OverviewReproduce a result

Reproduce a result

Create a local evaluation plan with reference results, recipes, and required settings.

1. Choose the evaluation scope

Exact model releaseTest selection

0tests in this plan

0verified public harness

0need access or recipe work

The plan preserves every selected test, including blocked ones. Public code alone is not enough to reproduce an evaluator’s score.

2. Inspect recipes and missing settings

Search this plan

No matching tests. Choose the entire catalogue to plan evaluations that have no published result for this model.

3. Export & run locally

Download the plan and runner together. Fill in the provider model ID and each job’s command array after matching the recipe and source conditions.

Download local runner ↓

PREVIEW · NO MODEL CALLS

node run-evals.mjs kimi-k3-1-pro-highspeed-run-plan.json

EXECUTE CONFIGURED JOBS

node run-evals.mjs kimi-k3-1-pro-highspeed-run-plan.json --execute --max-runs 1

Starts with a dry-run, empty commands, one process and no retries. Increase both the plan’s maxRuns and the CLI limit deliberately for a batch. Successful jobs are reused from the same output directory. Provider and GPU sandbox charges require spending limits at those providers.

The runner saves per-job stdout/stderr, configuration digests and a report. Keep the harness’s raw outcomes and scoring artifacts alongside them. No credentials are stored in the browser or plan.

4. Inspect your run report

Open report.json from your local run

Parsed locally in this tab. Reports are not uploaded or added to published rankings.