autor3search-typescript demo: −92.55% on code written slow on purpose
FIXTURE
not a third-party library
HEADLINE
−92.55%
own demo code, written slow on purpose
before → after
bar length is relative to the before valuecountWords ms/op−92.55%
what changed
A countWords whose hot loop rebuilt the whole array per character. chars.concat([c]) became chars.push(c).
3 experiments
1
kept
1
discarded
1
failed
note — the same live session discarded a no-op commit against the advanced baseline, and failed a scope violation. A big percentage off a demo is easy to earn. It is not a claim about real code.
how it was measured
- tool
vitest bench- rounds
- interleaved rounds per side
- verdict rule
- KEEP needs one significant improvement, no significant regression past max_regress_pct, and a geomean that beats min_effect_pct.
LIMITS
Results hold only for the benchmarks a run measures. Inputs they do not cover can get slower. Check your hot paths before you merge.