Validation

Score model

The validator uses a 100-point model with three classes of outcomes:

• Errors: required parts of the spec are missing or invalid
• Warnings: the spec is usable but incomplete
• Passes: required or recommended checks that succeeded

Required failures

These usually block registration:

• /ai endpoint not reachable
• Response is not valid JSON
• Missing aiendpoint
• Missing service.name
• Missing service.description
• Missing or empty capabilities

Recommended fields

These do not block registration, but they lower confidence and score:

• service.category
• auth
• token_hints
• meta.last_updated

For implementation details, see the validator logic in the registry service and the public validator page at /validate.