Kalyr Trust
Two questionsbefore every read.
Evidence and permission, answered separately, every time. Kalyr reads governed evidence and records proof. Your equipment and installed systems stay in command.
What every output must prove
Evidence first. Then permission.
Two separate answers, kept as two records. Then every output shows its belief, sealed prediction, authority and outcome.
What exists, and where does it live?
A reference to every source and where it lives. Your systems stay the source of truth.
May it be used for this purpose, now?
Your policies answer at the moment of use: this purpose, this recipient, this time.
What is believed.
A reading of a partly observable situation, uncertainty stated, traceable to its evidence.
What was expected.
Sealed before the move, so it can be scored honestly afterwards.
Your team decides.
Who is accountable, what they may see and for what purpose. Resolved before any output.
What reality confirmed.
Scored against the sealed prediction. Your learning rights admit, hold or exclude it.
Built today: records that keep these six apart. Next, in V1: both gates checked automatically, for every source.
Human state
The person acting, held on their terms.
A person's context enters only when it changes the question and permission allows it. Anything richer is research, governed separately.
Task-relevant.
Scoped to the question, estimated from recorded evidence, with the uncertainty stated. Authority is held separately.
Time-bound.
Every estimate about a person carries an expiry matched to its purpose.
Correctable.
The person it describes can inspect and correct it. A correction reaches every output that depended on it.
Purpose-governed.
Each use is bound to a stated purpose and recipient. Your people decide. The prediction is what gets scored.
See the research boundaryThe rights plane
Permission you can change at any time.
The chain governs what Kalyr may say. This plane governs what may be held, by whom, for how long and for what.
These controls are principles, designed into how records are kept. Enforcing them at every use is V1 product work.
One output · complete lineage
Trace every consequential decision.
Select a stage to see what survives into the authorised view.
Open where trust requires it
Inspectable process. Protected intelligence.
Every output can be audited. The methods behind it stay proprietary. Open means an auditable process.
The trust surface
- 01Source and provenance
- 02Belief and stated uncertainty
- 03The prediction, sealed before the move
- 04Purpose, recipient and authority
- 05Expiry, correction and revision
- 06Reason for refusal
The intelligence core
- 01How evidence is represented
- 02How change is modelled
- 03How signals are prepared
- 04When an episode may teach
- 05Model parameters
- 06What past episodes taught
Kalyr Trust
Control, by design.
Two questions before every read. Your evidence stays where it lives, and your team owns the decision.