Oct 3, 2026

Optimizations to the node's signature capture brought the batched evaluation rate up to the full capture rate. Formerly we would validate inline while batching was processing. This was better than waiting on the validation, but not as good as capturing while validating the last capture. So I added double-buffering, which creates a second set of signature batching tables. So now we batch validate all captures, and the capture rate against all sigs is ~99.5%. Interestingly this didn't reduce the wall time for the full validation run, it was almost identical to the batch leakage of around 8%. But it eliminated all of the CPU time spent validating that 8% of all signatures. On a lower CPU machine that becomes wall time.