Runtime Performance

For authoritative base/head comparisons or fixture work, read packages/effect/runtimeperf/README.md. Use pnpm runtimeperf-compare <suite-or-case> for local worktree comparisons and explicit --base/--head only for committed refs. Report the resolved comparison head and measurement configuration.

For exploratory scripts under benchmark/, match nearby Tinybench scripts, keep setup outside the measured callback unless setup is the operation, warm reusable state, close resources, and print the table. Label results exploratory unless the task defines a repeatable comparison protocol.

Focused runtimeperf work is complete when fixture validation and the selected run succeed. Comparison work additionally requires every configured round to succeed, the paired statistical classification to support the conclusion, and resolved refs and settings to be reported. Harness changes require its focused checks. Tinybench work is complete when resources close, the table is reported, and claims remain exploratory unless a repeatable protocol was run.