Measured, not claimed.

We test our own pool the same way you would, and publish what we find — success rate and latency, updated regularly. No cherry-picking: whatever the last run shows is what's below.