Terms
- pass
- The claimed value and the captured metadata value are canonically equal (reason: metadata_match).
- fail
- The claim disagrees with captured metadata: either the values are not canonically equal, or a capability the capture asserts is absent was claimed present (reason: metadata_disagreement).
- unknown
- There is no captured metadata for the claim, or the metadata value is null. Unmeasured is never reported as fail.
- claim
- A key/value assertion taken from a listing (for example a capability a publisher says their MCP server, API, or SDK provides).
- capability
- A key/value fact derived from captured metadata that a claim is compared against. Each carries a hashed evidence reference.
- adapter
- A pure function that turns a captured manifest into capabilities. Supported adapters: mcp-tools-list-v1 and openapi-operations-v1. No adapter performs a live fetch or edits a listing.
- manifest completeness
- A caller assertion (manifest_complete=true) that a captured snapshot is exhaustive. Only then does an absent, managed capability become an explicit false, so a claim for it can fail. Completeness is never independently verified.
- canonical equality
- Order-independent structural comparison of two JSON values (object keys sorted, arrays compared in order) used to decide pass versus fail.
- evidence reference
- A sha256: digest of the underlying value. Reports export these hashed references and reasons only, never raw listing text, prices, or secrets.
See the methodology for how these terms combine into a result, or return to the overview.