Metrics and Scores

There are two different things in this area, with different statuses:

	What it is	Status
SDK headline scores	Open-source score formulas in `biosignal-analytics` that builders can run on their own data	Experimental (source only, not on npm)
The biometric Score on Elata	One cross-app Score per user, built from derived results apps report with consent	Live

From signals to features

Before any score, the SDK computes features: well-defined, reproducible numbers.

EEG window features: band powers (absolute, relative, log), spectral entropy, dominant frequency, alpha peak, Hjorth parameters, and quality flags
Pulse features: heart rate, RMSSD, SDNN, and signal quality
Robust statistics: medians and median absolute deviations instead of means, so one bad sample doesn’t skew the result
Personal baselines: a rolling 30-day median and range for each person, with outlier rejection

Experimental. biosignal-analytics is available as source code and may change before its first npm release.

Score	Describes
Measurement Quality	The recording, not the person. Always available alongside any other score.
Activation	How far physiological arousal moved from the person’s baseline
Recovery	How well arousal came back down after an activation
Focus	Engagement during a task
Readiness	Day-to-day state against the personal baseline
Resilience	How consistently someone recovers over time

These are product interpretations, never measurements, and the SDK says so. Two design rules keep them honest:

Every contributor is visible. Each score exposes its inputs with their weights, standardized values, and quality.
Scores withhold themselves. Instead of guessing, a score returns no value plus a reason when it can’t be computed responsibly:

Reason	Meaning
Inputs missing	The measurement wasn’t made
Insufficient quality	It was made too poorly to read
Insufficient baseline	There’s no usable personal baseline to compare with
No task context	Focus: nothing was being attended to
Insufficient history	Readiness needs 14 qualified days; Resilience needs 21 days plus 6 recovered activations

A missing input never silently becomes a neutral middle value.

The Score is a single number from 0 to 100 that reflects a user’s results across all the apps they use, only if they opt in. How it’s built:

Apps report derived session results, never raw signal. A report says, for example, that the user’s calm level went from 0.42 at baseline to 0.61 during a five-minute session, with its signal quality and confidence.
Improvement relative to the user’s own baseline matters, not absolute values that differ between people.
Low-quality sessions are dropped, and the Score shows as “calibrating” until there are enough good sessions to be stable.
Consent is checked on the server for every report, so an app can’t contribute without it.
Only approved apps can contribute. An app needs a permission granted by Elata before its reports count.