Agent Service Benchmark: which Pocket portal service should an agent call, and what is the evidence? Reads the portal catalogue, the portal status snapshot and the official 9-rule acceptance audit, then returns each candidate's audit verdict and failed rules, serving flag, price, declared input and output schemas, portal example freshness, payment rails and how many services compete in the same category. Candidates are ordered by a rule stated in every response, and every input to that order is returned so the caller can re-rank. It invents no quality score and names what it does not measure. No paid call is made on the caller's behalf. Every response is a JSON object. Identity via GET /v1/version, readiness via GET /v1/health, function via GET /v1/categories, GET /v1/catalogue, GET /v1/service/{id} and POST /v1/compare {category|ids|question}.
