dmbench-examples / MANIFEST.json
teilomillet's picture
Publish 90 reviewed French DMBench v4 examples
291338c verified
Raw History Blame Contribute Delete
4.12 kB
{
"version": "dmbench-examples-v4",
"repository": "teilomillet/dmbench-examples",
"release_scope": "90 structured examples plus three alternative-candidate summaries; full benchmark and remaining paired variants withheld",
"examples": 90,
"scenario_families": 90,
"by_category": {
"conditions": 9,
"action": 9,
"quand-appeler": 9,
"securite": 9,
"routage": 9,
"preuves": 9,
"contraintes": 9,
"mise-a-jour": 9,
"donnees": 9,
"escalade": 9
},
"by_response_type": {
"choice": 30,
"score": 30,
"noul": 30
},
"by_category_and_type": {
"action/choice": 3,
"action/noul": 3,
"action/score": 3,
"conditions/choice": 3,
"conditions/noul": 3,
"conditions/score": 3,
"contraintes/choice": 3,
"contraintes/noul": 3,
"contraintes/score": 3,
"donnees/choice": 3,
"donnees/noul": 3,
"donnees/score": 3,
"escalade/choice": 3,
"escalade/noul": 3,
"escalade/score": 3,
"mise-a-jour/choice": 3,
"mise-a-jour/noul": 3,
"mise-a-jour/score": 3,
"preuves/choice": 3,
"preuves/noul": 3,
"preuves/score": 3,
"quand-appeler/choice": 3,
"quand-appeler/noul": 3,
"quand-appeler/score": 3,
"routage/choice": 3,
"routage/noul": 3,
"routage/score": 3,
"securite/choice": 3,
"securite/noul": 3,
"securite/score": 3
},
"by_length_band": {
"short": 30,
"medium": 30,
"long": 30
},
"source": {
"version": "typed-decisions-fr-v4",
"suite_sha256": "1d3f631c9bc96ae00dc26df6ddeeacfdfb631e342437821d53c381b7b772e3b5",
"manifest_sha256": "77579a769e21d86825ba9bac0c078c1ff778154ac2406794737cb7463155db70",
"quality_receipt_sha256": "84bbd1a7b3d29e44faa1bbc00e99d6f9d3bb1eaeaccabe44ffb73d9aefb74e23",
"exposure": "public-development"
},
"selection": {
"length_measure": "Unicode characters in request_json, including rubric and options",
"default": "Shortest native prompt, longest from another family, then a third family nearest their length midpoint. Source identity breaks ties.",
"curated_medium_categories": [
"action",
"quand-appeler",
"contraintes"
],
"curated_medium_type": "score",
"curated_medium_note": "Three reviewed reranking scenarios are reserved first as middle examples; their short/long selections exclude the reserved family. These three middle cases need not be nearest the exact length midpoint.",
"structured_paired_siblings_exported": false,
"worked_alternative_candidate_summaries": 3
},
"response_origin": "gold-derived reference; not an observed model prediction",
"quality_summary": {
"source_questions": 1200,
"source_families": 600,
"source_panels": 30,
"current_label_blind_endpoint_receipts": 1200,
"current_family_receipts": 600,
"unresolved_current_review_issues": 0,
"selected_native_reference_fixtures": 90,
"review_limit": "LLM authoring checks, not human certification, independent statistical evidence, or benchmark model performance."
},
"build_inputs": {
"exporter_sha256": "c84bb555ec126b1a991dc28a3be3cd58df8e8a16e1e509990e9f23f95aa7d792",
"native_reference_formatter_sha256": "47767a145e6eb0aabf7263b305c8ee41dbe0c1936c74ffd6ee19ba7d3b3b7b31",
"native_scorer_sha256": "7ec6a57804f575ba4c550ebdb2ae3b9fd51dd5b39f104aa823a9f9254652d162",
"dataset_card_sha256": "0d6593fbbccfef07de1f5207d05eba2f0d860ab9717ed1029636bdccc6139615",
"reranking_guide_sha256": "b3ca7ebe321fecf0238139a9b2e436edd44ad226c7ab2f1ad6ce62772df1784c",
"license_sha256": "f07fce42dbcd5e0ee8b2af1dac42f43f91839b0e3bc41a05c7e5436a62389b72"
},
"files": {
"EXAMPLES.md": "cfe90d744a3810534f98972d57ce5af2232798718d2f2e1b2a6360c35499f3cc",
"LICENSE": "f07fce42dbcd5e0ee8b2af1dac42f43f91839b0e3bc41a05c7e5436a62389b72",
"README.md": "0d6593fbbccfef07de1f5207d05eba2f0d860ab9717ed1029636bdccc6139615",
"RERANKING.md": "b3ca7ebe321fecf0238139a9b2e436edd44ad226c7ab2f1ad6ce62772df1784c",
"examples.jsonl": "e67eb16d66b969cb2d8eab22fb3d6b3fde2dbce8d12c2043ba468bead5cecb90"
}
}