Submit a run

Public submissions are not open yet. New rows are added only from validator-passed runner artifacts.

Current rule

A leaderboard row is accepted only when the runner has produced a complete run root with the frozen prompt, scenario manifest, variant config, five trials per scenario, cell summaries, validator report, aggregate summary, and leaderboard row. The website does not score submissions itself; it renders JSON generated from validated run roots.

RUN_PLAN.json
PROMPT.txt
SCENARIO_MANIFEST.json
VARIANT_CONFIGS.json
SCORING_RULE.json
<variant>/<scenario>/trial-1..5.json
<variant>/<scenario>/cell.json
<variant>/summary.json
<variant>/cells.json
run-state.json
validator-report.json
aggregate-summary.json
leaderboard-rows.json

What happens later

After the paper/release gate, outside submissions can come through GitHub pull request against the public runner. The maintainer check will require the same artifact shape, a passing validator report, and reproducibility on sampled scenarios.