SIX PRIMITIVES.
EXPLICIT SEMANTICS.
Explicit inputs. Explicit failures.
Inspect illustrative response states. Actual endpoints and schemas must be taken from the canonical deployed API definition.
{
"status": "verified",
"scope": "bounded",
"review_required": false
}What each call must answer.
High-level semantics only. The preview deliberately avoids fabricating request/response wire formats.
claim.check
Claim support and contradiction. A defensible response distinguishes grounded facts, missing evidence, unresolved conflicts and any human-review requirement.
citation.verify
Verify the cited proposition against its source. A defensible response distinguishes grounded facts, missing evidence, unresolved conflicts and any human-review requirement.
agent.action.verify
Pre-action evidence and authority checks. A defensible response distinguishes grounded facts, missing evidence, unresolved conflicts and any human-review requirement.
transaction.verify
Validate proposed transactional scope. A defensible response distinguishes grounded facts, missing evidence, unresolved conflicts and any human-review requirement.
provenance.receipt
Inspect what was checked and its limits. A defensible response distinguishes grounded facts, missing evidence, unresolved conflicts and any human-review requirement.
evidence.monitor
Track defined evidence states and change. A defensible response distinguishes grounded facts, missing evidence, unresolved conflicts and any human-review requirement.
Do not confuse discovery with execution.
A capability appears in public machine-readable metadata.
Machine identity and permission are checked.
A provider successfully performs the bounded task.
A validated result receives an inspectable receipt.