Let the signal
take shape.

The signal takes the screen. The controls wait nearby.

Work opens as a slow spiral galaxy; Rest uses a diffuse nebula. Both share the same visual scale: lower estimates are smaller and cooler aqua, middle values are lavender, and higher estimates spread farther with brighter coral/peach light. The visual key explains the mapping; the dotted orbit marks the zero-reference origin.

The initial source is Ambient: decorative movement with no device connected. Choose Try a demo to see explicitly synthetic variation. Choose Connect for real wearable input. No timer starts automatically.

Your personal estimate

The experimental inferred-arousal estimate now uses the original multivariate architecture: heart rate, pulse RMSSD, actual pulse SDNN, EEG beta/alpha, supplied breathing, and covariance distance, with bounded temporal modifiers. Let a fixed two-minute personal reference form. Zero is its origin on the one-sided 0–100 scale.

The available channels and their weights freeze with the reference. The current Muse/Hume bridge can supply four physiological channels; breathing remains unavailable. A missing required channel holds the estimate. Newly available channels require a new reference. HR-only use is explicitly limited. Weights and thresholds are experimental.

Open Reference & signals to inspect the weighted sum, modifiers, reference, contributions, prediction surprise, covariance, entropy, autocorrelation, signed lags, PCA, and onset events. You can save/open a reference or export the feature session for exact replay in Signal Lab.

An unavailable estimate shows a dash, not zero. The estimate sets the cloud's extent, brightness, warmth, and star-arm spread; raw EEG and optical variation add only minor filament texture. A low value means little response under this model, not proof of calmness.

This is an inference from physiological patterns, not direct sympathetic activity, a mental-state diagnosis, or a validated stress, attention, meditation, or health score. A change does not identify its cause. Formula, reference, and uncertainty

Muse EEG and optical variation add minor filament texture. Raw motion does not move the score cloud, and raw HR does not independently set its size or pace. The cloud encodes the personal estimate; its appearance is not extra evidence or another physiological score.

The bridge can extract eligible pulse-interval RMSSD and actual SDNN from Muse optical data. This is PRV—pulse rate variability—not ECG-derived HRV. Motion, contact, breathing, and peak detection can affect it. Hume's reported rate does not supply individual pulse intervals or RMSSD.

When Hume is absent, accepted Muse pulse intervals can provide rate after two optical channels agree. Rate and PRV then share one sensor; they are not independent confirmations. A changed rate source rebuilds the personal reference.

Open Connect to choose a device. A nearby-device list appears when scanning through the Mac bridge. Select the actual wearable; the sheet shows connection status and a small recent-signal preview.

The Mac bridge detects Classic or newer Athena Muse transports. Direct Web Bluetooth on supported desktop Chromium or Chrome on Android is Classic-only; use the Mac for an updated Athena Muse S and for iPhone. Bounded physical tests passed on one Athena headband and one Hume band: 40 seconds of Muse raw streams and 90 seconds of Hume replies after warm-up, with clean disconnects. This does not establish every firmware's compatibility, indefinite operation, or physiological accuracy.

More detail: Device guide · Visual mappings

Apple Watch: separate Health context

You can still review or archive Watch data in Apple's Health app. A web page does not have native HealthKit access or a direct Watch Bluetooth feed here; a live route would need a native companion. HealthKit HRV is SDNN, distinct from RMSSD.

Apple's Health export guide explains Export All Health Data. The canvas does not import that XML; the separate Signal Lab can load it as retrospective context, never a live Watch feed or reference seed. See Watch notes.

1. On the Mac, open the repository's Start-iPhone-Bridge.command in Terminal. The first run installs the bridge dependencies.
2. Keep the Mac awake, with both devices on the same trusted Wi-Fi.
3. In Connect on the Mac, scan the phone QR code with your iPhone camera and open the link in Safari. The printed phone address is a fallback.
4. Use Connect to scan and select your actual Muse or Hume. If a vendor app holds a wearable, disconnect it there when you are ready to switch.

The Mac does the Bluetooth work. The phone receives signals from the local bridge, rather than from GitHub Pages. Your Mac and phone use the same experience, with integrated connection controls.

Full setup and troubleshooting

Waiting, held, or stale estimates remove the score-encoded halo and show a distinct ambient sky. Unavailable input is never shown as a return to the reference. Continuing ambient light is not a continuing measurement. A raw source can remain Live while the estimate is unavailable.

Raw-signal variation is not proof of contact quality or a usable derived feature. Reported heart rate is not a series of measured heartbeat times. A Live badge can coexist with feature warm-up or reference collection. An HR-only estimate is an explicitly limited fixed profile.

Space offers less on-screen copy, gentler movement, a still image, richer or softer color, and full screen where supported. Gentler movement and the system's reduced-motion preference freeze decorative drift while allowing static score updates. Still image freezes canvas updates for fresh values. If the estimate becomes stale, held, or unavailable, its score encoding clears once even while Still image remains on. The number and status continue updating.

Rest adds an optional four-second inhale and six-second exhale cue. Its separate outer ring does not resize the score cloud. It guides timing; it does not measure breathing or promise a response. A lower personal estimate is an observed trend, not proof of recovery. Breathing itself can alter pulse variability.

Work's Elevation cue starts off. If enabled, a gentle message appears after the estimate remains at least 68 for 60 seconds of valid data. Rest shows change since entering that mode and a 60-second trend; interrupted data or a changed source, track, or reference resets the comparison. The same inference drives both modes.

A timer, only if you want one

Timer offers 10, 25, or 50 minutes, or a custom duration from 1 to 360 minutes. Pause, resume, or end it whenever you choose. It follows real elapsed time independently of animation speed. When time is up, the space stays open.

Signal Lab runs the same integrated model as the main visual, following the recording cursor automatically. Open an exported live analysis session to replay its exact features, sources, reference, calculation, and events.

The Lab also retains the earlier three-signal comparison score, separate research lenses, markers, and recording import/export. Apple Health and dated Hume entries remain retrospective context. Model and reference labels distinguish these analyses.

Lab formulas, input contract, and limits

The Lab's Original research lenses have separate research engine definitions for prediction error, multivariate distance, covariance/PCA, entropy, cross-lags, and metrics requiring independently supplied RR intervals.

Raw packets, derived features, personal references, and visual settings stay in memory. There is no account, recording database, automatic browser saving, or cloud upload of readings. Session and reference downloads are explicit local actions. Open a main-page feature export in Signal Lab to replay it; references can be opened on the main page.

Software tests and simulation check code behavior. Bounded physical tests establish transport on the tested devices; they do not validate feature extraction or the inferred-arousal model. The estimate is calculated from eligible cardiac features, not from the canvas's colors.

Methods · Validation · Repository guide