NKo — Notation, Memory, and Future Routing

NKo — Notation, Memory, and Future Routing

NKo's role in the LUME stack has two layers. The first is what to build now.
The second is what to build after.

Layer 1 — Notation and Memory (Build Now)

NKo is the durable naming and inscription layer for movement phrases.
This is the immediate role, and it should be built before anything else.

Every proven movement phrase should have:
- A technical name (`left_hand_raise_play`, `wave`, `weighted_hold`)
- A body description (landmarks, zone, timing, intent)
- An evidence record (session hash, clip count, gate result)
- A claim status (defined / observed / confirmed / promoted)
- Optional: a NKo name assigned after the phrase is stable

This is the movement alphabet. The archive of body inscriptions is how
the system stops treating movement as ephemeral sensor data and starts treating
it as a vocabulary with history.

The NKo writing system (Solomana Kanté, 1949) fits this role because it was
invented precisely to give Mande languages a durable computational structure:
a notation system, a memory layer, and a cultural claim about who the language
belongs to. Body movement — especially Mo's specific movement vocabulary — deserves
the same treatment.

Build Layer 1 before Layer 2. Do not force NKo into the live control loop until
the gesture schema is reliable and the body inscriptions are stable.

Layer 2 — ASR Correction And Possible Future Routing

Verified N'Ko ASR anchor:

audio -> trajectory-biased Transformer CTC -> N'Ko hypothesis

MAOE-NKo is not the trained acoustic ASR core. It is the correction,
admissibility, and routing layer around the ASR output unless a future same-run
experiment proves otherwise.

For movement, the transferable idea is not "use the ASR model." The useful
transfer is:

- trajectory bias: interpret a short motion window by where it appears to be
going, not only by the current frame;
- partition/routing discipline: stable, boundary, repair, recover, and block
states before promoting an output.

This is V6+. It requires:
- The notation layer (Layer 1) to be established first
- Enough paired movement+music sessions to train the anticipatory routing
- Full 128D training data with consistent Mocopi/Watch signal

Do not implement MAOE-NKo before the notation layer is stable and the
movement vocabulary is documented.

The Two Layers Do Not Conflict

Layer 1 (inscription/memory) is always present and grows with every session.
Layer 2 (live inference routing) is a future upgrade to the SAN pipeline.

A movement phrase can have an NKo inscription and a claim record (Layer 1)
without ever being used in live MAOE routing (Layer 2). The inscription
layer is the foundation. The routing layer builds on top of it when the
time comes.

Reference Files

• `07-nko/movement-notation.md` — body inscriptions and claim types
• `07-nko/maoe-routing.md` — MAOE correction/admissibility boundary
• `Desktop/computational-choreography/07-nko-synthesis/` — full Set A docs
• `Desktop/MotionMix/research/.../07-nko-computational-language/` — Set B docs