Download scoring/protocol.json from argentislabs/argentis-decision-bench: direct link, hf CLI and curl.
- Browser
- Download file 1.34 kB
-
https://huggingface.co/datasets/argentislabs/argentis-decision-bench/resolve/main/scoring/protocol.json
- Command line
-
hf download hf://datasets/argentislabs/argentis-decision-bench/scoring/protocol.json
-
curl -L -o protocol.json https://huggingface.co/datasets/argentislabs/argentis-decision-bench/resolve/main/scoring/protocol.json
1.34 kB
| { | |
| "protocol_sha256": "4a8363c6fc22655385390ca7c04d8489bd0e04dd5c0a2e6380afc9176fa5a0f8", | |
| "fields": { | |
| "dataset_manifest_sha256": "f2d78ed9d1307cf2b9b426386147b90e15887aa016943dc6a0a57b224ca8163b", | |
| "prompt_template": "You are a decision system for airline baggage fees. Apply the rules below to the case and respond with ONLY a JSON object, no other text.\n\nSchema:\n{{\"decision\": {{\"kind\": \"scalar\", \"value\": <number>, \"unit\": \"USD\"}} | null,\n \"status\": \"resolved\" | \"abstain\" | \"indeterminate\",\n \"evidence\": [<evidence IDs you relied on>],\n \"rules_applied\": [<rule names from the vocabulary>]}}\n\nstatus \"resolved\" requires a decision. If a required input is missing, use status \"abstain\", decision null, and cite the missing field in evidence.\ndecision.value is the TOTAL cost in USD: ticket price plus all baggage fees.\nEvidence ID inventory: {evidence_ids}\nRule vocabulary: {rules}\n\n=== RULES ===\n{rules_text}\n\n=== CASE ===\n{case_prompt}", | |
| "decoding": { | |
| "temperature": 0.0, | |
| "max_tokens": 512 | |
| }, | |
| "scorer_version": "s1.1.0", | |
| "transform_set_version": "suite-v1", | |
| "schema_version": "1", | |
| "harness_git_sha": "75e490d4fbd56a95815486b57923f0c1ceae768f" | |
| }, | |
| "note": "sha256 is the canonical-JSON hash of fields; every scored record must carry it" | |
| } | |