A query can return quickly because it examined only a small prefix of the source. That is a correctness defect, not a performance success. Establish the requested population and known boundary records before comparing response times. A cache can also make repeated work faster while retaining incomplete or stale data.
The executable tasks repair a connector-specific late-record failure and aggregate a complete local population. The reasoning tasks then ask how to investigate repeated remote calls and make a bounded performance claim. This browser fixture can prove its supplied results, but it cannot measure a real tenant’s latency or every production connector capability. Name those limits beside the evidence.