Core orchestration technology
A control engine for AI workflows that must recover visibly.
The SALTI-B Engine governs how Layer8 Adaptive explores alternatives, measures condition, records damage, attempts bounded repair, applies acceptance gates, and routes uncertain work to humans.
Six stages, one durable decision record
Grounding establishes the working evidence. Exploration produces candidates. Validation tests critical channels. Repair addresses reason-coded failures within limits. Calibration estimates operational confidence. Approval releases or escalates the result.
Weakest-link gates protect critical channels
Averages can hide a decisive failure. SALTI-B can evaluate critical validation channels with minimum or weakest-link logic, preventing strong performance elsewhere from compensating for a failed safety, evidence, or constraint channel.
Recovery is measured—not assumed
The Beaver Engineering Amplified Vector (BEAV) compares repair and capture rates with loss and decay. It is a design metric that requires domain calibration, not a certified engineering result or a substitute for accountable review.
Temperature is control state—not truth
The SALTI (Smith Agent Loop Temperature Indicator) expresses closed-loop controller behavior such as exploration pressure within the SALTI-B Engine. It is not a physical measurement, calibrated probability, or correctness guarantee. The interface and API must preserve those semantic boundaries.
Frequently asked questions
What does the SALTI-B Engine do?
It controls durable AI workflow stages, including grounding, exploration, validation, bounded repair, confidence handling, and human approval.
Can SALTI-B work with different AI providers?
Yes. Provider execution is handled through Layer8, allowing SALTI-B workflow stages to use approved models and specialist modules according to policy.