Grand Diomande Research · Full HTML Reader

LUME Derived Lanes After Capture Spec

This spec defines what the system is allowed to derive after a strict-real capture bundle exists. It prevents a common failure mode: treating generated outputs, semantic candidates, or learned summaries as if they were the original evidence.

Embodied Trajectory Systems research note experiment writeup candidate score 18 .md

Full Public Reader

LUME Derived Lanes After Capture Spec

Generated: 2026-06-10 20:18 EDT

This spec defines what the system is allowed to derive after a strict-real capture bundle exists. It prevents a common failure mode: treating generated outputs, semantic candidates, or learned summaries as if they were the original evidence.

Rule

Raw evidence comes first.

text
raw pose / raw video / raw SAN rows / source manifest / camera topology / operator notes
  -> verified bundle
  -> strict real-bundle preflight
  -> derived lanes

Derived lanes may explain, summarize, transform, annotate, map, or propose. They cannot replace required physical lanes, source identity, camera topology, operator review, or K11 command promotion.

Required Precondition

Do not produce promotion-grade derived lanes until:

text
required physical lanes are strict real
bundle origin is not fixture or ephemeral
source manifest is complete
camera topology has non-synthetic calibration status
K11 command/safety evidence is present
Mac4 read-only evidence is present
strict real-bundle preflight is green

Draft manifests may exist earlier, but must be marked:

text
status: research_manifest_only
promotion_eligible: false

Bundle Inputs

Canonical raw inputs:

text
bundle.json
raw/sources/source_manifest.json
raw/calibration/camera_topology.json
raw/pose/iphone_16_plus_motionmix.jsonl
raw/pose/iphone_14_pro_max_motionmix.jsonl
raw/pose/mac4_bodytruth_depth_or_wide.jsonl
raw/video/mac2_insta360_room_wide.mp4
raw/san/san_frames.jsonl
raw/operator/notes.json
dj/airdeck_events.jsonl

Optional raw inputs:

text
raw/pose/iphone_16_pro_max_motionmix.jsonl
raw/video/iphone_16_pro_max_motionmix.mp4
raw/ipad/stage_view.*
raw/ipad/shoot_view.*
raw/video/k11_pose_coach_gate.*
raw/room/room_sweep.*

Derived Lane Layout

Recommended layout:

text
derived/templates/template_candidates.json
derived/templates/template_review_report.md
derived/san/training_index.json
derived/san/phrase_windows.jsonl
derived/nko/body_inscriptions.jsonl
derived/enthea/request.json
derived/unity/visual_mapping_request.json
derived/mrt2/audio_mapping_request.json
derived/demon/request.json
derived/semantic/candidates.jsonl
derived/semantic/review_report.md
derived/room/gaussian_splat/*
derived/reconstruction/request.json
derived/mac5/reconstruction_report.json
derived/mac5/reconstruction_learning_audit.json

Template Lane

Purpose:

text
turn labeled motion windows into reusable movement hypotheses

Allowed outputs:

text
wave score candidates
weighted hold candidates
torso lean candidates
both-arm burst candidates
recovery candidates
false-reactive negative examples
AirDeck command mold windows
Dance Derivation Mold windows

Template candidate row:

json
{
  "schema": "lume.template_candidate.v1",
  "session_id": "session_001",
  "window": {
    "start_s": 12.3,
    "end_s": 14.1
  },
  "template_id": "wave",
  "score": 0.74,
  "evidence": {
    "source_ids": [
      "iphone_16_plus_motionmix",
      "iphone_14_pro_max_motionmix",
      "mac4_bodytruth_depth_or_wide"
    ],
    "confidence_min": 0.51,
    "confidence_mean": 0.68
  },
  "claim_status": "observed",
  "promotion_eligible": false,
  "review_required": true
}

Template candidates become stable templates only after repeated sessions and review.

SAN Lane

Purpose:

text
build phrase/audio/gesture policy training rows and timing windows

Allowed outputs:

text
SAN phrase windows
motion regime labels
gesture probabilities
camera score priors
track context references
latency and confidence diagnostics
training index rows

SAN cannot prove:

text
3D mesh quality
camera extrinsics
multi-view reconstruction fidelity
truth labels without review
Rekordbox command eligibility

SAN training index row:

json
{
  "schema": "lume.san_training_index.v1",
  "session_id": "session_001",
  "window_id": "window_00042",
  "source_rows": [
    "raw/san/san_frames.jsonl:1200-1320"
  ],
  "motion_regime": "grounded",
  "linked_templates": [
    "weighted_hold"
  ],
  "linked_audio_context": {
    "track_id": "optional",
    "position_s": 93.4
  },
  "label_strength": "review_pending",
  "training_admissible": false
}

Rows become training-admissible only after review.

NKo/MnKO Body Inscription Lane

Purpose:

text
make movement vocabulary durable

Each inscription should include:

text
technical phrase id
body description
landmarks/zones/timing
intent
evidence sessions
claim status
optional NKo/MnKO name
review notes
promotion status

Inscription row:

json
{
  "schema": "lume.nko_body_inscription.v1",
  "phrase_id": "left_hand_raise_play",
  "technical_name": "left hand raise play pause",
  "body_description": "left wrist moves above left shoulder while torso remains present",
  "evidence_sessions": [
    "session_001"
  ],
  "claim_status": "promoted",
  "nko_name": null,
  "mnko_notes": "",
  "command_eligible": true,
  "command_gate": "K11"
}

NKo/MnKO inscriptions do not command anything by themselves.

ENTHEA / Unity Visual Lane

Purpose:

text
map trusted body state and reviewed phrases into visual response requests

Allowed outputs:

text
overlay prompts
shader parameter mappings
Unity scene cue requests
ENTHEA visual phrase requests
confidence damping rules
false-reactive damping notes
room anchor references

Visual request:

json
{
  "schema": "lume.visual_mapping_request.v1",
  "target": "unity_enthea",
  "session_id": "session_001",
  "source_windows": [
    "window_00042"
  ],
  "template_refs": [
    "wave",
    "recovery"
  ],
  "mapping": {
    "ribbon_intensity": "bounded(template.wave)",
    "particle_burst": "damped_if_false_reactive",
    "body_visibility": "must_remain_readable"
  },
  "command_eligible": false,
  "review_required": true
}

Mac4 may render this read-only. It must not send Rekordbox commands.

MRT2-Style Audio Mapping Lane

Purpose:

text
turn reviewed body state into audio response parameters

Allowed outputs:

text
tempo-follow notes
filter curve suggestions
density curve suggestions
phrase-to-audio parameter maps
SAN-linked response windows

Audio mapping request:

json
{
  "schema": "lume.mrt2_audio_mapping_request.v1",
  "session_id": "session_001",
  "source_windows": [
    "window_00042"
  ],
  "controls": {
    "filter_open": "template.both_arm_burst * confidence",
    "density": "SAN.motion_energy",
    "cooling": "template.recovery"
  },
  "command_eligible": false,
  "review_required": true
}

This is audio response, not Rekordbox command authority.

DEMON Control-Curve Lane

Purpose:

text
prepare reviewed motion-to-audio-generation curves for a future suitable host

Current status should remain:

text
research_manifest_only
promotion_eligible: false

DEMON request:

json
{
  "schema": "lume.demon_request.v1",
  "status": "research_manifest_only",
  "session_id": "session_001",
  "source_windows": [
    "window_00042"
  ],
  "control_curves": [
    {
      "name": "grounded_pressure",
      "source": "template.weighted_hold",
      "range": [
        0.0,
        1.0
      ]
    }
  ],
  "host_ready": false,
  "requires_cuda_or_equivalent": true,
  "command_eligible": false,
  "review_required": true
}

DEMON must not receive raw camera frames as truth, must not send Rekordbox commands, and must not replace SAN or reconstruction evidence.

Semantic Candidate Lane

Purpose:

text
generate candidate descriptions, phrase names, review notes, and control-curve suggestions

Allowed inputs:

text
verified source manifests
SAN windows
template candidates
operator notes
NKo/MnKO inscriptions
reconstruction windows

Allowed outputs:

text
candidate phrase names
candidate NKo/MnKO notes
template descriptions
DEMON curve notes
ENTHEA/Unity prompt candidates
review annotations

Semantic sidecars cannot replace:

text
raw physical evidence
human labels
camera calibration
multi-view reconstruction
K11 promotion manifest

Room Reconstruction Lane

Purpose:

text
solve the room or stage context for future visuals and camera understanding

Allowed outputs:

text
COLMAP sparse solve
Gaussian Splat PLY
camera path report
room anchor notes
Unity/ENTHEA room asset references

Room reconstruction cannot satisfy required body evidence lanes and cannot claim the first real body reconstruction by itself.

Review And Promotion Rules

Derived outputs move through claim states:

text
draft
observed
reviewed
confirmed
promoted
retired

Only K11 command-related items can become live commands, and only after the K11 promotion process. Visual, SAN, NKo, DEMON, MRT2, room, and semantic outputs are not command gates.

Definition Of Done

This derived-lane contract is satisfied when a strict-real session can produce:

text
template candidates with evidence windows
SAN training index with review status
NKo/MnKO body inscription rows
ENTHEA/Unity visual mapping request
MRT2-style audio mapping request
DEMON research-manifest request
semantic candidate notes
room reconstruction outputs if available
Mac5 reconstruction-learning audit if heavy reconstruction was allowed

Every derived artifact must point back to source windows and preserve whether it is draft, observed, reviewed, confirmed, or promoted.

Promotion Decision

Attach run IDs, datasets, metrics, and reproduction commands.

Source Anchor

MotionMix/lume-wiring/lume-derived-lanes-after-capture-spec-current.md

Detected Structure

Evaluation · References