The guards allowed to influence this instance's verdicts, read from the
engine's own datasource only. A guard applies solely within the datasource
it physically lives in, and resourceType/resourceId are plain document
fields any writer in that datasource can set — so a verdict load must be
physically scoped: a guard doc in a watched content dataset that
self-declares the engine's resource id must never reach a verdict. This
keeps the advisory layer consistent (verdicts are UX, not enforcement);
it is not itself a security boundary. Reads through the same filter
definition the reactive adapters subscribe with
(instanceGuardQuery, the one-id form of the set-shaped
instancesGuardQuery), so the stateless and reactive paths agree.
The guards allowed to influence this instance's verdicts, read from the engine's own datasource only. A guard applies solely within the datasource it physically lives in, and
resourceType/resourceIdare plain document fields any writer in that datasource can set — so a verdict load must be physically scoped: a guard doc in a watched content dataset that self-declares the engine's resource id must never reach a verdict. This keeps the advisory layer consistent (verdicts are UX, not enforcement); it is not itself a security boundary. Reads through the same filter definition the reactive adapters subscribe with (instanceGuardQuery, the one-id form of the set-shaped instancesGuardQuery), so the stateless and reactive paths agree.