Load test
Generate a stream of plausible name searches, feed them through the real API at a pace the fleet can absorb, and watch what comes back — answers, errors, or Google deciding it has seen enough of us.
Outcome over time
Jobs as they finish. Red creeping in from the right is Google catching on.
Pages per identity
Counted in pages, not jobs — each page of a depth walk takes its own identity, and the page is what Google blocks.
Fetch time
Median over the p10–p90 band, from the agent's own per-page timing — not this page's wall clock, which is quantised by the poll interval. A steady median with a fanning spread is a fleet starting to struggle.
Codes
Results
Click a row to read the SERP that came back. Served by fills in as soon as the first page reports — nothing decides the identity at submit, the sidecar's pool picks one when it acquires it, so that is the earliest it can be known. A row still in flight shows its status and, on a multi-page walk, which page it is on.