| 1 | # Runtime Performance |
| 2 | |
| 3 | For authoritative base/head comparisons or fixture work, read |
| 4 | `packages/effect/runtimeperf/README.md`. Use `pnpm runtimeperf-compare |
| 5 | <suite-or-case>` for local worktree comparisons and explicit `--base`/`--head` |
| 6 | only for committed refs. Report the resolved comparison head and measurement |
| 7 | configuration. |
| 8 | |
| 9 | For exploratory scripts under `benchmark/`, match nearby Tinybench scripts, |
| 10 | keep setup outside the measured callback unless setup is the operation, warm |
| 11 | reusable state, close resources, and print the table. Label results exploratory |
| 12 | unless the task defines a repeatable comparison protocol. |
| 13 | |
| 14 | Focused runtimeperf work is complete when fixture validation and the selected |
| 15 | run succeed. Comparison work additionally requires every configured round to |
| 16 | succeed, the paired statistical classification to support the conclusion, and |
| 17 | resolved refs and settings to be reported. Harness changes require its focused |
| 18 | checks. Tinybench work is complete when resources close, the table is reported, |
| 19 | and claims remain exploratory unless a repeatable protocol was run. |