CultureHub spatial audio — Will’s Codex handoff #

Build a read-only web UI for the current two-machine ACOS rehearsal on CULTUREHUB LA Wi-Fi. Seat 1 is left; seat 2 is right. Both are online and synthesize their own speaker output from the shared spatial score. Microphones must stay closed; do not request microphone access, record audio, or issue device commands.

The rehearsal host provides a read-only bridge at http://192.168.1.235:8787. Start with GET /api/seats or subscribe to /events; no source checkout or credentials are needed to consume the running bridge. Addresses are local to this Wi-Fi.

Connect #

Current rehearsal addresses, September 18, 2026:

Screen label	JSON seat	Name	HTTP base
Seat 1 — left	0	ac-device	http://192.168.1.236
Seat 2 — right	1	ac5	http://192.168.1.237

Both currently run build dusted-mantella-tideline. Only these two machines are in the active bridge and presence feed.

Use explicit IPs. Read each snapshot’s machineName and ip, also shown on its screen. DHCP addresses can change; verify the runtime identity and seat assignment whenever connecting. Preserve seat numbers when a device goes offline; do not renumber the remaining laptops.

curl --fail --max-time 2 http://192.168.1.237/status
curl --fail --max-time 2 http://192.168.1.237/pieces/spatial-rehearsal-status.json

Read endpoint	Meaning
/status	Runtime {name, build, piece, ip, uptime, port}. Check that piece is spatial-rehearsal; uptime is LAN service uptime in seconds.
/pieces/spatial-rehearsal-status.json	Latest rehearsal snapshot, rewritten about every 100 ms while the piece runs.
/pieces/spatial-rehearsal-config.json	{seat, seats, maxSeconds}. Use the live snapshot’s duration for the current run length.
/pieces/spatial-rehearsal.nsscore	Shared score JSON: duration, lanes/events, and spatial control ribbons. Fetch once per score change.
/pieces/spatial-rehearsal.mjs	Deployed piece source, including bundled deterministic world math. Read as text for implementation reference; do not execute fetched source automatically.

These reads need no credentials. Keep this integration to GET requests.

Read the snapshot #

Field	Interpretation
machineName, ip	Device name and current IP, also displayed on its screen.
seat, seats	Original zero-based seat index and registered ensemble size. Currently two seats, indices 0–1.
geometry, seatOrder	Runtime layout and ordered zero-based seat indices. Currently "line" and [0,1]: seat 1 left, seat 2 right. Build the layout from these fields.
connectivity	{stale, seats: [{seat, state}]} from a separate controller presence publisher. States are online, offline, unstable, unknown, or error. Honor stale; this is separate from the bridge’s polling health.
phase	ready, prepared, countdown, playing, finished, or error.
mode	beeps or score. Beep mode gives each laptop the same scheduled pulse.
runId	Preparation identifier for the run; null before preparation. Compare score times only within the same run.
duration, scoreDuration, scoreName	Current run length, full score length (both seconds), and score name. Read these dynamically; beep mode lasts 20 seconds.
error	Empty string normally; display nonempty errors.
command	Last processed command identifier. Useful context; not a sample sequence number.
audioTime	This device’s audio clock, in seconds. Origins differ between devices: never subtract one seat’s audioTime from another’s.
origin	Scheduled score start in that same device’s audio clock; null before scheduling.
scoreTime	Seconds relative to origin; null when unarmed, negative before the scheduled start. Compare between seats during the same playing run, allowing for unequal sample arrival times. A value alone does not prove playback: always inspect phase.
output.left, output.right	Current digital speaker amplitudes. Use these for live meters and screen brightness; these are not calibrated sound-pressure measurements.
glow	Screen brightness, 0–1. Lights only when a routed source has at least 90% power on this seat (seatGain² ≥ 0.9) and output amplitude exceeds 0.002. Use it to mirror the laptop’s focus display.
outputPeak	Largest observed output amplitude since preparation. Historical peak, not a live meter.
sources	Live source positions and this seat’s gains; see Spatial view below.
beepCount	Pulses emitted in the current beep run; resets on preparation.
maxFrameGap	Largest simulation step in audio-clock seconds since preparation. Shows scheduling stalls, not measured acoustic skew.
networkHalfRttMs	Controller’s half-round-trip timing estimate, in milliseconds; may be null before preparation.
timing	Human-readable timing limitation. Current scheduling estimates audio clocks over Wi-Fi; output latency and drift are uncalibrated.
microphone.hot, microphone.recording	Both must be false. Display any unexpected true value; do not try to calibrate using microphones.
battery	{percent, charging, status, minutesLeft}. Negative percentage or minutes means unavailable. minutesLeft is an estimate when discharging.
screen	{width, height} in runtime pixels.

Treat missing or additional fields defensively. This is a rehearsal interface without a versioned schema. The telemetry contains numeric synthesis state, not a microphone recording or an audio stream.

Browser bridge #

Bridge endpoint	Response
http://192.168.1.235:8787/api/seats	Latest {updatedAt, seats: [...]} snapshot.
http://192.168.1.235:8787/events	Server-Sent Events named seats, carrying the same JSON snapshot.
http://192.168.1.235:8787/health	Bridge health. Check seat freshness separately.

Each item in seats is {host, connected, stale, ageMs, receivedAt, data, error}. data is the native snapshot described above; it is null before the first successful read. updatedAt and receivedAt are host-side ISO timestamps; receivedAt is the last valid fetch time. ageMs measures time since the last change in that seat’s audioTime, or is null before any sample. connected describes the latest poll; error is null on success. Failed reads retain the last valid data.

The bridge polls each seat at 5 Hz with at most one request in flight per seat; it marks data stale on a failed read or when the audio clock stops advancing for two seconds. HTTP success alone does not establish that the piece is running. Preserve host associations and show connection failures and stale data separately. Exclude stale data when choosing the current score or layout.

The bridge sends Access-Control-Allow-Origin: *. Save this as index.html, serve its directory with python3 -m http.server 8000 --bind 127.0.0.1, and open http://127.0.0.1:8000 for a minimal live reader:

<!doctype html>
<meta charset="utf-8">
<title>Spatial audio data</title>
<p id="connection" role="status"></p>
<pre id="seats"></pre>
<script type="module">
const connection = document.querySelector('#connection');
const seats = document.querySelector('#seats');
function renderFleet(data) {
  connection.textContent = '';
  seats.textContent = JSON.stringify(data, null, 2);
}
function showBridgeDisconnected() {
  connection.textContent = 'Bridge disconnected; reconnecting. Last data below.';
}
const base = 'http://192.168.1.235:8787';
async function loadSnapshot() {
  const response = await fetch(base + '/api/seats', {
    cache: 'no-store', signal: AbortSignal.timeout(2000),
  });
  if (!response.ok) throw new Error(`Bridge HTTP ${response.status}`);
  return response.json();
}

loadSnapshot().then(renderFleet).catch(showBridgeDisconnected);
const feed = new EventSource(base + '/events');
feed.addEventListener('seats', event => renderFleet(JSON.parse(event.data)));
feed.onerror = showBridgeDisconnected; // EventSource reconnects automatically.
// Call feed.close() when the view is disposed.
</script>

An HTTPS page may block local HTTP through mixed-content or local-network restrictions. If browser access fails, run a Node proxy in your own UI server and consume it from the same origin. Browser permission to access the local network is distinct from microphone access; this UI never needs a microphone.

Direct browser reads from the laptops have an additional limitation: /status sends a CORS header, but /pieces/* currently does not. JSON files also arrive with Content-Type: text/javascript; a server-side client can read the body and parse JSON.

If running your own bridge from a source checkout:

node fedac/native/tools/spatial-data.mjs --bind 127.0.0.1 --port 8787 \
  192.168.1.236 192.168.1.237

For a custom polling implementation, fetch with cache: 'no-store', bounded timeouts, and no overlapping requests per host. Files can be read during a write: keep the last valid JSON and retry on the next tick. Track host receive time and when audioTime last advanced; a stale file can still return HTTP 200. Verify /status and seat assignments on connection and periodically.

Show registered seats with live output, battery, phase, score time, data age, and microphone state; retain offline seats visibly. Use glow for focused-seat brightness and output for meters. Keep stale values visibly stale. Do not label differences between asynchronously received snapshots as measured speaker synchronization.

Spatial view #

The current composition is Octave Climb (octave-climb.nsscore): 112 BPM, 70.97 seconds including tails, seven sources, and 64 cycles. Four successive registers rise by an octave every 16 cycles; the seventh source adds sparse high-note answering phrases. Three melodic attacks occupy the same time as two soft tap pulses, with 60:40 swing. Melody and taps bounce independently; three diminishing echo sources alternate across the laptops and a quiet reflection source adds a diffuse tail. These are synthesized repeats, not microphone input or convolution reverb. Read each source's live position independently. The deployed score remains readable at /pieces/spatial-rehearsal.nsscore; derive duration, sources and layout from fresh data. Earlier Sine Line, Rhythm Bounce and 3-against-2 scores remain available.

Use data.sources directly to draw the world. Each source contains {lane, name, color, position, seatGain, active}. lane is its zero-based score index and color is an RGB array. In line geometry, position.line runs from 0 (leftmost seat) to 1 (rightmost); x spans −1.6 to 1.6 in virtual units, and y/z are zero. angle is a zero placeholder here: do not use it to draw the line. In ring geometry, position.angle is in radians. Coordinates describe the score, not measured room positions.

seatGain is the source’s 0–1 equal-power panning gain for the reporting seat. Line geometry hands off between neighbors in seatOrder without wrapping the last seat to the first. active means that source has a currently scheduled synthesized event on the seat during score playback; its gain may still be zero. Check both fields for participation, and use output.left/right for actual digital output level.

Playback is currently output-only: a controller estimates each local audio clock over Wi-Fi, prepares a future local start on every seat, then starts the ensemble. The web UI observes this system. Its network feed must not become the audio scheduling clock.

Rehearsal changes — September 18, 2026 #

• Established output-only synthesis, coordinated starts, battery displays, and a read-only HTTP/SSE feed; microphones stay closed.

• Explored Monosine and Melodic Orbit, then moved to Sine Line with four active seats while seat 1 was offline.

• Restored seat 1 on a newer build and added seat 6 at .87; expanded the line to 1 → 2 → 3 → 4 → 5 → 6 and Sine Line to 97.92 seconds. Added device names/IPs and focused-seat brightness.

• Seat 5 later went offline after reporting 2% battery during the six-seat rehearsal.

• Moved to CULTUREHUB LA Wi-Fi for a two-machine rehearsal: seat 1 ac-device at 192.168.1.236, seat 2 ac5 at 192.168.1.237, both on dusted-mantella-tideline. The controller is 192.168.1.235; Sine Line now uses the 40.32-second duet, with only these two seats in the bridge and presence feed.

• Current work deploys local pieces and tracks source changes in Git. No OTA release was requested.

• Added Rhythm Bounce and 3-against-2 polyrhythm tests. Per-source line paths allow independent movement; idle sources no longer trigger focused brightness.

• Softened the rhythm test and added Soft Swing Echo with longer attacks/decays, swing and spatial echo tails.

• Added Octave Climb: three octave rises, gentler high registers and a seventh source for little top-line phrases.