Files
zeekayandHanzo Dev c2484d012e landing: stop publishing benchmarks nobody ran
The performance section led with 'Real benchmarks on Apple M-series silicon'
and then showed two tables that no harness in this repository produces. Two
rows ran backwards: scalar multiplication at 6.6ms for 10 items and 2.3ms for
100, and signing at 25ms with 3 parties against 21ms with 10. Work does not get
cheaper as it grows — a table that says so was not measured, and calling it
real made the claim worse rather than better.

What the section was actually communicating is true and needs no numbers:
signing is O(t) and touches only the threshold signers, so a 10-of-10,000
scheme signs like a 10-of-10, while keygen and reshare are O(n^2) and touch
everyone. That is now the whole section, plus a note that keygen benchmarks
exist under e2e/ and that their results belong here once run on a named machine
and a named build.

This is a public page. It is the first performance claim a reader meets, and it
should be one we can defend.

Co-authored-by: Hanzo Dev <[email protected]>
2026-07-29 10:04:27 -07:00
..