Skip to main content

HCS-25 (Signal): NANDA Simple Evals (Informative)

PurposeDirect link to Purpose

Run lightweight “baseline correctness” prompts against NANDA A2A endpoints and store results as trust signals, including error classification for upstream billing/limits.

Stored fields (example schema)Direct link to Stored fields (example schema)

Stored in subject.metadata.additional (example keys):

FieldTypeMeaning
nandaSimpleMathScorenumber | nullScore [0,100]
nandaSimpleMathStatusstringStatus token (e.g., correct, wrong, unparseable, timeout, missing, empty, skipped, upstream-error, error)
nandaSimpleMathQuestionIdstringQuestion identifier
nandaSimpleMathResponsestring | nullRaw response (optional)
nandaSimpleMathErrorstring | nullOptional error classification (timeouts, upstream limits, etc.)
nandaSimpleMathUpdatedAtISO timestampRefresh time
nandaSimpleScienceScorenumber | nullScore [0,100]
nandaSimpleScienceStatusstringStatus token
nandaSimpleScienceQuestionIdstringQuestion identifier
nandaSimpleScienceResponsestring | nullRaw response (optional)
nandaSimpleScienceErrorstring | nullOptional error classification
nandaSimpleScienceUpdatedAtISO timestampRefresh time

See ../simple-evals.md for the general evaluation methodology.

Interop note (informative)Direct link to Interop note (informative)

Some implementations normalize NANDA results into the shared a2aSimple* key family so that a single SimpleMath/SimpleScience scoring adapter can be reused across ecosystems. See ./a2a-simple-evals.md for that shared schema.

Production example (Registry Broker; informative)Direct link to Production example (Registry Broker; informative)

  • Endpoint: https://hol.org/registry/api/v1/agents/{uaid}
  • Example UAID: uaid:aid:3RomW1LwBJ7ZM1PWrLCEro9w4YtY9xwGNWMM2i21mANv8BcWpKg4a7zXxcoNMPDJ7B