|
| 1 | + # Router-cache benchmark fixtures |
| 2 | + |
| 3 | + Checked-in corpus used by router-cache benchmarks so CI and local runs share the same |
| 4 | + inputs (manifest, campaign, schedule, synthetic PowerPlan / observation samples). |
| 5 | + |
| 6 | + ## Layout |
| 7 | + |
| 8 | + - `campaign.json` |
| 9 | +- `cost-model.json` |
| 10 | +- `legacy-offline-campaign.json` |
| 11 | +- `legacy-offline-manifest.json` |
| 12 | +- `manifest.json` |
| 13 | +- `minimal-live-pilot.json` |
| 14 | +- `oracles/README.md` |
| 15 | +- `oracles/ripgrep-crlf-trace.json` |
| 16 | +- `oracles/ripgrep-glob-case-insensitive-locate.json` |
| 17 | +- `oracles/ripgrep-no-ignore-vcs-locate.json` |
| 18 | +- `oracles/ripgrep-null-data-trace.json` |
| 19 | +- `power-plan.json` |
| 20 | +- `schedule.json` |
| 21 | +- `synthetic-calibration-observations.jsonl` |
| 22 | +- `synthetic-final-observations.jsonl` |
| 23 | +- `synthetic-sealed/docs/ripgrep-crlf-trace-calibration.json` |
| 24 | +- `synthetic-sealed/docs/ripgrep-glob-case-insensitive-locate-calibration.json` |
| 25 | +- `synthetic-sealed/docs/ripgrep-no-ignore-vcs-locate-holdout.json` |
| 26 | +- `synthetic-sealed/docs/ripgrep-null-data-trace-holdout.json` |
| 27 | +- `synthetic-sealed/index.json` |
| 28 | + |
| 29 | + ## How these are used |
| 30 | + |
| 31 | + - Treat this tree as **read-only fixtures**: do not commit regenerated large artifacts |
| 32 | + or machine-specific paths. |
| 33 | + - Benchmarks should resolve paths relative to the repository root (or the package |
| 34 | + `CARGO_MANIFEST_DIR`) so clones remain portable. |
| 35 | + - When adding a fixture, keep payloads small, deterministic, and document purpose |
| 36 | + plus which bench target consumes them. |
| 37 | + |
| 38 | + ## Regenerating (if applicable) |
| 39 | + |
| 40 | + Prefer committed golden files over ad-hoc generation. If a generator exists under |
| 41 | + `benchmarks/`, document the exact command here and keep outputs stable across platforms. |
0 commit comments