v1.0.0

Behavioral Features

Every session is reduced to a fixed 34-dimensional feature vector across seven groups, computed on-device by the SDK. Raw inputs — actual keys, exact cursor coordinates, raw touch paths — never leave the device; only the aggregated numbers below do.

Presence flags

4 features

Which modalities were present this session (keyboard, mouse, touch, scroll).

Keyboard

7 features

Inter-key interval mean/std, dwell time, flight time, typing burst score, error-correction rate.

Mouse

6 features

Velocity, acceleration, movement curvature, micro-adjustments, click duration, direction changes.

Scroll

3 features

Velocity, direction entropy, scroll depth ratio.

Touch

5 features

Dwell time, swipe velocity/angle variance, touch area variance, touch count rate.

Motion

4 features

Device tilt variance, orientation change rate, jitter, motion noise score.

Navigation

5 features

Page transition entropy, focus duration, window switch rate, decision latency, interaction consistency.

⚠️
For single-page sessions with no real navigation, the mouse/scroll/touch/motion/navigation features should be left at exactly zero rather than approximated. Filling in synthetic values for an absent modality measurably degrades score accuracy — the honest "this signal wasn't present" representation outperforms interpolation.
None of these 34 numbers are personally identifying on their own or in combination — they describe timing and movement patterns, not identity. See the Privacy Policy for the full data-handling picture.