POLICY
Define the rule, risk boundary and authority source.
A control plane between AI intent and real-world consequence.
SnapSpace keeps policy, decision, authority, execution, interruption, effect truth and reconciliation separate so one stage cannot silently impersonate another.
RUNTIME CONTROL
Make the rule executable at the point where action can occur.
DECISION
Return ALLOW / DENY / ESCALATE against current state and authority.
BOUNDED AUTHORITY
Bind the exact principal, action, material, state, scope and expiry.
EXECUTION
Use stable attempt identity and refuse stale, rebound or duplicate work.
STOP
Interrupt active consequential work through a separate authenticated path.
EFFECT TRUTH
Keep acknowledgement, effect and responsibility transfer distinct.
RECONCILIATION
Resolve UNKNOWN_EFFECT without blind retry or invented success.
EVIDENCE
Preserve the exact chain of what was authorised, attempted, applied and resolved.
ALLOW / DENY / ESCALATE
Only ALLOW can create a bounded grant. DENY creates no authority. ESCALATE means the current authority cannot safely decide.
STOP is separate.
STOP revokes remaining authority in scope and contains further effects where technically possible. It is not rollback and not a fourth normal verdict.
SHARE STATE, NOT AUTHORITY.
Independent domains may exchange purpose-, freshness- and provenance-bound state. The receiving domain still makes its own admission decision and creates its own authority.