Files
Imbad0202 3a5e59405d feat(schema): benchmark_report.schema.json + fillable template
JSON Schema (draft-2020-12) defining required fields for ARS
benchmark reports. Catches 'n=2 authors' trap, self-scoring
disclosure, and forces non-empty caveats.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-15 16:16:42 +08:00

30 lines
765 B
JSON

{
"ars_version": "3.3.5",
"task_definition": {
"description": "FILL IN: one-sentence description of the task",
"task_type": "open-ended",
"outcome_gradable": false
},
"human_baseline": {
"sample_size": 0,
"author_independence": "author-conducted",
"hours_spent": 0,
"recruitment": "FILL IN: how participants were recruited",
"tools_allowed": ["FILL IN: list of tools"]
},
"ars_run": {
"hours_spent": 0,
"cost_usd": 0,
"skills_used": ["academic-pipeline"],
"data_access_level_declared": "raw"
},
"metrics": {
"primary_metric": "FILL IN: metric name",
"primary_metric_value": 0,
"scoring_independence": "authors-scored"
},
"caveats": [
"FILL IN: at least one known limitation"
]
}