It has somewhere to go
Most adaptive systems mirror the state they detect. Prism holds a target, measures the gap, and works to close it over time.
State in. Generated output out.
The Prism Engine reads four dimensions of your state, knows how much to trust each one, and generates sound, haptics and interface to close the gap toward where you want to be.
THE ARCHITECTURE
One beam in. Four dimensions. Three outputs.
Signals from the device converge into the Prism Context Engine, which splits them into a state vector. Everything downstream is driven by that vector — and by how much of it can be trusted.
INTERACTION
MOTION
AMBIENT LIGHT
HEART RATE
PRESENCE
CLOCK
Prism Context Engine
Fuses the available signals into state dimensions — and derives a confidence value for each one independently.
AROUSAL
How activated you are
VALENCE
Which way it's leaning
COGNITIVE LOAD
How much you're carrying
READINESS
What you can take on
VALUE
CONFIDENCE
PRISM STATE VECTOR — INDEPENDENT PER DIMENSION
Target difference
The engine holds a target state and works on the gap, per dimension. Default target is a neutral baseline.
Actuation authority
Each dimension is only allowed to push as hard as its own confidence permits. Low confidence, small hand on the wheel.
PGAE — SOUND
Generative audio, composed continuously rather than played back.
PGHE — HAPTICS
Vibrotactile output on the same state read, for when sound isn't wanted.
PGUIE — INTERFACE
Pacing and presentation that ease off when you're already loaded.
CLOSED LOOP — THE STATE VECTOR IS RE-ESTIMATED CONTINUOUSLY, AND PARAMETERS MOVE RATE-LIMITED SO NOTHING JUMPS
CONFIDENCE, PER DIMENSION
When a signal drops, only its dimension loses authority.
Confidence isn't one number for the whole read. Each dimension carries its own, derived from which signals were actually available. Turn signals off below and watch what happens.
AROUSAL
VALENCE
COGNITIVE LOAD
READINESS
All signals available. Every dimension is at full authority.
Confidence is a weighted fusion of how well each contributing signal was actually covered in the last window. It degrades — it doesn't fall over.
Three things that make it ours.
Uncertainty is a control input
Confidence doesn't just annotate the read — it caps how hard each dimension may act, so a shaky estimate can't produce a strong intervention.
The read is a clean seam
Sensing and generation only meet at the state vector. Swap the surface or the modality, keep the engine — which is why a phone, a room and a cabin share one core.
Runs where it reads
The whole loop executes locally, on the device doing the sensing. No round trip, no dependency on a connection, and no raw signal leaving the room it came from.
What we claim, and what we don't
The architecture is filed. The effects are still being tested, and we say so.