Benchmarks
Every push to cfgaudit's main runs the benchmarks and records the medians of ten runs: 2 commits from 2026-10-28 to 2026-10-28. The input is a corpus generated from a fixed seed (neutral project documentation, skills, rules files, settings and MCP configs), not files from real repositories.
Allocations and bytes do not depend on the machine, so they are shown as they are. Time does, and the hosted runners change hardware, so it is shown only as the ratio to the latest release measured in the same job: below 1.0 is faster. release the runner CPU changed. Each point links to its commit; hover a point for its value.
Scan_Large
Scan_Repo
Run_EdgeCases/DenseLine
Run_EdgeCases/LargeSettings
Run_EdgeCases/LongLine
Run_Fixtures
Run_InstructionContent
Init_Rules
Data: bench.jsonl on the bench-data branch, written by the Benchmark history workflow.