LUME Chain 3 — Closed-Loop Integration: Body → iPhone (Echelon) → Audio Bridge → Mac4 LUMF → Sky Garden Visuals

LUME Chain 3 — Closed-Loop Integration: Body → iPhone (Echelon) → Audio Bridge → Mac4 LUMF → Sky Garden Visuals

Status: RELEASED (plan + scripts ready, live verification pending)
Subject: Performer body conducts BOTH the audio AND the visuals simultaneously. Chain 3 closes the loop between Chain 1 (released, body→audio) and Chain 2 (released, body-aware visuals) by routing iPhone-generated Echelon audio onto Mac4's LUMF publisher so the existing visual-side `LumeAudioFftReceiver` consumes it.
Started: 2026-05-08
Released: 2026-05-08
Chain owner: Mohamed
Execution model: META:OMEGA + META:HYDRA collapsed (single pass), 8-lens reviewed
Prerequisites:
- Chain 1 (Echelon Layer 4 + 128D Temporal Closure) released same day, see `docs/chains/RELEASE-CHAIN-1.md`. Live verification of Chain 1 is REQUIRED before Chain 3 setup scripts are run.
- Chain 2 (Sky Garden Duncan-parity push) released same day, see `docs/chains/RELEASE-CHAIN-2.md`. Live verification of Chain 2 is REQUIRED before Chain 3 setup scripts are run.

---

What this release contains

A VALIDATED PLAN + READY-TO-RUN SETUP SCRIPTS for closing the body→music→visuals loop on the LUME rig. It does NOT contain a live-verified integrated system. Live verification is gated on:
1. Chain 1 patches APPLIED + verified (iPhone audio measurably tracks body skeleton)
2. Chain 2 patches APPLIED + verified (Sky Garden reads as Duncan-family at 1080p60)
3. Chain 3 setup scripts run on Mac4 (audio bridge installed + routed + verified)
4. 60-second demo cut MP4 captured per the runbook

When all 4 are green, Chain 3 transitions from RELEASED to FULLY SHIPPED.

This chain is mostly plumbing — config, setup scripts, and a small handful of integration patches. The novelty is at the SYSTEM level (body conducting both audio and visuals) rather than at the bridge level.

---

Audio bridge selection

After 8-lens review (`Desktop/omega-output/closed-loop-integration-20260508/03-review.md`):

Selected: Path 3 — USB audio interface. iPhone connects via Lightning-to-USB to a class-compliant USB audio interface (Apogee Jam, Behringer UMC22, Focusrite Scarlett); Mac4 reads the interface input via the existing `audio_pub.py --mic` path. Latency 10-20ms, hardware $80-200, no new code.

Fallback: Path 1 — AirPlay receive + BlackHole 2ch on Mac4. Zero hardware cost; iPhone AirPlays to Mac4, Mac4 system output is a Multi-Output Aggregate Device combining BlackHole 2ch + speakers, audio_pub.py reads BlackHole 2ch input. Latency 30-60ms.

Disqualified: Path 4 — direct UDP audio from iPhone. Violates Chain 2 acceptance #3 (no new endpoints) and creates a duplicate iOS-side audio path. The setup scripts ship support for both Path 3 (default) and Path 1 (fallback) via a `--mode` flag.

---

Acceptance criteria — verification status

Honest summary: All 5 criteria are BLOCKED on live capture, NOT on chain artifacts. The plan + setup scripts are complete. What is NOT done is the actual recording. That requires Chains 1 + 2 to be live-verified first, then setup scripts run on Mac4, then the body cue performance.

---

Honest gap statement

This chain produced VALIDATED PLANS + READY-TO-RUN SETUP SCRIPTS + READY-TO-APPLY PATCHES, NOT a live-verified integrated body→music→visuals system.

What IS done:
- 8-lens review applied; 0 CRITICAL findings; 3 HIGH (Path 4 coherence, BlackHole stuck-route, runaway-loop oscillation) all resolved in path selection or accompanying scripts/runbook
- Audio bridge path selected with rationale (Path 3 USB; Path 1 AirPlay fallback) and a clean disqualification of Path 4
- 3 idempotent setup scripts (install, route, verify) for both modes
- 2 small integration patches (audio_pub.py DC high-pass + iPhoneSourceMode tag; LumeAudioFftReceiver iPhoneSourceMode bool toggle)
- 1 closed-loop smoke test that measures Echelon tempo lock-in time
- 1 performance runbook with body cue script, OBS scene config, post-flight verification, failure-recovery playbook
- All entanglement findings between Chain 3 and Chain 2's visual layers documented (F-7: iPhone-source audio under-reads at default thresholds)
- All paradox-fuel risks captured and choreographically mitigated (F-3: held-breath bar every 16 to break runaway loop)

What is NOT done:
- Setup scripts NOT run on Mac4 (no `brew install`, no AppleScript automation triggered)
- Patches NOT applied to source trees (read-only on `audio_pub.py` and `LumeAudioFftReceiver.cs`)
- Closed-loop smoke test NOT executed
- 60s demo cut NOT recorded
- No spectrogram diff baseline-vs-active produced
- Chain 1 + Chain 2 patches not yet applied as of 2026-05-08, so Chain 3 cannot be live-verified TODAY anyway

The execution gate (Chain 1 live + Chain 2 live + bridge active + recording captured) is human-in-loop work for Mohamed + Codex over multiple sessions.

---

Artifact pointers (all absolute paths)

### 8-lens review
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/03-review.md`

### Setup scripts (run in order on Mac4 AFTER Chains 1+2 live-verified)
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/patches/setup/01-audio-bridge-install.sh` — installs BlackHole/switchaudio per `--mode`, checks USB interface presence, idempotent
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/patches/setup/02-audio-bridge-route.sh` — sets system input/output per mode, builds Multi-Output Aggregate Device for `--mode airplay`
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/patches/setup/03-audio-bridge-verify.sh` — measures datagram rate + RMS on a shadow port, asserts ≥30Hz and RMS≥0.02

### Integration patches (apply with `git apply` from `lume-commerce` repo root)
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/patches/audio-pub-extension.patch` — adds `--dc-cut` and `--iphone-mode` flags to `services/audio-pub/audio_pub.py`
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/patches/lume-audio-receiver-iphone-mode.patch` — adds Inspector-toggleable `iPhoneSourceMode` to `viz/lume-pcloud/Assets/Scripts/LumeAudioFftReceiver.cs`

### Smoke test + runbook
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/patches/integration-test/closed-loop-smoke.sh` — body cue tempo-lock test, asserts lock within 6s
- `[home]/Desktop/omega-output/closed-loop-integration-20260508/patches/runbook/CLOSED-LOOP-PERFORMANCE-RUNBOOK.md` — pre-flight, body cue script, OBS recording protocol, post-flight verification, failure recovery

### Source files affected (no edits applied — patches reference these targets)
- `[home]/Desktop/lume-commerce/services/audio-pub/audio_pub.py`
- `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/LumeAudioFftReceiver.cs`

---

Promotion criteria — when does Chain 3 become "FULLY SHIPPED"

Chain 3 transitions from RELEASED to FULLY SHIPPED when ALL of the following hold:

1. Chain 1 patches applied + live-verified per `RELEASE-CHAIN-1.md` § Promotion criteria (10 patches, mp4 proof of body→audio causation)
2. Chain 2 patches applied + live-verified per `RELEASE-CHAIN-2.md` § Promotion criteria (12 patches, side-by-side mp4 vs Duncan reference at 1080p60)
3. `setup/01-audio-bridge-install.sh` reports `ok` on Mac4 (`--mode usb` or `--mode airplay`)
4. `setup/02-audio-bridge-route.sh` reports `ok` (system input + output set correctly)
5. `setup/03-audio-bridge-verify.sh` reports `PASS` (datagram rate ≥30Hz, RMS ≥0.02)
6. `integration-test/closed-loop-smoke.sh` reports `PASS — tempo locked at <T> s`
7. `Desktop/lume-commerce/docs/chains/closed-loop-demo-cut-001.mp4` exists, 60s ±0.5s, 1080p60, all 5 acceptance criteria from §"Acceptance criteria" verify per runbook §4

When 1-7 are all green, this release doc gets §Acceptance criteria checkmarks updated and `[home-path]` is set.

---

Rollback recipe

If the closed-loop introduces issues at performance time:

1. Instant rollback (no patch revert):
- Untick `LumeAudioFftReceiver.iPhoneSourceMode` in Inspector → reverts thresholds to Wave 4 defaults
- In `audio_pub.py`, drop the `--dc-cut` and `--iphone-mode` flags → bit-preserved old behavior

2. Audio bridge rollback (mode flip):
- Re-run `setup/01-audio-bridge-install.sh --mode airplay` then `setup/02-airplay --mode airplay`
- This swaps from USB interface to AirPlay path without losing any other state

3. Hard rollback (kill the closed loop, keep Chains 1+2):
- `SwitchAudioSource -t input -s 'MacBook Pro Microphone'` (or whichever default mic) → audio_pub.py reads ambient room sound, not iPhone
- Sky Garden visuals continue to react to whatever's audible — same Wave 4 baseline behavior

4. Full revert:
- `git revert` the audio-pub-extension and lume-audio-receiver-iphone-mode commits if applied
- Each patch is independently revertible (no inter-patch fixups)
- Chain 1 + Chain 2 stay applied; only the closed-loop-specific tuning reverts

---

Constraints honored this chain

• NO touch K11 (Sky Garden display target unchanged)
• NO new endpoints on Mac4 LUMF/LUMD/LUMM/LUME (Chain 2 acceptance #3 — Path 4 disqualified for this reason)
• NO modification to Chain 1 or Chain 2 patches (treated as inputs)
• NO patches actually applied (read-only on source repos)
• NO origin push
• Latency budgets respected by Path 3 selection (sums ~80-100ms vs 250ms / 80ms acceptance budgets)
• Total deliverable size: ~32KB review + ~30KB scripts/patches + ~8KB runbook + ~6KB this doc = ~76KB across all output files (within 80-100KB budget)

---

Why this is a release, not a "completed integration"

Chain 3's job was to validate a path, write the scripts, identify the entanglement risks. The actual closed-loop demonstration happens at performance time, on the bar rig, with Mohamed driving his body through the cue script. No amount of LLM execution closes that loop. We delivered the readiness; the body delivers the proof.

---

Sign-off

Chain 3 release artifact written by Claude Opus 4.7 (1M context) at `2026-05-08`. The release flag at `[home-path]` is set on the same date. Capstone document at `docs/chains/FULL-OMEGA-QUEUE-COMPLETE.md` consolidates Chains 1+2+3.