KERN โ€” deterministic benchmark

Proof-cache hash quality, regenerated per commit with python3 tools/bench_json.py --html docs/bench/index.html.

MEASURED
avalanche % (SHA-256 key)50.01 %
target: 45-55 % (ideal 50)
max collision chain7
target: <= 8 links
chi-square uniformity99952
target: ~= 100000 (buckets)
collisions0
target: 0
avalanche % (naive hash, CONTROL)3.06 %
target: expected bad
max collision chain (naive, CONTROL)100000
target: expected bad
WHY THESE METRICS (AND NOT GB/s)

Throughput depends on the machine: a CI chart of GB/s would chart the runner, not the code. These numbers are exact and must stay exact on every commit โ€” a drift is a real defect, not noise. The last two rows are a control: a naive sum-of-bytes hash, which must look bad (and does). Local throughput numbers are in the README, measured on the author's machine.

PROVENANCE
date2026-10-17T17:40:19Z
commit6893063
python3.14.7
machineLinux x86_64
cpuunknown
formulas100000
hashSHA-256 (ProofCache.keyOf)

Back to the landing page ยท repository