PLATFORM

Runtime Guardrails.
Turn evaluation into control.

A runtime guardrail is useful when it changes what the system is allowed to do, not merely what a dashboard reports.

Decision modes

PASS

Allow

Action meets the configured evidence and policy threshold.

WARN

Proceed with signal

Action can proceed while preserving a warning or remediation.

ESCALATE

Require review

Route to human or higher-authority policy when confidence is insufficient.

BLOCK

Stop

Do not invoke the external tool until the reason is resolved.

Place guardrails at the tool boundary

The strongest runtime location is immediately before a consequential tool call because the trust layer can inspect the exact action and arguments that would reach the outside world.

Bounded claims

Guardrails reduce defined classes of risk; they do not guarantee perfect correctness, compliance or safety. The implemented checks and policy contract should always be inspectable.