Comp-Core / LUME CC / N'Ko Unification

Comp-Core / LUME CC / N'Ko Unification

Purpose

This document is the correction layer for the computational choreography package.
It aligns the choreography docs with the actual implementation in Comp-Core,
MotionMixApp, and the N'Ko audio/ASR work.

The short version:

body / sensors
  -> LUME CC / Echelon latent state
  -> audio + visuals + training captures
  -> ClaimBridge / cc-inscription
  -> N'Ko motion inscription claims

speech / audio
  -> N'Ko trajectory-biased Transformer CTC
  -> N'Ko text hypothesis
  -> MAOE-N'Ko correction/admissibility layer
  -> corrected or preserved N'Ko text

These are connected lanes. They are not the same model, and the docs should not
pretend they are.

The Three Current Lanes

1. LUME CC / Motion Lane

The motion lane is the body-to-latent runtime.

Implemented pieces:

• MotionMixApp `EchelonBridge.swift`
• Comp-Core `core/audio-media/cc-echelon/`
• Rust `cc-brain`
• SAN / Echelon dynamics
• `echelon_get_dynamics_128`
• audio bridge and real-time synth controls
• `SANTrajectoryLogger` for training capture

Its job is to turn body state into a stable real-time latent representation that
can drive music, visual state, camera behavior, training capture, and inscription.

2. N'Ko Motion Inscription Lane

The inscription lane is the body-to-words path.

Implemented pieces:

• MotionMixApp `ClaimBridgeService.swift`
• Rust `cc-brain/src/san/claim_bridge.rs`
• Rust FFI in `cc-brain/src/ffi.rs`
• `cc-inscription`
• Convex `inscriptions.ts`
• Comp-Core `docs/nko-ecosystem/`
• Comp-Core `docs/motion-language/`

This lane does not need full natural-language generation to be real. It already
turns movement into typed N'Ko claim signals. The current "words" are controlled
technical inscriptions:

ߛ stabilization
ߜ dispersion
ߕ transition
ߙ return
ߡ dwell
ߚ oscillation
ߞ recovery
ߣ novelty
ߠ place-shift
ߥ echo

That is the current computational vocabulary. Natural N'Ko phrasing can be added
later on top of these operator claims, but the operator layer is already the
right foundation because it is deterministic, learnable, and traceable.

3. N'Ko Audio / ASR Lane

The N'Ko audio lane is speech-to-N'Ko text.

Current evidence says:

- The verified ASR anchor is the trajectory-biased Transformer CTC model in
`Desktop/nko-brain-scanner`.
- The local confirmed number is 20.57
- The important mechanism is trajectory bias inside the CTC model, not MAOE.
- MAOE-N'Ko is the correction, routing, and admissibility layer around the ASR
anchor.

So the correct ASR wording is:

audio
  -> trajectory-biased Transformer CTC
  -> N'Ko hypothesis + trajectory telemetry
  -> MAOE partition router
  -> bounded correction / preserve / quarantine decision

Do not call MAOE the trained ASR model unless a future same-snapshot run proves a
MAOE-trained or MAOE-corrected system beats the anchor.

How These Lanes Actually Meet

They meet at inscription, not at a magical shared model.

MotionMix / LUME CC
  -> 128D dynamics
  -> ClaimBridge
  -> N'Ko sigil claim
  -> Convex / inscription corpus

N'Ko ASR
  -> N'Ko text hypothesis
  -> MAOE admissibility
  -> speech-sourced inscription
  -> same corpus family

Motion and speech can both write into the inscription layer. That is the common
record. The future shared latent can train on this record, but it should not be
claimed as already trained.

What "Mutation" Means Here

There are two real mutation systems in the repo.

Convex Mutations

`Comp-Core/apps/convex-memory/convex/inscriptions.ts` exposes write mutations:

• `inscribeMotion`
• `inscribeSpeech`
• `inscribeConversation`

These accept the ten valid sigils, gate low-confidence claims, and write
`inscriptionUnits`.

This is the application-level mutation surface: new motion, speech, or
conversation evidence becomes a persisted inscription row.

Semantic Lifecycle Mutation

`cc-semantic-language` handles semantic lifecycle transitions:

proto -> provisional -> canonical -> deprecated

The important files are:

• `core/semantic/cc-semantic-language/src/lifecycle/transitions.rs`
• `core/semantic/cc-semantic-language/src/storage/event_log.rs`
• `core/semantic/cc-semantic-language/src/compiler/nko/operator_assignment.rs`

This is not "rewrite the past." It is append-only semantic evolution. A word,
operator sequence, or surface form can mature, be deprecated, or be reinterpreted
through a new lifecycle state, but the historical evidence remains traceable.

What Is Already Real

These are implemented and should be described as real:

- LUME CC/Echelon motion-to-audio and 128D dynamics runtime.
- ClaimBridge detection from 104D/128D latent state to ten N'Ko claim signals.
- MotionMixApp posting motion inscriptions through `ClaimBridgeService`.
- Convex inscription mutations for motion, speech, and conversation sources.
- `cc-inscription` typed claim IR and N'Ko surface rendering.
- `cc-semantic-language` N'Ko segmentation, operator assignment, grammar
validation, invariance scoring, and lifecycle transitions.
- N'Ko trajectory-biased Transformer CTC ASR anchor.
- MAOE-N'Ko as a routed correction/admissibility architecture.

What Is Not Yet Real

These should be framed as future work:

• A fully trained body + speech + N'Ko shared latent space.
• Natural N'Ko sentence generation from every movement phrase.
• A proven MAOE-improved ASR result over the 20.57
• A single model that learns all DJ/AirDeck gestures and all N'Ko claims together.
• A runtime where K11/Rekordbox directly consumes N'Ko ASR or Mac4 Unity events.

Source Files To Treat As Ground Truth

Motion / LUME CC:

• `Desktop/MotionMixApp/MotionMixApp/Services/EchelonBridge.swift`
• `Desktop/MotionMixApp/MotionMixApp/Services/SANTrajectoryLogger.swift`
• `Desktop/Comp-Core/core/audio-media/cc-echelon/`
• `Desktop/Comp-Core/core/audio-media/cc-echelon/crates/cc-brain/src/ffi.rs`

Motion inscription:

• `Desktop/MotionMixApp/MotionMixApp/Services/ClaimBridgeService.swift`
• `Desktop/Comp-Core/core/audio-media/cc-echelon/crates/cc-brain/src/san/claim_bridge.rs`
• `Desktop/Comp-Core/core/semantic/cc-inscription/`
• `Desktop/Comp-Core/apps/convex-memory/convex/inscriptions.ts`

N'Ko semantics:

• `Desktop/Comp-Core/docs/nko-ecosystem/README.md`
• `Desktop/Comp-Core/docs/nko-ecosystem/SIGIL_SEMANTICS.md`
• `Desktop/Comp-Core/docs/motion-language/MOTION_SEMANTICS.md`
• `Desktop/Comp-Core/core/semantic/cc-semantic-language/`

N'Ko audio / ASR:

• `Desktop/nko-brain-scanner/HANDOFF.md`
• `Desktop/nko-brain-scanner/asr/trajectory_asr.py`
• `Desktop/nko-brain-scanner/constrained/trajectory_bias.py`
• `Desktop/MAOE-NKo-Technical-Architecture.md`
• `Desktop/Comp-Core/experiments/agp_mlx/asr_bridge/`

Wording Rules Going Forward

Use:

> LUME CC maps body state into a realtime latent. ClaimBridge reads that latent
> and emits N'Ko motion-inscription claims. The N'Ko audio lane separately uses a
> trajectory-biased Transformer CTC ASR anchor, with MAOE as a bounded correction
> and admissibility layer.

Avoid:

> MAOE is the trained ASR core.

Avoid:

> ClaimBridge is a placeholder.

Avoid:

> The N'Ko connection is decorative.

Avoid:

> Motion and speech already share one trained latent model.