Reference
Benchmarks and memory
Reproduce performance, heap and load measurements on your own hardware.
pnpm bench:validate
pnpm bench:query
pnpm bench:binary
pnpm bench:binary-compositions
pnpm bench:binary-strings
pnpm bench:lazy
pnpm bench:load
pnpm bench:allMitata results are persisted under bench/results/*.latest.json with timing, heap and GC
statistics. Comparisons that do different work are marked as not comparable and excluded
from win/loss summaries.
Representative local results on Node 22.17.1 and Ryzen 7 5800H:
| Scenario | JIT | Baseline |
|---|---|---|
| Binary adaptive load + query, 1M | 52.67 ms / 28.26 MiB | Zod 454.89 ms / 110.31 MiB |
| Tagged union count, 1M | 0.78 ms | string discriminator 2.64 ms |
| Lazy direct visitor, 800k outputs | 3.81 ms / 760 B | generator 15.21 ms / 7.75 MiB |
Always benchmark the production shape. Cardinality, selectivity, string uniqueness, engine version and cache reuse can change the winning physical plan.