML positionally and
relies on those defaults, so the module moves to 1000/1.0/0.001 — 5x the max
iterations, 10x the learning rate, and a 100x looser early-stop threshold. The
old values were never justified anywhere; they arrived with fbc68eb alongside
an unrelated module rewrite and no tuning rationale. Firmware and WASM have
shipped 1.0/1000 all along. It is now runtime-settable if this turns out worse.
The generated header lands in nisps/ml/generated/, not nisps/modes/generated/
where the rest of codegen output lives: training hyperparameters are an ML
fact, and nisps/ml sits below nisps/modes, so emitting them there would make
mlp.hpp include upward. The agent that built this flagged the directory-crossing
rather than hiding it; this is the fix. CI's generated-freshness gate learns the
new directory.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS (max delta 2.38e-7),
lint clean, manifold typecheck + 17 unit + 33 e2e (which exercise train() and
trainAsync() through a real browser).
2026-07-21 17:20:10 +02:00
- `nisps/ml/` — the MLP core, written once against a storage policy (`mlp.hpp` `MLPCore<Storage>` ): `storage.hpp` (`FixedStorage` — template-sized `std::array` , zero heap; `MLP<NIn,NH1,NH2,NH3,NOut>` alias preserves the classic compile-time surface) and `dynamic_storage.hpp` (`DynamicStorage` — runtime dims, single arena alloc at construction; `#error` s on RP2350 builds, sole lint heap-allowlist entry). Fixed↔dynamic bit-parity enforced by `tests/cpp/test_mlp_storage_parity.cpp` . Files: `mlp.hpp` , `activations.hpp` , `loss.hpp` (MSE, no double-scaling), `training.hpp` (SGD + grad clipping), `init.hpp` (spread-aware uniform↔Xavier), `rl.hpp` (`move_weights` with output pin mask + per-layer scaling + weight decay), `jolt.hpp` (`Jolt` — held continuous weight-morph over the flat weight buffer + post-release LR ramp; ported from upstream InterfaceRL), `ou_noise.hpp` (`OUNoise< N > ` — Ornstein-Uhlenbeck exploration walk on the output vector; ported from upstream InterfaceRL), `feedback.hpp` (`FeedbackControllerCore< FbStorage > ` — the "Down Action" state machine: Avoid (geometric push-away default / Diffuse legacy) / RandomiseOutputs / RandomiseMlp / ExploreAndPlace; storage-policied like the MLP, own deterministic RNG, exposed via `nisps_ml_feedback_*` C API), `replay.hpp` (`ReplayView` — reward-tagged memory: dedup/deepen, k-NN positive centroid with deterministic tie-break, proportional decay+eviction), `geo_push.hpp` (push-away target computation, upstream InterfaceRL @ 0a541cc), `warm_start.hpp` (overlapping-weights copy for reshape), `stats.hpp` . `generated/ml_defaults.hpp` is codegen output (do not edit): `nisps::ml::generated::kMlTrainDefaults` , the ONE learning-rate / max-iterations / min-error default shared by firmware, WASM and VCV (source `schemas/ml_defaults.json` ); `MLPCore::set_train_config()` and `nisps_ml_set_train_config()` override it at runtime. It lives under `ml/` rather than `modes/generated/` because `nisps/ml` sits below `nisps/modes` — mlp.hpp must not include upward. Jolt + OU are inert by default and wired into `ModeBase` , so every mode exposes `jolt_press/jolt_release` , `jolt_lr_scale` , and `set_explore_intensity` .
2026-07-18 12:23:14 +02:00
- `nisps/pipeline/` — the control-rate input/output processing chains (P4): `input_chain.hpp` (`InputChain` — invert→deadzone→circular clamp→momentum-modulated zoom→centred power→EMA→momentum; caller-supplied dt, internal clock, fixed velocity ring, serialisable state) and `output_chain.hpp` (`OutputChain< NMax > ` — curve→EMA→slew→freeze(+mask), capacity-templated). Behaviour contract = the retired manifold TS pipelines, pinned by `manifold/tests/fixtures/` and parity stage 7.
2026-07-21 14:03:07 +02:00
- `nisps/dsp/` — `biquad.hpp` , `delay.hpp` , `reverb.hpp` , `filter.hpp` , `env.hpp` , `osc.hpp` , `pitch_shift.hpp` , `dc_blocker.hpp` , plus the sequencer primitives shared by the sequencer engines: `ratio_seq.hpp` and `seq_clock.hpp` (bar phasor + MIDI clock + bpm). Lean primitives extracted from maximilian; daisysp PitchShifter replaced with custom granular impl.
2026-04-29 18:57:27 +02:00
- `nisps/engines/` — eight audio engines, each satisfying `AudioEngine` : `paf_synth.hpp` , `channel_strip.hpp` , `xiasri.hpp` , `verb_fx.hpp` , `memlcelium.hpp` , `breakor.hpp` (sequencer, NoOp audio), `elysiamorf.hpp` (sequencer, NoOp audio), `analysis.hpp` (input-side spectral features). Plus `base.hpp` (`NoOpEngine`, engine_id "thru").
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- `nisps/modes/` — platform-agnostic modes binding `{ML config, engine, voice space lambdas, abstract I/O channels}` . Files: `paf_synth.hpp` , `channel_strip.hpp` , `xiasri.hpp` , `verb_fx.hpp` , `memlcelium.hpp` , `slp_workshop.hpp` (`SLPWorkshopMode` — the Synth Library Portland workshop build; reuses the MEMLCelium engine + MLP shape, foregrounds the Jolt + OU explore gestures), `breakor.hpp` , `elysiamorf.hpp` , `sound_analysis_midi.hpp` , `external_synth_midi.hpp` (`ExternalSynthMIDIMode< const MidiDevice &, NOut > ` — joystick→MLP→MIDI CC for an external synth; compile-time device from `nisps/midi` ; `consteval pick_cc_slots` curates which params fill the NOut slots; NoOpEngine, `kRouteOutputsToEngine=false` ). `base.hpp` provides a CRTP scaffold eliminating the duplication that previously plagued firmware modes; it also owns `driver_config()` — the audio-driver setup (mic vs line, gain staging, sample rate) the platform glue reads at mode start. Defaults to `engine().driver_config()` ; a mode overrides it with an optional `on_driver_config()` hook only when its engine is not what consumes the audio input (`sound_analysis_midi`, whose analyser rather than its NoOp engine owns the mic). `generated/` contains codegen output (do not edit by hand): per-mode `k<Mode>Schema` ParamSchema instances, the `<Mode>MLP` type aliases built from the schema's own dims, and `schema_types.hpp` which now owns the `ParamSchema` struct itself. Mode headers no longer hand-write either their schema aggregate or their net shape.
2026-04-29 18:57:27 +02:00
- `nisps/wasm/bindings.cpp` — flat C API exported to WASM (Emscripten target only).
feat(midi-devices): canonical external-synth CC templates + dual codegen
Add a durable, committed source of truth for external MIDI synth control:
- synth-midi-cc.json: verified CC maps + provenance/sources (8 devices researched)
- schemas/midi_device.schema.json + schemas/midi_devices/*.json: 6 CC-controllable
device templates (Moog Sub 37/Sub Phatty, Creamware Pro-12 ASB, Elektron Analog
Keys, ASM Hydrasynth, Roland JD-800), params keyed {id, cc, label, min, max,
default, group}.
- codegen/generate-midi-devices.ts (isolated from the mode golden test) emits both
nisps/midi/generated/midi_devices.hpp (no-heap constexpr, firmware+WASM) and
manifold/src/midi-devices/generated/ (typed catalogue for the browser).
- codegen/seed-midi-devices.ts: reproducible seed from the research artifact.
Lets a performer pick a device and address its parameters by name (not CC number)
on both the firmware and the Manifold browser engine.
2026-06-28 20:03:17 +02:00
- `nisps/midi/generated/midi_devices.hpp` — codegen output: no-heap `constexpr` external-MIDI-synth templates (`nisps::midi::generated`; `MidiParam` /`MidiDevice` + `kMidiDevices` registry). Source = `schemas/midi_devices/` ; do not edit by hand.
2026-04-29 18:57:27 +02:00
- `nisps/CMakeLists.txt` + `nisps/build/` — host-target builds + ctest.
build(firmware): migrate to PlatformIO and vendor memllib (plan §5)
One cut, no dual path. Closes ALIGNMENT defect 3 ("Arduino-CLI build
machinery is actively hostile") and vision bullet 4.
platformio.ini carries 16 [env:], one per variant, each passing
-DMEMLNAUT_MODE_TYPE; selftest passes -DNISPS_SELFTEST=1 instead. The env list
IS the registry now — the .ino comment-registry and the NISPS_ST_* token-paste
table are deleted rather than migrated. L12 noted that table was already
silently missing the currently-shipped SLPWorkshop variant, which is the whole
argument against having a second list.
Also deleted: the Python/sed machinery that rewrote the COMMITTED .ino on every
build, the sketch symlink forest, the global TFT_eSPI User_Setup.h mutation
(now -D flags — TFT_eSPI's own documented PlatformIO recipe), the UF2
boot-mount detection stack (upload_protocol=picotool talks to the bootloader
directly), and build-firmware-arch.sh entirely. Scripts 683 -> 435 lines.
memllib is vendored at lib/memllib/ from upstream e291192; no submodules
remain. VENDORED.md records provenance and the re-sync procedure.
S9: a firmware-build CI job compiles three representative envs against a cached
toolchain and reports per-variant flash/RAM. Firmware is in an automated gate
for the FIRST time. The old ci.yml comment justified excluding it as "low
verification value" — an assessment that did not survive contact, since the
SelfTest variant sat broken for an unknown period calling a DisplayDriver
method that did not exist at the pinned memllib commit, and nothing noticed
because nothing built it.
Verified: all 16 envs build from an empty cache, each within ~520 bytes of the
arduino-cli binary it replaces, flash and RAM. Measured as .text+.rodata /
.data+.bss+vector+uninitialized — NOT PlatformIO's console line, which
double-counts .data on this board. This does not prove the hardware boots; no
flash+smoke test was possible and that stays an operator chokepoint.
slpworkshop 248232/145028 pafsynth 256880/149716 selftest 216228/17960
(all 16 in the CI log format; none exceeds 2% of a 16 MB flash)
Two traps recorded so nobody rediscovers them: vendoring memllib's subdirs
without a src/ wrapper makes PlatformIO's library builder silently compile
NOTHING while still linking; and project build_flags land BEFORE the
framework's own -std=gnu++17 -Os, so build_unflags is required.
CORRECTION carried in this commit: the firmware sizes in c19d846's message and
the first version of the memllib recon doc were wrong — SLPWorkshop 145348,
PAFSynth 145300, SelfTest 141840. They came from building variants in sequence
through a SHARED incremental arduino-cli build directory, which reused stale
objects and under-reported by ~75 KB. Clean-cache rebuilds of the identical
commit give 216736/18492 for SelfTest. The real cost of the memllib upstream
bump is +216 bytes flash, not +316. Never measure firmware size through a
reused build dir.
HISTORY NOTE: this commit and the docs commit before it were rebuilt (force-push,
2026-07-21) so that each contains only what its message describes. The first
versions had the firmware deletions stranded in the docs commit by a shared-index
race between concurrent agents; content is byte-identical to the originals.
Gates: run-all-tests.sh ALL GREEN (nisps/ untouched by this change beyond
include paths); 16/16 pio envs build.
2026-07-21 20:17:58 +02:00
### `firmware/` — PlatformIO project + hardware glue
- `firmware/MEMLNaut-NISPS/platformio.ini` — **the variant registry** : one `[env:<alias>]` per firmware variant (16 of them), each passing `-DMEMLNAUT_MODE_TYPE=<alias>` ; `selftest` passes `-DNISPS_SELFTEST=1` instead. There is no second list to keep in sync. Shared `[env]` base pins the platform wrapper + arduino-pico framework, sets `-std=gnu++20 -O3` (via `build_unflags` , because the framework appends its own `-std=gnu++17 -Os` AFTER project flags), reaches `nisps/` with `-I${PROJECT_DIR}/../..` , and carries the TFT_eSPI panel config as `-D` flags. Build: `pio run -e <alias>` , or `scripts/build-firmware.sh [--all]` .
- `firmware/MEMLNaut-NISPS/src/main.cpp` — entry point (was `MEMLNaut-NISPS.ino` ). Forks on `NISPS_SELFTEST` : normal modes run the engine/ML path; the `SelfTest` variant delegates all four entry points to `glue/selftest.hpp` .
Stream 6: extract firmware glue under firmware/
Move the Arduino sketch into firmware/MEMLNaut-NISPS/ and bridge the
hardware (memllib) to the platform-agnostic nisps/ library through a
slim glue layer. Delete the legacy root-level *AudioApp.hpp,
modes/MEMLNautMode*.hpp, voicespaces/, IMLInterface.hpp, XiasriAnalysis,
and the src/memlp submodule.
Glue layout (firmware/MEMLNaut-NISPS/glue/):
audio_driver.hpp - bridge memllib block callback to Mode::process
via per-Mode templated trampoline (no virtual dispatch)
peripherals.hpp - joystick/pots/buttons -> Mode::set_input + ML primitives
midi_io.hpp - MIDI in -> mode.note_on/update_bpm/set_playing,
drains mode ControlEvent ring -> MIDI UART
mode_select.hpp - using-aliases mapping MEMLNautMode<Name> to
nisps::modes::*Mode (build script rewrites the
#define MEMLNAUT_MODE_TYPE line)
input_router.hpp / output_router.hpp - top-level wire/drain entry points
The sketch tree uses src/{memllib,daisysp,nisps} symlinks because
Arduino-CLI rejects ".." in include paths from sketch-tree headers.
mode_select.hpp #undefs Arduino's sq/min/max/abs/round macros before
including nisps headers (some nisps engines use those identifiers as
method names). The audio bridge struct is extern in the header and
defined in the .ino because inline + __not_in_flash section attribute
collide at link time.
Verification: arduino-cli compile succeeds for PAFSynth, ChannelStrip,
and BreakOr (rp2040:rp2040:solderparty_rp2350_stamp_xl:opt=Optimize3,
-std=gnu++20). Host C++ tests under nisps/build still pass (3 binaries,
110+ tests). Build script (scripts/build-firmware.sh) updated to point
at the new sketch path; mode-rewrite logic unchanged.
Closes meml-gkm.
2026-04-29 16:05:38 +02:00
- `firmware/MEMLNaut-NISPS/glue/` — hardware bindings:
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- `audio_driver.hpp` — bridges memllib `AudioDriver` callback → `Mode::process(stereosample_t)` , and brings the codec up on the **active mode's** `driver_config()` (`setup_audio_driver`) plus publishes its preferred sample rate before the system clock is derived from it (`apply_mode_sample_rate`). Mic vs line input is therefore a mode-level declaration, not a firmware constant.
- `codec_config.hpp` — pure, Arduino-free clamping of a `nisps::DriverConfig` to SGTL5000-representable values + sample-rate resolution (unsupported/"don't care" → 48 kHz, because `AudioDriver::GetSysClockSpeed()` `panic()` s otherwise). Host-tested by `tests/cpp/test_mode_driver_config.cpp` .
2026-06-28 23:44:57 +02:00
- `peripherals.hpp` — joystick / pots / buttons → `Mode::set_input` and ML primitives. Wires the shared `FeedbackController` ExploreAndPlace lifecycle (MomA1 = enter/exit explore, MomA2 = freeze/place, TogB2 = commit; MomB1/MomB2 = reroll/nudge while exploring **or** grab/drop *reposition* while idle) plus the adaptive-learning gestures: **TogB1** = Jolt (held weight morph), **RVX1** = exploration amount (OU output walk). Reposition relocates an existing positive example's output to a new input position (`feedback.hpp` `begin_reposition` /`commit_reposition`) — no scratchpad, no weight restore.
2026-04-29 18:57:27 +02:00
- `midi_io.hpp` — MIDI in → mode `note_on` /`update_bpm`/`set_playing`; drains `ControlEvent` ring → MIDI UART.
build(firmware): migrate to PlatformIO and vendor memllib (plan §5)
One cut, no dual path. Closes ALIGNMENT defect 3 ("Arduino-CLI build
machinery is actively hostile") and vision bullet 4.
platformio.ini carries 16 [env:], one per variant, each passing
-DMEMLNAUT_MODE_TYPE; selftest passes -DNISPS_SELFTEST=1 instead. The env list
IS the registry now — the .ino comment-registry and the NISPS_ST_* token-paste
table are deleted rather than migrated. L12 noted that table was already
silently missing the currently-shipped SLPWorkshop variant, which is the whole
argument against having a second list.
Also deleted: the Python/sed machinery that rewrote the COMMITTED .ino on every
build, the sketch symlink forest, the global TFT_eSPI User_Setup.h mutation
(now -D flags — TFT_eSPI's own documented PlatformIO recipe), the UF2
boot-mount detection stack (upload_protocol=picotool talks to the bootloader
directly), and build-firmware-arch.sh entirely. Scripts 683 -> 435 lines.
memllib is vendored at lib/memllib/ from upstream e291192; no submodules
remain. VENDORED.md records provenance and the re-sync procedure.
S9: a firmware-build CI job compiles three representative envs against a cached
toolchain and reports per-variant flash/RAM. Firmware is in an automated gate
for the FIRST time. The old ci.yml comment justified excluding it as "low
verification value" — an assessment that did not survive contact, since the
SelfTest variant sat broken for an unknown period calling a DisplayDriver
method that did not exist at the pinned memllib commit, and nothing noticed
because nothing built it.
Verified: all 16 envs build from an empty cache, each within ~520 bytes of the
arduino-cli binary it replaces, flash and RAM. Measured as .text+.rodata /
.data+.bss+vector+uninitialized — NOT PlatformIO's console line, which
double-counts .data on this board. This does not prove the hardware boots; no
flash+smoke test was possible and that stays an operator chokepoint.
slpworkshop 248232/145028 pafsynth 256880/149716 selftest 216228/17960
(all 16 in the CI log format; none exceeds 2% of a 16 MB flash)
Two traps recorded so nobody rediscovers them: vendoring memllib's subdirs
without a src/ wrapper makes PlatformIO's library builder silently compile
NOTHING while still linking; and project build_flags land BEFORE the
framework's own -std=gnu++17 -Os, so build_unflags is required.
CORRECTION carried in this commit: the firmware sizes in c19d846's message and
the first version of the memllib recon doc were wrong — SLPWorkshop 145348,
PAFSynth 145300, SelfTest 141840. They came from building variants in sequence
through a SHARED incremental arduino-cli build directory, which reused stale
objects and under-reported by ~75 KB. Clean-cache rebuilds of the identical
commit give 216736/18492 for SelfTest. The real cost of the memllib upstream
bump is +216 bytes flash, not +316. Never measure firmware size through a
reused build dir.
HISTORY NOTE: this commit and the docs commit before it were rebuilt (force-push,
2026-07-21) so that each contains only what its message describes. The first
versions had the firmware deletions stranded in the docs commit by a shared-index
race between concurrent agents; content is byte-identical to the originals.
Gates: run-all-tests.sh ALL GREEN (nisps/ untouched by this change beyond
include paths); 16/16 pio envs build.
2026-07-21 20:17:58 +02:00
- `mode_select.hpp` — type aliases mapping firmware mode identifiers to `nisps::modes::*Mode` C++ types, selected by the `-D` from platformio.ini. Includes the six `MEMLNautModeExtSynth*` external-synth variants (one per device template in `nisps/midi` , e.g. `MEMLNautModeExtSynthSub37` ). The `NISPS_ST_*` /`NISPS_ST_CAT` token-paste table and the `SelfTestRig` tag type are GONE — selftest is now just an env with its own `-D` .
2026-06-28 04:14:30 +02:00
- `selftest.hpp` — standalone guided hardware self-test rig (`SelfTest` variant; no engine/ML). Step-driven state machine on a `SelfTestView` : TFT prompts the operator through every control, auto-advances on detection, encoder-press skips. Ends with optional L/R/BOTH sine-sweep headphone check (core 1 block callback) + MIDI loopback-cable test. Lives firmware-side (touches TFT + raw pins) so it stays out of platform-agnostic `nisps/` .
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
- `output_router.hpp` — top-level `drain_outputs()` entry point. (Inputs are wired directly by `peripherals.hpp` 's `bind_peripherals()` .)
2026-06-28 21:02:23 +02:00
- `settings_view.hpp` — `wire_settings(mode)` : adds on-device settings views to the MEMLNaut display carousel (TFT + rotary encoder). Joystick Dual/Single toggle for the 4-input ("two 2-D joystick") modes — "Single" pins ML input channels 2,3 to neutral via `ModeBase::set_input_pinned` (no net rebuild). Registered in the `.ino` after `addSystemInfoView()` .
build(firmware): migrate to PlatformIO and vendor memllib (plan §5)
One cut, no dual path. Closes ALIGNMENT defect 3 ("Arduino-CLI build
machinery is actively hostile") and vision bullet 4.
platformio.ini carries 16 [env:], one per variant, each passing
-DMEMLNAUT_MODE_TYPE; selftest passes -DNISPS_SELFTEST=1 instead. The env list
IS the registry now — the .ino comment-registry and the NISPS_ST_* token-paste
table are deleted rather than migrated. L12 noted that table was already
silently missing the currently-shipped SLPWorkshop variant, which is the whole
argument against having a second list.
Also deleted: the Python/sed machinery that rewrote the COMMITTED .ino on every
build, the sketch symlink forest, the global TFT_eSPI User_Setup.h mutation
(now -D flags — TFT_eSPI's own documented PlatformIO recipe), the UF2
boot-mount detection stack (upload_protocol=picotool talks to the bootloader
directly), and build-firmware-arch.sh entirely. Scripts 683 -> 435 lines.
memllib is vendored at lib/memllib/ from upstream e291192; no submodules
remain. VENDORED.md records provenance and the re-sync procedure.
S9: a firmware-build CI job compiles three representative envs against a cached
toolchain and reports per-variant flash/RAM. Firmware is in an automated gate
for the FIRST time. The old ci.yml comment justified excluding it as "low
verification value" — an assessment that did not survive contact, since the
SelfTest variant sat broken for an unknown period calling a DisplayDriver
method that did not exist at the pinned memllib commit, and nothing noticed
because nothing built it.
Verified: all 16 envs build from an empty cache, each within ~520 bytes of the
arduino-cli binary it replaces, flash and RAM. Measured as .text+.rodata /
.data+.bss+vector+uninitialized — NOT PlatformIO's console line, which
double-counts .data on this board. This does not prove the hardware boots; no
flash+smoke test was possible and that stays an operator chokepoint.
slpworkshop 248232/145028 pafsynth 256880/149716 selftest 216228/17960
(all 16 in the CI log format; none exceeds 2% of a 16 MB flash)
Two traps recorded so nobody rediscovers them: vendoring memllib's subdirs
without a src/ wrapper makes PlatformIO's library builder silently compile
NOTHING while still linking; and project build_flags land BEFORE the
framework's own -std=gnu++17 -Os, so build_unflags is required.
CORRECTION carried in this commit: the firmware sizes in c19d846's message and
the first version of the memllib recon doc were wrong — SLPWorkshop 145348,
PAFSynth 145300, SelfTest 141840. They came from building variants in sequence
through a SHARED incremental arduino-cli build directory, which reused stale
objects and under-reported by ~75 KB. Clean-cache rebuilds of the identical
commit give 216736/18492 for SelfTest. The real cost of the memllib upstream
bump is +216 bytes flash, not +316. Never measure firmware size through a
reused build dir.
HISTORY NOTE: this commit and the docs commit before it were rebuilt (force-push,
2026-07-21) so that each contains only what its message describes. The first
versions had the firmware deletions stranded in the docs commit by a shared-index
race between concurrent agents; content is byte-identical to the originals.
Gates: run-all-tests.sh ALL GREEN (nisps/ untouched by this change beyond
include paths); 16/16 pio envs build.
2026-07-21 20:17:58 +02:00
- `firmware/MEMLNaut-NISPS/lib/memllib/` — **vendored** memllib (was the `src/memllib` submodule): hardware abstraction (audio driver, TFT display, MIDI, peripherals), ~1.9 MB / 100 files, `examples/` dropped. `VENDORED.md` records the upstream commit and the re-sync procedure; `LICENSE` is MPL-2.0, copied verbatim. **Sources must sit under `lib/memllib/src/`** — PlatformIO's library builder falls back to a flat root-only scan without it and silently compiles nothing while still linking (see VENDORED.md).
2026-04-29 18:57:27 +02:00
- `firmware/README.md` — structure + build instructions.
2026-07-13 22:15:46 +02:00
- `firmware/useq-celium/` — standalone RP2040 firmware (PlatformIO, Arduino-Pico core) that turns a uSEQ module + CV expander into a USB→CV/gate converter driven by the manifold `cvgate` backend. `shared/protocol.h` is the v2 wire-protocol single source of truth (mirrored by `manifold/src/backends/useq-protocol.ts` ); `main/` (USB serial → CV1– – `expander/` (I2C slave → CV4– `docs/specs/useq-cv-protocol.md` . Restored from the April-2026 "uSEQ-Celium" mode.
2026-04-29 18:57:27 +02:00
2026-07-13 23:27:56 +02:00
### `manifold/` — Vite + React + TS convertible-mode app (the sole browser app)
2026-06-28 04:14:30 +02:00
The Manifold "convertible" Console on the real engine, deployed at `meml.lnfinitemonkeys.org/next` (staging,
2026-07-13 22:15:46 +02:00
alongside the live vanilla a-immersive at `/` ). Built 2026-06-27/28; see `docs/specs/plans/BUILD-PLAN.md` (resume
anchor + locked decisions) and the `docs/specs/*-spec.md` set.
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
- `manifold/src/engine/` — the parity-tested TS engine (same `nisps.wasm` ), made
2026-06-28 04:14:30 +02:00
framework-neutral: `wasm-iml.ts` (rewired off Solid stores onto an injected `EngineSink` ), `engine-host.ts` +
2026-07-18 12:23:14 +02:00
`worklet/nisps-processor.ts` (audio), thin WASM wrappers over the core pipelines + curve catalog (the TS
`input-pipeline` /`output-pipeline`/`curves` implementations died at P4), `wasm-worker.ts` ,
2026-06-28 04:14:30 +02:00
`spine.ts` (the reactive spine BELOW React — `setInput` derives processed→ml→routed eagerly off-render),
`engine-api.ts` (`EngineApi` façade incl. `feedback.*` wrappers over the `nisps_ml_feedback_*` C ABI),
`EngineProvider.tsx` /`useEngine.ts` (React binding via `useSyncExternalStore` version counter). nisps.js is
loaded via fetch+indirect-eval (Emscripten MODULARIZE glue has no ES exports), base-aware via `document.baseURI`
for the `/next` sub-path.
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
- `manifold/src/primitives/` — the 7 design primitives as typed React (Badge, Button, PillToggle, Slider, Switch, VirtualJoystick, XYPad). Five unused ones were deleted in the 2026-07 sweep (L22).
2026-06-28 04:14:30 +02:00
- `manifold/src/console/` — the convertible Console: `ConsoleApp` , `CompositeStage` (single-divider convertible
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
with snap/magnetism/minimap-demotion), `OutputStage` /`SandwichStage`/`ParticleStage`/`Manifold` (canvas,
rect↔circular + feedback markers), `Dock` (top Mode selector + 5 vertically-centred drawers), `Drawers`
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
(Learning/Inputs/Outputs/Settings/Help), `TrainingHealth` (real per-iteration loss curve from
`nisps_ml_loss_history` + per-layer weight health from `nisps_ml_get_layer_stats` ; rendered only at
the Learning drawer's `expanded` depth — that IS the advanced-surface flag), `VerdictCluster`
(mode-aware), `OutputEditor` /`CurvePad`, `icons.tsx`
feat(manifold): P5.2/P5.3 — derive MF_MODES from schema truth + per-mode engine dims
Schema-backed modes in console/model.ts are now DERIVED from the codegen
schemas in src/modes/generated/ (source of truth): real param names/groups/
count, plus each mode's ml net shape (MFMode.ml) and schema engine_id. A thin
manifold OVERLAY supplies only label/glyph/ModeClass/input/ordering. New
browser-viable modes xiasri + slp_workshop get derived entries; schema-less
visualizer + c15 stay hand-written on DEFAULT_MODE_ML. Schema min/max/default/
label/curve surface as engine-unit metadata (schemaMin/... on MFParam) without
touching the 0..1 routing semantics.
Switching instrument mode reshapes the runtime-shaped WASM net to the mode's
schema ml config (ConsoleApp effect keyed on [engine, modeId]; no confirm
modal). Boot lands paf_synth dims (4->[10,10,14]->33) once WASM is ready. The
P2.3 axis-count reshape offer still reads the engine's live inputSize and does
not spuriously prompt on a mode switch.
Adds schema-modes.spec.ts (P5 gate): drives switches via a new window.__mf
debug seam and asserts describe() dims, getWeights count, output length/bounds,
and UI param count FROM the imported schemas; spot-checks trainAsync after a
switch. Updates reshape/probe-api/geo-dislike specs to assert from the boot
mode schema instead of the retired fixed 32/126 shape.
All gates green: typecheck, unit (9), build, e2e (33).
2026-07-18 12:45:06 +02:00
(monochrome currentColor SVG), `model.ts` (`MF_MODES` catalogue — schema-backed modes DERIVED from
`manifold/src/modes/generated/` ; carries per-mode `ml` net shape + `engineId` ), `output-mode.ts` .
- `manifold/src/modes/generated/` — codegen output (`*_schema.ts`, do NOT hand-edit): `ModeSchema`
consts (mode_id, engine_id, ml dims, params, voice_spaces, ui) — the SOURCE OF TRUTH for `MF_MODES` .
Switching mode reshapes the WASM net to the mode's `ml` dims (ConsoleApp P5.3; boot mode paf_synth →
4→[10,10,14]→33).
2026-06-28 04:14:30 +02:00
- `manifold/src/dock/` — `OutputControlRow` (off/fixed/live + mute + solo/arm + min/max/curve), `output-state.ts` ,
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
`OutputsBackendConfig.tsx` (per-backend specialised Outputs panel — the sole per-backend editor).
2026-06-28 04:14:30 +02:00
- `manifold/src/backends/` — `OutputBackend` adapter + `BackendManager` (spine consumer); `midi-backend.ts`
2026-06-28 21:28:03 +02:00
(WebMIDI), `osc-backend.ts` +`osc-client.ts` (OSC-over-WS), `vcv-backend.ts` (VCV-over-WS), `cv-backend.ts`
(`UseqCvBackend` — uSEQ CV/gate over USB Web Serial, backend id `cvgate` ) + `useq-protocol.ts` (v2 wire
protocol, mirrors `firmware/useq-celium/shared/protocol.h` ; `useq-protocol.test.ts` runs via `bun test` ),
`particle-backend.ts` , `passthrough-backend.ts` , `presets.ts` (named presets), `manager.ts` .
feat(midi-devices): canonical external-synth CC templates + dual codegen
Add a durable, committed source of truth for external MIDI synth control:
- synth-midi-cc.json: verified CC maps + provenance/sources (8 devices researched)
- schemas/midi_device.schema.json + schemas/midi_devices/*.json: 6 CC-controllable
device templates (Moog Sub 37/Sub Phatty, Creamware Pro-12 ASB, Elektron Analog
Keys, ASM Hydrasynth, Roland JD-800), params keyed {id, cc, label, min, max,
default, group}.
- codegen/generate-midi-devices.ts (isolated from the mode golden test) emits both
nisps/midi/generated/midi_devices.hpp (no-heap constexpr, firmware+WASM) and
manifold/src/midi-devices/generated/ (typed catalogue for the browser).
- codegen/seed-midi-devices.ts: reproducible seed from the research artifact.
Lets a performer pick a device and address its parameters by name (not CC number)
on both the firmware and the Manifold browser engine.
2026-06-28 20:03:17 +02:00
- `manifold/src/midi-devices/` — external-synth device templates. `generated/` is codegen output from
`schemas/midi_devices/` (`MIDI_DEVICES` catalogue + `MIDI_DEVICES_BY_ID` , params by name+CC). The MIDI Outputs
config (`dock/OutputsBackendConfig.tsx`) reads it for the device picker + param-select that fills the CC table.
feat(manifold): MIDI + game controller inputs; widen ML net to N-D
Wire the modular input layer into the Console and reshape the browser
engine so input axes are genuine independent dimensions.
Inputs (manifold/src/inputs/):
- gamepad-source: emit press+release edges with standard-mapping labels
(enables hold-and-move); single/double-stick already present.
- midi-input-source: single-device selection + batch "MIDI Learn"
(every CC swept while armed becomes an axis); notes stay discrete.
- input-layer: compose() forwards each axis 1:1 (no mean-blend);
add onReducedInput so the manifold tracks gamepad/MIDI position.
- types: InputAction.phase, InputMode.
Console (manifold/src/console/):
- ConsoleApp: bind gamepad buttons to verdicts (RB up / LB down /
X randomise / Y nudge / B undo / A-hold reposition); mirror composed
position onto the manifold.
- Drawers: rebuilt Inputs drawer (source picker, gamepad legend, MIDI
device picker + batch-learn flow, learned-control meters).
Engine (nisps/wasm, manifold/src/engine):
- DefaultMLP widened MLP<2,..> -> MLP<32,..> (32 = MAX_AXES); each
active axis gets a dedicated slot, unused slots held at 0 (inert).
Rebuilt nisps.wasm (playground + manifold).
- spine/engine-api: setInputs writes the full N-D vector (was dropping
arr[2+]); primary pair keeps the 2-D pipeline; process() re-ticks the
whole vector via spine.reprocess().
Tests:
- parity_check/parity_wasm: ParityMLP -> 32 inputs, widen example bufs.
- CMakeLists: build parity binary with -ffp-contract=off so native
matches FMA-free WASM (training amplified the gap past 1e-5).
Inputs dock is still an exclusive picker; mixing toggles, reshape modal,
and the >2-D slider view (inputs-spec.md) are groundwork-laid but not
yet wired. See docs/redesign/midi-gamepad-inputs-worklog.md.
2026-06-28 21:05:30 +02:00
- `manifold/src/inputs/` — modular INPUT layer feeding the ML head. The Inputs dock picks ONE exclusive mode
(`InputMode` = `internal` | `gamepad` | `midi` ; Internal/XY-pad is default). `input-layer.ts` owns a single rAF
docs: the specs disposition pass (plan §8)
Roughly 20k lines were deleted from this repo in the last week and much of the
corpus still described the pre-deletion world in the present tense. Executes
the §8 table: archive the retired, reclassify the executed, prune the stale.
aimmersive-clone-spec -> _archive/ with a deprecated-by note
feedback-modes-port-spec -> plans/, kind: plan, status: executed
manifold-parity-features -> plans/, kind: plan, status: active
playground-2.0-rewrite -> status: superseded
engine-architecture 434 -> ~120 lines; seam + spine kept, rewritten
present-tense against the shipped engine/
MAIN.md six contradicted claims fixed; registry resynced
vcv-module.md pruned to the current 8->16 contract and made the
single .nisps format spec
vcv/NISPS-FORMAT.md DELETED — documented a v1 format that no longer loads
vcv/README.md, BUILDING.md rewritten to the real contract, menu, OSC table
inputs/backends/dock trio grounding sections marked historical, dead cites fixed
Two rows of the §8 table were themselves wrong, corrected here: the deleted
full-state sync lives in backends-spec.md §6.3, not vcv-module.md (which has no
§6.3), and codegen/README.md was already a MAP pointer with no port-solidjs
trigger left to remove.
Beyond the table — found by sweeping every backticked path in the changed docs
against `git ls-files`, which is how these should have been caught before:
manifold/ONBOARDING.md documented a UI that Phase 1 deleted, as if current:
SplitStage, ReadoutStrip, InputMini, BackendAdvanced, AltitudeNav, and a
shot.spec.ts that does not exist. The whole stage table was keyed on a `focus`
axis that no longer exists — selection is now sandwich > particles >
composite. This matters more than the rest: CLAUDE.md tells every agent to
read ONBOARDING.md first for Manifold work, so it was actively teaching a
fiction. Rewritten against ConsoleApp.tsx.
MAP.md claimed the input layer reduces axes to the engine arity with an
"even/odd blend". input-layer.ts says the opposite in its own header: one
dedicated slot per axis, 1:1, into a 32-input over-provisioned head, and
mean-blending was removed deliberately because it diluted every source.
AGENT-REFERENCE.md still promised TS emission "returns at P5" (landed),
per-mode dims "become schema-real at P5" (landed at P5.3), and pointed at
nisps::FixedBuffer (deleted).
Doc-right/code-suspect, filed rather than fixed: VCV computes derivedMean/Std/
Delta and cachedNovelty behind a live context-menu toggle that nothing reads;
vcv/plugin.json points at the MusicallyEmbodiedML org rather than this repo's
origin; and the module defaults to UDP 7001+id%64 while bridge.ts defaults to
9000, so out of the box they do not meet.
Firmware-build docs are deliberately untouched — the PlatformIO migration
lands next and rewrites all of them.
2026-07-21 20:17:58 +02:00
loop composing the active source's axes → **one dedicated engine input slot per axis, 1:1, no blending** → one
`setInputs` , plus an `onReducedInput` callback the manifold tracks. The WASM net is over-provisioned to a
32-input head (`MAX_AXES`, `nisps/wasm/bindings.cpp` ); unused slots are zero-padded and a zero input is inert,
so idle sources cannot perturb the net. Mean-blending was removed deliberately — it diluted every source and
biased the net toward idle sources' resting values. Changing the ACTIVE axis count offers a reshape
(`ConsoleApp` → `ReshapeModal` ): new net at the new arity, warm-started from overlapping weights, examples and
feedback state reset; declining keeps the over-provisioned head. Sources: `xy-pad-source` (push-driven),
feat(manifold): MIDI + game controller inputs; widen ML net to N-D
Wire the modular input layer into the Console and reshape the browser
engine so input axes are genuine independent dimensions.
Inputs (manifold/src/inputs/):
- gamepad-source: emit press+release edges with standard-mapping labels
(enables hold-and-move); single/double-stick already present.
- midi-input-source: single-device selection + batch "MIDI Learn"
(every CC swept while armed becomes an axis); notes stay discrete.
- input-layer: compose() forwards each axis 1:1 (no mean-blend);
add onReducedInput so the manifold tracks gamepad/MIDI position.
- types: InputAction.phase, InputMode.
Console (manifold/src/console/):
- ConsoleApp: bind gamepad buttons to verdicts (RB up / LB down /
X randomise / Y nudge / B undo / A-hold reposition); mirror composed
position onto the manifold.
- Drawers: rebuilt Inputs drawer (source picker, gamepad legend, MIDI
device picker + batch-learn flow, learned-control meters).
Engine (nisps/wasm, manifold/src/engine):
- DefaultMLP widened MLP<2,..> -> MLP<32,..> (32 = MAX_AXES); each
active axis gets a dedicated slot, unused slots held at 0 (inert).
Rebuilt nisps.wasm (playground + manifold).
- spine/engine-api: setInputs writes the full N-D vector (was dropping
arr[2+]); primary pair keeps the 2-D pipeline; process() re-ticks the
whole vector via spine.reprocess().
Tests:
- parity_check/parity_wasm: ParityMLP -> 32 inputs, widen example bufs.
- CMakeLists: build parity binary with -ffp-contract=off so native
matches FMA-free WASM (training amplified the gap past 1e-5).
Inputs dock is still an exclusive picker; mixing toggles, reshape modal,
and the >2-D slider view (inputs-spec.md) are groundwork-laid but not
yet wired. See docs/redesign/midi-gamepad-inputs-worklog.md.
2026-06-28 21:05:30 +02:00
`gamepad-source` (sticks→axes single/double; buttons emit press+release actions, bound in `ConsoleApp` to
verdicts — LB/RB=down/up, X/Y/B=randomise/nudge/undo, A-hold=reposition), `midi-input-source` (device picker +
BATCH "MIDI Learn": every CC swept while armed becomes an axis, shown as read-only meters). `useInputLayer.ts`
is the React binding; `base-source.ts` shared status/action plumbing; `types.ts` the adapter contract.
2026-07-21 14:03:07 +02:00
`backends/base-backend.ts` is its output-side counterpart (status + throttle + lastSent) used by the midi/osc/vcv transports.
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
- `manifold/src/feedback/` — `controller.ts` (Explore-and-place scratchpad + geometric-dislike + solo; a thin
driver over the shared C++ core).
2026-06-28 04:14:30 +02:00
- `manifold/src/settings/` — `settings-store.ts` (monochrome icons, input-map shape, corner radius).
- `manifold/src/serial/` — `memlnaut-serial.ts` Web Serial scaffold + `EditorPanel.tsx` (MEMLNaut Editor mode).
docs(audit): simplification audit 2026-07 — recon findings, phased plan, ALIGNMENT rewrite
66-agent adversarially-verified audit vs the five-bullet one-core vision.
113 findings: CI red since 2026-07-13 (memllib pin on no remote), ungated
deploys, unshared mode layer, no curated/advanced split, dead-mass inventory.
Recon: docs/specs/recon/simplification-audit-2026-07.md (immutable).
Plan: docs/specs/plans/simplification-plan.md (proposal; phases gated on
operator adoption, §7 decisions). ALIGNMENT rewritten; MAP flatly-false
lines fixed (phantom MEMLCelium-upstream entry, exploration.ts, daisysp
non-submodule, pre-P5 sentence, perf-attr claims); MAIN registry updated.
2026-07-21 01:24:35 +02:00
- `manifold/src/engine/exploration.ts` — Jolt press + OU explore gestures (Learning drawer): a thin
timer-driver over the shared C++ core via the `nisps_ml_jolt_*` /`nisps_ml_ou_*` bindings (the interim
TS math and `jolt.ts` /`ou-explore.ts` were deleted when P3 landed).
2026-07-13 23:30:28 +02:00
- `manifold/src/debug/probe.ts` — `window.__nisps` (`?debug=1`). `manifold/tests/e2e/` — `smoke` ,
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
`probe-api` (engine-contract port), `spine` (spine invariant + probe-survives-mode-switch),
`geo-dislike` , `reshape` , `schema-modes` , `training-health` (the loss/layer-stats panel + its
expanded-depth gating). E2E on the VPS runs via non-snap node (see BUILD-PLAN).
`manifold/tests/fixtures/` — golden parity
2026-07-13 23:30:28 +02:00
fixtures (gesture trace, curves, input/output pipelines) captured 2026-07-13 pre-P4, guarded by
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
`tests/pipeline-golden.test.ts` (in `bun run test` ). `tests/loss-history.test.ts` drives the
`nisps_ml_loss_history` C ABI straight at the committed WASM — the training path parity-check
never touches. `manifold/osc-bridge/` — Deno WS↔UDP-OSC bridge.
2026-06-28 04:14:30 +02:00
### `vcv/` — VCV Rack 2 plugin (MEMLNaut module, WIP)
2026-07-18 13:01:28 +02:00
Native C++ Rack module: ML CV-mapper with RL feedback + a browser bridge. **8 inputs ×
LED rings**, palette from the frontend tokens, WS↔OSC browser bridge (see `docs/specs/vcv-module.md` ).
`src/MEMLNaut.cpp` (module, 8→[16,24,16]→16), `src/iml.hpp` (**thin adapter over `nisps::ml::MLPCore<DynamicStorage>`
+ core `nisps::Rng` ** — P6 reunification 2026-07-18, closes vcv-module.md delta #5 ; behaviour is now core-exact,
pinned by `tests/cpp/test_vcv_iml_parity.cpp` ), `src/osc_server.hpp` (bridge, transport-only), `src/plugin.{hpp,cpp}` ,
`res/*.svg` (panels), `Makefile` (needs `RACK_DIR` ). Builds against the current `../nisps/` core via relative
includes; no `nisps-core` .
2026-06-28 04:14:30 +02:00
2026-04-29 18:57:27 +02:00
### `schemas/` — JSON parameter contracts (firmware/browser source of truth)
- `schemas/schema.json` — Draft 2020-12 meta-schema validating mode files.
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- `schemas/modes/<mode>.json` (× `engine_id: memlcelium` .) `params[].curve` is the mode-wide DEFAULT response curve; a `voice_spaces` entry may be an object `{name, curve_overrides}` declaring only the slots where THAT voice space deviates (index i == `VoiceSpace` ordinal i). Descriptive throughout: the curve is applied exactly once, inside the engine's voice space.
- `schemas/modes/params_notes.md` — provenance notes and judgement calls per mode, plus the exact `square` /`sqrt`/`linear` predicate the drift check enforces.
feat(midi-devices): canonical external-synth CC templates + dual codegen
Add a durable, committed source of truth for external MIDI synth control:
- synth-midi-cc.json: verified CC maps + provenance/sources (8 devices researched)
- schemas/midi_device.schema.json + schemas/midi_devices/*.json: 6 CC-controllable
device templates (Moog Sub 37/Sub Phatty, Creamware Pro-12 ASB, Elektron Analog
Keys, ASM Hydrasynth, Roland JD-800), params keyed {id, cc, label, min, max,
default, group}.
- codegen/generate-midi-devices.ts (isolated from the mode golden test) emits both
nisps/midi/generated/midi_devices.hpp (no-heap constexpr, firmware+WASM) and
manifold/src/midi-devices/generated/ (typed catalogue for the browser).
- codegen/seed-midi-devices.ts: reproducible seed from the research artifact.
Lets a performer pick a device and address its parameters by name (not CC number)
on both the firmware and the Manifold browser engine.
2026-06-28 20:03:17 +02:00
- `schemas/midi_device.schema.json` — Draft 2020-12 meta-schema for external-MIDI-synth templates.
2026-07-21 14:03:07 +02:00
- `schemas/midi_devices/<device>.json` (× `{id, cc, label, min, max, default, group}` . Canonical source for both firmware + browser device pickers. Verified-CC provenance + sources live in `schemas/midi_devices/sources/synth-midi-cc.json` (a `sources/` subdir, because the generator ajv-validates every `*.json` directly under `midi_devices/` as a device template).
2026-04-29 18:57:27 +02:00
### `codegen/` — schema → C++/TS code
2026-07-18 12:46:24 +02:00
- `codegen/generate.ts` — Bun script: validates schemas via ajv (incl. the P5 firmware-fit check: exactly 3 hidden layers, dims ≤4096), emits per-mode C++ `nisps/modes/generated/<mode>_schema.hpp` (`constexpr`, `nisps::modes::generated` ) AND TS `manifold/src/modes/generated/<mode>_schema.ts` (+ `types.ts` , `index.ts` ). Idempotent; golden-tested in `run-all-tests.sh` stage 5. The TS output is the SOURCE OF TRUTH consumed by `MF_MODES` (`manifold/src/console/model.ts`).
feat(midi-devices): canonical external-synth CC templates + dual codegen
Add a durable, committed source of truth for external MIDI synth control:
- synth-midi-cc.json: verified CC maps + provenance/sources (8 devices researched)
- schemas/midi_device.schema.json + schemas/midi_devices/*.json: 6 CC-controllable
device templates (Moog Sub 37/Sub Phatty, Creamware Pro-12 ASB, Elektron Analog
Keys, ASM Hydrasynth, Roland JD-800), params keyed {id, cc, label, min, max,
default, group}.
- codegen/generate-midi-devices.ts (isolated from the mode golden test) emits both
nisps/midi/generated/midi_devices.hpp (no-heap constexpr, firmware+WASM) and
manifold/src/midi-devices/generated/ (typed catalogue for the browser).
- codegen/seed-midi-devices.ts: reproducible seed from the research artifact.
Lets a performer pick a device and address its parameters by name (not CC number)
on both the firmware and the Manifold browser engine.
2026-06-28 20:03:17 +02:00
- `codegen/generate-midi-devices.ts` — separate Bun script (isolated from the mode golden test): validates `schemas/midi_devices/` via ajv, emits `nisps/midi/generated/midi_devices.hpp` (no-heap `constexpr` ) and `manifold/src/midi-devices/generated/{types,devices,index}.ts` . Idempotent.
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- `codegen/curve-audit.ts` — reads `nisps/engines/*.hpp` and derives, per voice space, which response curve the engine applies to each NN-output slot. Handles the four idioms a regex misses (`const float v = p[n]; v*v`, memlcelium's implicit-counter `sq()` lambda, loop-generated indices, `smooth_params_[n]` ) and RAISES on anything it cannot reduce rather than defaulting to `linear` .
- `codegen/tests/curve_drift_test.ts` — the gate: schema-declared curves (JSON **and** generated TS) must equal what `curve-audit.ts` extracts, and schema `voice_spaces` order must equal the engine's `kVoiceSpaceNames` . Source-level by necessity — the curve is not recoverable from engine output. Run by `run-all-tests.sh` stage 5, CI, and `bun run test` in `codegen/` .
2026-07-21 14:03:07 +02:00
- `codegen/lib.ts` — helpers shared by both generators. `codegen/tests/golden/` — golden snapshot for paf_synth (C++ + TS).
2026-04-29 18:57:27 +02:00
### `tests/cpp/` — host C++ tests
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
- Per-component tests: `test_dsp_*.cpp` , `test_engine_*.cpp` , `test_mlp_*.cpp` , `test_mode_*.cpp` , `test_ring_buffer.cpp` , `test_rng.cpp` , `test_math.cpp` . Helpers in `test_helpers.hpp` .
2026-04-29 18:57:27 +02:00
- Verification: `ml_golden_vectors.cpp` , `engine_impulse.cpp` (+ `engine_impulse_baseline.bin` ), `parity_check.cpp` + `parity_wasm.mjs` + `parity_diff.mjs` — native-vs-WASM bit-equivalence within 1e-5.
test(ml): behavioural benchmark + 20 invariants for the control mapping
NISPS is a controller, not a synth: the object of study is the mapping
f: control-space -> parameter-space and how a musician's gestures deform
it. Loss measures fit to points the user dictated, which is the one thing
they never experience. So this measures geometry and gesture-response.
tests/cpp/ml_bench.cpp 61 scenarios, REPORTS never asserts (same
discipline as engine_bench.cpp). Shape-
agnostic via MLPCore<DynamicStorage>
(--shape, default 2,16,16,16,8), seeded
RNG throughout, branch points replayed
from scratch rather than snapshotted.
tests/cpp/test_ml_behaviour.cpp 20 asserting invariants, wired into
nisps_core_tests.
scripts/bench-ml.sh native + WASM from one source; --compare,
--sweep-shape, --smoke, --scenario, --seed.
Documents two contracts that fail SILENTLY (both now pinned by tests):
a thumbs-up must call BOTH mlp.add_example() and fb.store_positive(),
since dislike_geometric k-NNs the replay buffer and not the MLP dataset;
and placed_output() is valid only while state == Placing, after which an
empty span whose l2() is 0 scores a broken lifecycle as a perfect place.
ALIGNMENT defect 6 re-ranked (SGD-vs-RMSProp is not a research axis - it
silently invalidated every ported hyperparameter) and split into 6b (the
geometric dislike was ported from a superseded upstream design) and 6c
(InterfaceRL, the reference impl, is not in the tree).
Gates: build-cpp-tests (138 tests, ctest 4/4), parity-check PASS,
lint-cpp clean, bench-ml.sh --smoke runs end to end.
2026-07-25 11:02:24 +02:00
- Behaviour: `test_ml_behaviour.cpp` — 20 INVARIANTS of the interaction model (not tuning). Asserts what must hold however the knobs are set: a like is reachable and a re-like overwrites; dislike never yields non-finite weights/outputs in either `AvoidStyle` , and is safe at cold start; repeat dislikes inside `kReplayDedupRadius` deepen ONE negative while distant ones store separately; explore→reroll→undo→exit restores weights *bit-exactly* and over-undoing is safe; `RandomiseOutputs` leaves weights untouched (the outputs-vs-weights randomisation distinction); same seed + same gesture sequence ⇒ bit-identical weights (what makes any behavioural benchmark comparable); example-ring overflow keeps the newest; contradictory examples stay finite; the ExploreAndPlace accessor contract (`placed_output()` while Placing, `committed_output()` after commit — for both place and reposition); a FULLY-masked focus gate freezes every weight; switching mode mid-exploration never strands the net in a randomised scratchpad; and all of it holds at 1× × × × ×
- Measurement (asserts nothing): `ml_bench.cpp` — the BEHAVIOURAL benchmark. NISPS is a controller, so this measures the shape of the control→parameter mapping and how interaction journeys deform it, never loss alone. Shape-agnostic (`--shape N_IN,H1,H2,H3,N_OUT`, default `2,16,16,16,8` ) via `MLPCore<DynamicStorage>` , so "does a wider/deeper net change the UX?" becomes a number; sample points come from a deterministic Kronecker low-discrepancy sequence rather than a raster, which is what makes it work at any input arity. Field metrics: local gain p50/p95, cliff index, dead fraction, range utilisation, rail occupancy, effective dimensionality (participation ratio — trace²/‖C‖²_F, no eigendecomposition). Displacement metrics: at-point, rings, global, **blast ratio** , and collateral damage at the protected positives. **61 scenarios.** Diagnostic (D1 — the geometric-dislike dose decomposed: intended push vs effective LR vs measured movement at 1/10/100/1000 presses). Atomic probes A1– – – – `interfaceRL` (a DDPG actor-critic, recoverable from this repo's own git history at blob `755ff8b` ) differs structurally — the user HEARS `actorTarget` , a soft copy updated `target += alpha*(online-target)` at alpha=0.005, and it trains a batch of 4 from replay only every `optimiseDivisor=40` gestures. The critic half is not reproducible here (MLPCore has `train_targets` but not the per-layer gradient extraction the policy-gradient step needs), but both OUTPUT-PATH ideas are: U1 sweeps the soft-target alpha (alpha=1 IS NISPS today, a free control), U2 sweeps the train-every-Nth divisor, U3 compares actor shapes, U4 sweeps the positive-path training dose. **U4 carries a load-bearing caveat** : upstream LRs are RMSProp LRs (memlp `Layer.h:239` , `StaticMLP.h:268` ) and `nisps/ml/training.hpp` is SGD-only, so the numbers are not comparable — see ALIGNMENT.md defect 6. Their shared metric is **lurch** — how far the mapping the musician is playing moves per single gesture, averaged over the whole field. Knobs: `--spread` (1 = Xavier, 0 = uniform with NO fan_in coupling — i.e. the post-removal behaviour, measurable before paying for the refactor), `--geo-lr` , `--geo-iters` . Driven by `scripts/bench-ml.sh` . **Two contracts that fail SILENTLY and are pinned by tests:** (1) a thumbs-up must go through BOTH `mlp.add_example` AND `fb.store_positive` — `dislike_geometric` k-NNs the replay buffer, not the MLP dataset, so a harness that only calls `add_example` measures the cold-start branch instead; (2) `placed_output()` is valid ONLY while state is `Placing` — after `commit_place()` /`commit_reposition()` the vector moves to `committed_output()` , and reading the wrong one yields an empty span whose `l2()` is 0, i.e. a broken lifecycle scored as a perfect placement.
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- Measurement (asserts nothing): `engine_bench.cpp` + `bench_report.mjs` — per-engine throughput (ns/sample, blocks/s, realtime factor) for the `process()` hot path. ONE source compiled twice (CMake `nisps_engine_bench` natively, emcc for WASM) so the two targets are comparable without adding a single export to `nisps/wasm/bindings.cpp` . Engines are driven into a working state (transport running + event drain for the sequencers, periodic `note_on` for paf_synth, a noise+sine input bed for the fx/analysis engines) and every row prints its own working-state evidence, so a number produced by an idle engine is visible rather than plausible. Driven by `scripts/bench-engines.sh` .
2026-04-29 18:57:27 +02:00
### `scripts/` — build + verify entry points
- `build-firmware.sh` , `flash-firmware.sh` , `build-and-flash-firmware.sh` , `firmware-common.sh` — Arduino-CLI wrapper for RP2350 target with C++20 flag.
2026-07-13 23:27:56 +02:00
- `build-wasm.sh` — Emscripten compile producing `manifold/public/nisps.{wasm,js}` .
2026-04-29 18:57:27 +02:00
- `build-cpp-tests.sh` — CMake configure + build + ctest (Ninja).
- `parity-check.sh` — runs native + WASM and diffs binary outputs.
test(ml): behavioural benchmark + 20 invariants for the control mapping
NISPS is a controller, not a synth: the object of study is the mapping
f: control-space -> parameter-space and how a musician's gestures deform
it. Loss measures fit to points the user dictated, which is the one thing
they never experience. So this measures geometry and gesture-response.
tests/cpp/ml_bench.cpp 61 scenarios, REPORTS never asserts (same
discipline as engine_bench.cpp). Shape-
agnostic via MLPCore<DynamicStorage>
(--shape, default 2,16,16,16,8), seeded
RNG throughout, branch points replayed
from scratch rather than snapshotted.
tests/cpp/test_ml_behaviour.cpp 20 asserting invariants, wired into
nisps_core_tests.
scripts/bench-ml.sh native + WASM from one source; --compare,
--sweep-shape, --smoke, --scenario, --seed.
Documents two contracts that fail SILENTLY (both now pinned by tests):
a thumbs-up must call BOTH mlp.add_example() and fb.store_positive(),
since dislike_geometric k-NNs the replay buffer and not the MLP dataset;
and placed_output() is valid only while state == Placing, after which an
empty span whose l2() is 0 scores a broken lifecycle as a perfect place.
ALIGNMENT defect 6 re-ranked (SGD-vs-RMSProp is not a research axis - it
silently invalidated every ported hyperparameter) and split into 6b (the
geometric dislike was ported from a superseded upstream design) and 6c
(InterfaceRL, the reference impl, is not in the tree).
Gates: build-cpp-tests (138 tests, ctest 4/4), parity-check PASS,
lint-cpp clean, bench-ml.sh --smoke runs end to end.
2026-07-25 11:02:24 +02:00
- `bench-ml.sh` — the ML BEHAVIOUR benchmark on native + WASM from one source (same trick as `bench-engines.sh` ). `--shape` , `--scenario` , `--smoke` , `--seed` , `--compare` , and `--sweep-shape` (runs the corpus across a ladder of architectures and arities — the knob-sensitivity instrument). **Reports, never asserts** : a cliff index is a description, not a pass/fail. Invariants live in `tests/cpp/test_ml_behaviour.cpp` instead. Reports land in `nisps/build/bench-ml/` .
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- `bench-engines.sh` — engine throughput on native + WASM; `--compare <report.json>` prints per-engine Δ%. **Reports, never asserts** (a wall-clock threshold on shared hardware is meaningless or flaky — same call as the firmware size job). Reports land in `nisps/build/bench/` .
2026-04-29 18:57:27 +02:00
- `lint-cpp.sh` — `.f` literal warn + heap/`Arduino.h` violation fail.
- `run-all-tests.sh` — master verification script.
### `.github/workflows/`
2026-07-13 23:27:56 +02:00
- `ci.yml` — GitHub Actions: cmake build + ctest + WASM build + parity check + lint + Playwright (cpp-tests + manifold-tests jobs). Firmware compile is documented as manual.
2026-04-29 18:57:27 +02:00
docs(audit): simplification audit 2026-07 — recon findings, phased plan, ALIGNMENT rewrite
66-agent adversarially-verified audit vs the five-bullet one-core vision.
113 findings: CI red since 2026-07-13 (memllib pin on no remote), ungated
deploys, unshared mode layer, no curated/advanced split, dead-mass inventory.
Recon: docs/specs/recon/simplification-audit-2026-07.md (immutable).
Plan: docs/specs/plans/simplification-plan.md (proposal; phases gated on
operator adoption, §7 decisions). ALIGNMENT rewritten; MAP flatly-false
lines fixed (phantom MEMLCelium-upstream entry, exploration.ts, daisysp
non-submodule, pre-P5 sentence, perf-attr claims); MAIN registry updated.
2026-07-21 01:24:35 +02:00
### `src/` — submodule + vendored trees
build(firmware): migrate to PlatformIO and vendor memllib (plan §5)
One cut, no dual path. Closes ALIGNMENT defect 3 ("Arduino-CLI build
machinery is actively hostile") and vision bullet 4.
platformio.ini carries 16 [env:], one per variant, each passing
-DMEMLNAUT_MODE_TYPE; selftest passes -DNISPS_SELFTEST=1 instead. The env list
IS the registry now — the .ino comment-registry and the NISPS_ST_* token-paste
table are deleted rather than migrated. L12 noted that table was already
silently missing the currently-shipped SLPWorkshop variant, which is the whole
argument against having a second list.
Also deleted: the Python/sed machinery that rewrote the COMMITTED .ino on every
build, the sketch symlink forest, the global TFT_eSPI User_Setup.h mutation
(now -D flags — TFT_eSPI's own documented PlatformIO recipe), the UF2
boot-mount detection stack (upload_protocol=picotool talks to the bootloader
directly), and build-firmware-arch.sh entirely. Scripts 683 -> 435 lines.
memllib is vendored at lib/memllib/ from upstream e291192; no submodules
remain. VENDORED.md records provenance and the re-sync procedure.
S9: a firmware-build CI job compiles three representative envs against a cached
toolchain and reports per-variant flash/RAM. Firmware is in an automated gate
for the FIRST time. The old ci.yml comment justified excluding it as "low
verification value" — an assessment that did not survive contact, since the
SelfTest variant sat broken for an unknown period calling a DisplayDriver
method that did not exist at the pinned memllib commit, and nothing noticed
because nothing built it.
Verified: all 16 envs build from an empty cache, each within ~520 bytes of the
arduino-cli binary it replaces, flash and RAM. Measured as .text+.rodata /
.data+.bss+vector+uninitialized — NOT PlatformIO's console line, which
double-counts .data on this board. This does not prove the hardware boots; no
flash+smoke test was possible and that stays an operator chokepoint.
slpworkshop 248232/145028 pafsynth 256880/149716 selftest 216228/17960
(all 16 in the CI log format; none exceeds 2% of a 16 MB flash)
Two traps recorded so nobody rediscovers them: vendoring memllib's subdirs
without a src/ wrapper makes PlatformIO's library builder silently compile
NOTHING while still linking; and project build_flags land BEFORE the
framework's own -std=gnu++17 -Os, so build_unflags is required.
CORRECTION carried in this commit: the firmware sizes in c19d846's message and
the first version of the memllib recon doc were wrong — SLPWorkshop 145348,
PAFSynth 145300, SelfTest 141840. They came from building variants in sequence
through a SHARED incremental arduino-cli build directory, which reused stale
objects and under-reported by ~75 KB. Clean-cache rebuilds of the identical
commit give 216736/18492 for SelfTest. The real cost of the memllib upstream
bump is +216 bytes flash, not +316. Never measure firmware size through a
reused build dir.
HISTORY NOTE: this commit and the docs commit before it were rebuilt (force-push,
2026-07-21) so that each contains only what its message describes. The first
versions had the firmware deletions stranded in the docs commit by a shared-index
race between concurrent agents; content is byte-identical to the originals.
Gates: run-all-tests.sh ALL GREEN (nisps/ untouched by this change beyond
include paths); 16/16 pio envs build.
2026-07-21 20:17:58 +02:00
- **There are no submodules.** `src/memllib` was one until the Phase 4 PlatformIO migration; it is now vendored at `firmware/MEMLNaut-NISPS/lib/memllib/` . Fresh clones need no `git submodule` step.
2026-04-29 18:57:27 +02:00
### Top-level docs
- `CLAUDE.md` — long-form architecture narrative.
- `MAP.md` — this file.
- `ALIGNMENT.md` — strategic gaps + open mission questions, dated, opinionated.
2026-04-15 17:32:32 +02:00
- `README.md` — short quickstart.
2026-07-11 23:19:01 +02:00
- `AGENTS.md` — canonical agent contract: architecture, build/test, scope, and Ergo workflow.
2026-04-15 17:32:32 +02:00
## Entry points
2026-04-29 18:57:27 +02:00
- **Firmware**: `scripts/build-firmware.sh [VARIANT]` (interactive prompt if omitted), `scripts/flash-firmware.sh` , `scripts/build-and-flash-firmware.sh` . Target: `rp2040:rp2040:solderparty_rp2350_stamp_xl:opt=Optimize3` , `-std=gnu++20` .
2026-07-13 23:27:56 +02:00
- **Manifold dev**: `cd manifold && bun install && bun run dev` (Vite, COOP/COEP headers).
- **Manifold build**: `cd manifold && bun run build` .
2026-04-29 18:57:27 +02:00
- **WASM rebuild**: `bash scripts/build-wasm.sh` (needs `emcc` ).
- **Host C++ tests**: `bash scripts/build-cpp-tests.sh` .
- **Parity check**: `bash scripts/parity-check.sh` .
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- **Engine benchmark**: `bash scripts/bench-engines.sh` (add `--compare nisps/build/bench/latest.json` to diff against the previous run).
test(ml): behavioural benchmark + 20 invariants for the control mapping
NISPS is a controller, not a synth: the object of study is the mapping
f: control-space -> parameter-space and how a musician's gestures deform
it. Loss measures fit to points the user dictated, which is the one thing
they never experience. So this measures geometry and gesture-response.
tests/cpp/ml_bench.cpp 61 scenarios, REPORTS never asserts (same
discipline as engine_bench.cpp). Shape-
agnostic via MLPCore<DynamicStorage>
(--shape, default 2,16,16,16,8), seeded
RNG throughout, branch points replayed
from scratch rather than snapshotted.
tests/cpp/test_ml_behaviour.cpp 20 asserting invariants, wired into
nisps_core_tests.
scripts/bench-ml.sh native + WASM from one source; --compare,
--sweep-shape, --smoke, --scenario, --seed.
Documents two contracts that fail SILENTLY (both now pinned by tests):
a thumbs-up must call BOTH mlp.add_example() and fb.store_positive(),
since dislike_geometric k-NNs the replay buffer and not the MLP dataset;
and placed_output() is valid only while state == Placing, after which an
empty span whose l2() is 0 scores a broken lifecycle as a perfect place.
ALIGNMENT defect 6 re-ranked (SGD-vs-RMSProp is not a research axis - it
silently invalidated every ported hyperparameter) and split into 6b (the
geometric dislike was ported from a superseded upstream design) and 6c
(InterfaceRL, the reference impl, is not in the tree).
Gates: build-cpp-tests (138 tests, ctest 4/4), parity-check PASS,
lint-cpp clean, bench-ml.sh --smoke runs end to end.
2026-07-25 11:02:24 +02:00
- **ML behaviour benchmark**: `bash scripts/bench-ml.sh` (`--smoke` for a fast run, `--shape 2,16,16,16,8` , `--sweep-shape` for the architecture/arity sweep, `--compare` to diff).
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- **All tests**: `bash scripts/run-all-tests.sh` (stage 6 is a bench smoke report; it does not gate).
2026-07-13 23:27:56 +02:00
- **Playwright**: `cd manifold && node node_modules/.bin/playwright test` (non-snap node runner on the VPS — BUILD-PLAN gotcha; `bunx playwright test` works elsewhere).
refactor(ml): one runtime-configurable training default (S26)
The operator's call: "there should be one default learning rate and one
default max iterations and they should both be configurable at runtime."
There were SIX copies, not the four the audit described, and they did not
agree:
nisps/ml/mlp.hpp no-arg train() hardcoding 1.f / 1000u / 0.001f —
and firmware's ONLY training path calls exactly
this, so firmware had no runtime knob at all
wasm-iml.ts train() and trainAsync() TS default params (x2)
engine-api.ts learningRate ?? 1.0, with no maxIterations knob
vcv/src/iml.hpp 200 / 0.1 / 0.00001 — silently divergent
external_synth_midi.hpp its own kDefaultLearningRate/kDefaultMaxIterations
schemas/modes/*.json x9, identical, read by nobody at runtime
Now: schemas/ml_defaults.json is the single declaration (validated against a
sibling meta-schema, matching the midi_device.schema.json convention), codegen
emits it to C++ and TS in the same run, and MLPCore carries a TrainConfig whose
default member initialisers read the generated constant.
set_train_config()/nisps_ml_set_train_config() make it runtime-overridable on
every target; the explicit-argument train() overload is untouched. min_error
joins the tuple — it was duplicated identically and belongs with the other two.
The per-mode ml block loses default_learning_rate/default_max_iterations.
default_spread stays (genuinely wired on both targets) and input_channels stays
(codegen-time validated, real information for sound_analysis_midi).
VCV BEHAVIOUR CHANGE, deliberate: MEMLNaut.cpp constructs IML positionally and
relies on those defaults, so the module moves to 1000/1.0/0.001 — 5x the max
iterations, 10x the learning rate, and a 100x looser early-stop threshold. The
old values were never justified anywhere; they arrived with fbc68eb alongside
an unrelated module rewrite and no tuning rationale. Firmware and WASM have
shipped 1.0/1000 all along. It is now runtime-settable if this turns out worse.
The generated header lands in nisps/ml/generated/, not nisps/modes/generated/
where the rest of codegen output lives: training hyperparameters are an ML
fact, and nisps/ml sits below nisps/modes, so emitting them there would make
mlp.hpp include upward. The agent that built this flagged the directory-crossing
rather than hiding it; this is the fix. CI's generated-freshness gate learns the
new directory.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS (max delta 2.38e-7),
lint clean, manifold typecheck + 17 unit + 33 e2e (which exercise train() and
trainAsync() through a real browser).
2026-07-21 17:20:10 +02:00
- **Codegen**: `cd codegen && bun run generate.ts` (regenerates `nisps/modes/generated/` + `nisps/ml/generated/` C++ and `manifold/src/modes/generated/` TS).
2026-04-15 17:32:32 +02:00
## Conventions
2026-04-29 18:57:27 +02:00
build(firmware): migrate to PlatformIO and vendor memllib (plan §5)
One cut, no dual path. Closes ALIGNMENT defect 3 ("Arduino-CLI build
machinery is actively hostile") and vision bullet 4.
platformio.ini carries 16 [env:], one per variant, each passing
-DMEMLNAUT_MODE_TYPE; selftest passes -DNISPS_SELFTEST=1 instead. The env list
IS the registry now — the .ino comment-registry and the NISPS_ST_* token-paste
table are deleted rather than migrated. L12 noted that table was already
silently missing the currently-shipped SLPWorkshop variant, which is the whole
argument against having a second list.
Also deleted: the Python/sed machinery that rewrote the COMMITTED .ino on every
build, the sketch symlink forest, the global TFT_eSPI User_Setup.h mutation
(now -D flags — TFT_eSPI's own documented PlatformIO recipe), the UF2
boot-mount detection stack (upload_protocol=picotool talks to the bootloader
directly), and build-firmware-arch.sh entirely. Scripts 683 -> 435 lines.
memllib is vendored at lib/memllib/ from upstream e291192; no submodules
remain. VENDORED.md records provenance and the re-sync procedure.
S9: a firmware-build CI job compiles three representative envs against a cached
toolchain and reports per-variant flash/RAM. Firmware is in an automated gate
for the FIRST time. The old ci.yml comment justified excluding it as "low
verification value" — an assessment that did not survive contact, since the
SelfTest variant sat broken for an unknown period calling a DisplayDriver
method that did not exist at the pinned memllib commit, and nothing noticed
because nothing built it.
Verified: all 16 envs build from an empty cache, each within ~520 bytes of the
arduino-cli binary it replaces, flash and RAM. Measured as .text+.rodata /
.data+.bss+vector+uninitialized — NOT PlatformIO's console line, which
double-counts .data on this board. This does not prove the hardware boots; no
flash+smoke test was possible and that stays an operator chokepoint.
slpworkshop 248232/145028 pafsynth 256880/149716 selftest 216228/17960
(all 16 in the CI log format; none exceeds 2% of a 16 MB flash)
Two traps recorded so nobody rediscovers them: vendoring memllib's subdirs
without a src/ wrapper makes PlatformIO's library builder silently compile
NOTHING while still linking; and project build_flags land BEFORE the
framework's own -std=gnu++17 -Os, so build_unflags is required.
CORRECTION carried in this commit: the firmware sizes in c19d846's message and
the first version of the memllib recon doc were wrong — SLPWorkshop 145348,
PAFSynth 145300, SelfTest 141840. They came from building variants in sequence
through a SHARED incremental arduino-cli build directory, which reused stale
objects and under-reported by ~75 KB. Clean-cache rebuilds of the identical
commit give 216736/18492 for SelfTest. The real cost of the memllib upstream
bump is +216 bytes flash, not +316. Never measure firmware size through a
reused build dir.
HISTORY NOTE: this commit and the docs commit before it were rebuilt (force-push,
2026-07-21) so that each contains only what its message describes. The first
versions had the firmware deletions stranded in the docs commit by a shared-index
race between concurrent agents; content is byte-identical to the originals.
Gates: run-all-tests.sh ALL GREEN (nisps/ untouched by this change beyond
include paths); 16/16 pio envs build.
2026-07-21 20:17:58 +02:00
- Firmware mode selection is compile-time only — one `-DMEMLNAUT_MODE_TYPE` per `[env:]` in `platformio.ini` .
docs: sync MAP/ALIGNMENT/AGENT-REFERENCE with the Phase 1 sweep
The three top-level orienting docs are each falsified by several of the eight
preceding commits, so they land here as one sync rather than being split
across commits that would each leave them half-true. Same phase, same push.
MAP.md — removed fixed_buffer.hpp, voice_space.hpp, the src/daisysp entry and
its symlink from the sketch-tree list, input_router.hpp/wire_inputs (inputs are
wired by bind_peripherals now), test_fixed_buffer.cpp, SplitStage/ReadoutStrip/
InputMini, BackendAdvanced.tsx, and feedback/rng.ts; corrected the primitives
count 12 -> 7; replaced the "engine LIFTED from playground/src" provenance;
rewrote the perf-attribute convention and deleted the NISPS_AUDIO_FUNC gotcha
(both macros and gotcha are gone).
docs/AGENT-REFERENCE.md — the memory-section-attribute instruction now names
only the macros that exist; input_router.hpp dropped from the firmware tree;
symlink list and the submodule-init note no longer mention daisysp.
ALIGNMENT.md — defect 5 (dead mass and registry sprawl) rewritten: the deletion
half is done, so the entry now scopes to what actually remains, which is the
registry/dual-truth half (Phase 3) plus the stale specs (docs pass). Per the
ALIGNMENT convention this is a rewrite-to-current, not a checkbox.
2026-07-21 12:49:39 +02:00
- `nisps/` follows Chris's RP2350 perf rules: no heap, `static const float` for non-trivial constants, strict `.f` suffix. `perf.hpp` now carries only `NISPS_HOT` /`NISPS_FORCE_INLINE`; the three dead/misshapen SRAM-section macros were deleted in the 2026-07 sweep (S21/L13).
2026-04-29 18:57:27 +02:00
- C++ identifiers: `PascalCase` types, `snake_case` functions/variables, `kPascalCase` constexpr. JSON keys `snake_case` . TS types `PascalCase` , components `PascalCase.tsx` , modules `kebab-case.ts` .
2026-07-18 12:23:14 +02:00
- `Curve` enum lives in `nisps/core/math.hpp` (lowercase: `linear/exp/log/square/sqrt/sigmoid/cubic` , plus the parameterised `centered_power` free function); generated mode headers re-export via `using Curve = ::nisps::Curve;` . Since P4 there is NO TS mirror — the browser samples the WASM catalog (`nisps_curve_apply(+batch)`).
2026-07-18 12:45:45 +02:00
- Modes are TSX components composed of primitives; mode parameter contracts are JSON schemas with codegen → C++ **and** TS types (`MF_MODES` derives params/ml-config from the generated schemas since P5; labels/ordering stay a manifold overlay). **No declarative JSON UI.**
docs(audit): simplification audit 2026-07 — recon findings, phased plan, ALIGNMENT rewrite
66-agent adversarially-verified audit vs the five-bullet one-core vision.
113 findings: CI red since 2026-07-13 (memllib pin on no remote), ungated
deploys, unshared mode layer, no curated/advanced split, dead-mass inventory.
Recon: docs/specs/recon/simplification-audit-2026-07.md (immutable).
Plan: docs/specs/plans/simplification-plan.md (proposal; phases gated on
operator adoption, §7 decisions). ALIGNMENT rewritten; MAP flatly-false
lines fixed (phantom MEMLCelium-upstream entry, exploration.ts, daisysp
non-submodule, pre-P5 sentence, perf-attr claims); MAIN registry updated.
2026-07-21 01:24:35 +02:00
- WASM and firmware share the same C++; the browser MLP is runtime-shaped (`MLPCore< DynamicStorage > `, since P2): `nisps_ml_create` honours `(input, output, hidden[3])` with non-positive/null args defaulting to `32→[10,14,18]→126` ; `nisps_ml_reshape` warm-starts a new shape. Firmware keeps compile-time `MLP<...>` (zero heap). Per-mode dims are schema-real on both targets since P5.3 (the browser reshapes on mode switch).
2026-04-29 18:57:27 +02:00
- Cross-platform parity: `scripts/parity-check.sh` enforces native vs WASM agreement within 1e-5.
2026-04-15 17:32:32 +02:00
## Gotchas
2026-04-29 18:57:27 +02:00
build(firmware): migrate to PlatformIO and vendor memllib (plan §5)
One cut, no dual path. Closes ALIGNMENT defect 3 ("Arduino-CLI build
machinery is actively hostile") and vision bullet 4.
platformio.ini carries 16 [env:], one per variant, each passing
-DMEMLNAUT_MODE_TYPE; selftest passes -DNISPS_SELFTEST=1 instead. The env list
IS the registry now — the .ino comment-registry and the NISPS_ST_* token-paste
table are deleted rather than migrated. L12 noted that table was already
silently missing the currently-shipped SLPWorkshop variant, which is the whole
argument against having a second list.
Also deleted: the Python/sed machinery that rewrote the COMMITTED .ino on every
build, the sketch symlink forest, the global TFT_eSPI User_Setup.h mutation
(now -D flags — TFT_eSPI's own documented PlatformIO recipe), the UF2
boot-mount detection stack (upload_protocol=picotool talks to the bootloader
directly), and build-firmware-arch.sh entirely. Scripts 683 -> 435 lines.
memllib is vendored at lib/memllib/ from upstream e291192; no submodules
remain. VENDORED.md records provenance and the re-sync procedure.
S9: a firmware-build CI job compiles three representative envs against a cached
toolchain and reports per-variant flash/RAM. Firmware is in an automated gate
for the FIRST time. The old ci.yml comment justified excluding it as "low
verification value" — an assessment that did not survive contact, since the
SelfTest variant sat broken for an unknown period calling a DisplayDriver
method that did not exist at the pinned memllib commit, and nothing noticed
because nothing built it.
Verified: all 16 envs build from an empty cache, each within ~520 bytes of the
arduino-cli binary it replaces, flash and RAM. Measured as .text+.rodata /
.data+.bss+vector+uninitialized — NOT PlatformIO's console line, which
double-counts .data on this board. This does not prove the hardware boots; no
flash+smoke test was possible and that stays an operator chokepoint.
slpworkshop 248232/145028 pafsynth 256880/149716 selftest 216228/17960
(all 16 in the CI log format; none exceeds 2% of a 16 MB flash)
Two traps recorded so nobody rediscovers them: vendoring memllib's subdirs
without a src/ wrapper makes PlatformIO's library builder silently compile
NOTHING while still linking; and project build_flags land BEFORE the
framework's own -std=gnu++17 -Os, so build_unflags is required.
CORRECTION carried in this commit: the firmware sizes in c19d846's message and
the first version of the memllib recon doc were wrong — SLPWorkshop 145348,
PAFSynth 145300, SelfTest 141840. They came from building variants in sequence
through a SHARED incremental arduino-cli build directory, which reused stale
objects and under-reported by ~75 KB. Clean-cache rebuilds of the identical
commit give 216736/18492 for SelfTest. The real cost of the memllib upstream
bump is +216 bytes flash, not +316. Never measure firmware size through a
reused build dir.
HISTORY NOTE: this commit and the docs commit before it were rebuilt (force-push,
2026-07-21) so that each contains only what its message describes. The first
versions had the firmware deletions stranded in the docs commit by a shared-index
race between concurrent agents; content is byte-identical to the originals.
Gates: run-all-tests.sh ALL GREEN (nisps/ untouched by this change beyond
include paths); 16/16 pio envs build.
2026-07-21 20:17:58 +02:00
- Firmware needs PlatformIO: `nix-shell -p platformio-core` . Use `platformio-core` , NOT `platformio` — the latter is nixpkgs' bubblewrap-wrapped FHS build and fails without a working user namespace. First build pulls ~1-2 GB into `~/.platformio` .
2026-04-29 18:57:27 +02:00
- `firmware/MEMLNaut-NISPS/glue/mode_select.hpp` `#undef` s Arduino macros (`sq`, `min` , `max` , `abs` , `round` ) before pulling nisps headers — engines use those identifiers as method names.
build(firmware): migrate to PlatformIO and vendor memllib (plan §5)
One cut, no dual path. Closes ALIGNMENT defect 3 ("Arduino-CLI build
machinery is actively hostile") and vision bullet 4.
platformio.ini carries 16 [env:], one per variant, each passing
-DMEMLNAUT_MODE_TYPE; selftest passes -DNISPS_SELFTEST=1 instead. The env list
IS the registry now — the .ino comment-registry and the NISPS_ST_* token-paste
table are deleted rather than migrated. L12 noted that table was already
silently missing the currently-shipped SLPWorkshop variant, which is the whole
argument against having a second list.
Also deleted: the Python/sed machinery that rewrote the COMMITTED .ino on every
build, the sketch symlink forest, the global TFT_eSPI User_Setup.h mutation
(now -D flags — TFT_eSPI's own documented PlatformIO recipe), the UF2
boot-mount detection stack (upload_protocol=picotool talks to the bootloader
directly), and build-firmware-arch.sh entirely. Scripts 683 -> 435 lines.
memllib is vendored at lib/memllib/ from upstream e291192; no submodules
remain. VENDORED.md records provenance and the re-sync procedure.
S9: a firmware-build CI job compiles three representative envs against a cached
toolchain and reports per-variant flash/RAM. Firmware is in an automated gate
for the FIRST time. The old ci.yml comment justified excluding it as "low
verification value" — an assessment that did not survive contact, since the
SelfTest variant sat broken for an unknown period calling a DisplayDriver
method that did not exist at the pinned memllib commit, and nothing noticed
because nothing built it.
Verified: all 16 envs build from an empty cache, each within ~520 bytes of the
arduino-cli binary it replaces, flash and RAM. Measured as .text+.rodata /
.data+.bss+vector+uninitialized — NOT PlatformIO's console line, which
double-counts .data on this board. This does not prove the hardware boots; no
flash+smoke test was possible and that stays an operator chokepoint.
slpworkshop 248232/145028 pafsynth 256880/149716 selftest 216228/17960
(all 16 in the CI log format; none exceeds 2% of a 16 MB flash)
Two traps recorded so nobody rediscovers them: vendoring memllib's subdirs
without a src/ wrapper makes PlatformIO's library builder silently compile
NOTHING while still linking; and project build_flags land BEFORE the
framework's own -std=gnu++17 -Os, so build_unflags is required.
CORRECTION carried in this commit: the firmware sizes in c19d846's message and
the first version of the memllib recon doc were wrong — SLPWorkshop 145348,
PAFSynth 145300, SelfTest 141840. They came from building variants in sequence
through a SHARED incremental arduino-cli build directory, which reused stale
objects and under-reported by ~75 KB. Clean-cache rebuilds of the identical
commit give 216736/18492 for SelfTest. The real cost of the memllib upstream
bump is +216 bytes flash, not +316. Never measure firmware size through a
reused build dir.
HISTORY NOTE: this commit and the docs commit before it were rebuilt (force-push,
2026-07-21) so that each contains only what its message describes. The first
versions had the firmware deletions stranded in the docs commit by a shared-index
race between concurrent agents; content is byte-identical to the originals.
Gates: run-all-tests.sh ALL GREEN (nisps/ untouched by this change beyond
include paths); 16/16 pio envs build.
2026-07-21 20:17:58 +02:00
- `nisps_firmware::g_active_mode_bridge` is `extern` in `glue/audio_driver.hpp` and defined in `src/main.cpp` ; combining `inline` with `__not_in_flash` produces a comdat conflict at link time.
- `pio run` 's own "Flash: NN%" console line double-counts `.data` on this board (PlatformIO's generic size checker counts every PROGBITS+ALLOC section). Compare `arm-none-eabi-size -A` — flash = `.text+.rodata` — before believing a size regression.
feat: curve truth, DriverConfig, real telemetry, engine benchmark
Four items from one workflow, committed together because their build and CI
wiring genuinely interleaves — nisps/CMakeLists.txt, run-all-tests.sh and
ci.yml each carry hunks from two of them, and the stage renumbering (1/5 ->
1/6) touches every line. Splitting would produce commits that do not build,
which is worse than a commit that does four things and says so.
S26 part 2 — the curve declaration now matches reality. params[].curve stays
the mode-wide DEFAULT; a voice_spaces entry may now be {name, curve_overrides}
declaring only the slots where THAT voice space deviates. The 6 modes with one
voice space are byte-identical. The values were derived MECHANICALLY by a new
codegen/curve-audit.ts that models the four idioms a p[N]*p[N] regex misses
(alias form, memlcelium's implicit-counter sq() lambda, loop-generated indices,
smooth_params_), inlines helpers, and RAISES rather than guessing when it
cannot reduce an expression. A drift gate cross-checks 1179 (voice space x
param) slots against engine source on every run and was proved to fail loudly
on three drift classes. Application stays in the engine: nisps/engines,
nisps/pipeline and nisps/core are untouched, generated output is pure insertion
(755 insertions, 0 deletions), and the rebuilt nisps.wasm was byte-identical.
S4 / 7.2 — firmware reads the active mode's driver config at mode start, and
mic/line is real. My brief assumed the engine owns this; the code disagreed and
the code was right. sound_analysis_midi's EngineT is NoOpEngine — the mic lives
on a separately-composed AnalysisEngine member — so engine-level wiring would
have compiled, passed every gate, and left the one mic mode on line input.
Hence a mode-level seam defaulting to engine().driver_config(). Separately,
DriverConfig's defaults (line_level 0, output_volume 1.0) had drifted from
memllib's actual 3/0.8 because nothing had ever read them; wiring them as-is
would have made every silent mode louder and its line input maximally
insensitive — a behaviour change disguised as plumbing. Now pinned by a test.
Also: GetSysClockSpeed() panic()s on unsupported sample rates and runs on the
first line of setup(), so sample_rate needed a fallback ahead of clock setup.
CI's firmware env list gains soundanalysismidi — it is the only mic variant and
nothing else compiles that path.
Plan 5e — telemetry is real. A loss_history C-API entry across the full 5-layer
chain lets the browser read the per-iteration loss the core already records.
The audit named one fabrication site; there were two — wasm-iml.ts's
synchronous train() published lossHistory: [loss] as well. A third, ctx.loss,
was not merely dead but actively synthetic (fallbacks of prev * 0.82 and a
literal 0.5, rendered by nothing) and is deleted. The firmware buffer stays
untouched, per the L25 call. EngineApi.lossHistory() reads spine state rather
than the MLP handle, because trainAsync() fits on the worker's mirror net and
the handle would give a subtly-wrong second answer.
Plan 5f — engine throughput is measurable. One source compiled twice (CMake
natively, emcc for WASM) so the targets compare directly and no WASM export is
added. Sequencers are driven into a working state, and every row prints its own
working-state evidence so a number produced by an idle engine is visible rather
than plausible. Reports, never asserts: a wall-clock threshold on shared
hardware is meaningless or flaky, same call as the firmware size job.
ALIGNMENT: the telemetry defect is deleted (built, not deferred); the
performance defect is rewritten to what is actually left — these are HOST
numbers, and nothing measures the RP2350 at 150 MHz, which is the target the
mission's constraint is about. Q4 (memllib ownership) and Q5 (legacy feedback
modes) are closed.
Corrections to my own earlier claims, both found by agents contradicting the
brief: manifold/ONBOARDING.md was NOT "now accurate" — its primitives list
still named five deleted primitives and cited a seededGradient() that does not
exist. And the parity harness misses the sequencer engines because it runs 128
frames while their sequencers evaluate every 400-500 samples, NOT because
all-params-0.5 fails to trigger them (it does trigger: 0.5 maps to ratio 2,
firing three times per bar). The fix is a longer window, not different params.
Gates: run-all-tests.sh ALL GREEN — 4/4 ctest, parity PASS, lint clean, curve
drift 1179 slots ok, 39 e2e (was 33). Firmware: 5 envs built including the mic
variant.
2026-07-21 22:02:23 +02:00
- `nisps_modes_tests` builds against generated schemas under `nisps/modes/generated/` ; if you add a new mode, regenerate via `bun run codegen/generate.ts` before building. It also compiles one firmware header (`glue/codec_config.hpp`), so the repo root is on its include path.
- `nisps::DriverConfig` 's member defaults are load-bearing: they reproduce memllib's historical hardcoded codec setup, so a mode that declares nothing gets exactly the pre-wiring behaviour. Changing them changes the codec setup of every mode that expresses no opinion (pinned by `tests/cpp/test_mode_driver_config.cpp` ).
2026-04-29 18:57:27 +02:00
## Smells / strategic concerns
See `ALIGNMENT.md` .
2026-06-28 22:36:58 +02:00
## Specs
- **Root**: `docs/specs/`
2026-07-13 22:15:46 +02:00
- **Entry**: `MAIN.md`
- **Layout**: flat (with `plans/` , `recon/` , `_archive/` subdirs)
- **Index**: none (intentionally — generate when a consumer exists)
2026-06-28 22:36:58 +02:00
- **Skill**: invoke `/specs` to review/maintain/add/navigate.
2026-07-13 22:15:46 +02:00
- **Conventions**: Four-genre ontology — `kind: spec` (timeless contract, wins by intent), `kind: plan` (status: active|executed|superseded, never authority for behaviour), `kind: finding` (dated, immutable, exempt from drift lint), ADRs in `docs/adr/` .
2026-06-28 22:36:58 +02:00
2026-07-13 22:15:46 +02:00
The corpus holds platform-level specs (engine architecture, I/O backends, feedback design), implementation specs (port specs, wire protocols), feature specs (e.g. slp-workshop-firmware.md), historical findings (dated research artifacts), and finite build plans. Before changing behaviour a spec covers, find it via `/specs` ; the spec wins by intent — if it's wrong, update it in the same commit as the code.
