nusantara-validator / claim-schema.json
ariffazil's picture
Upload claim-schema.json with huggingface_hub
7050912 verified
Raw History Blame Contribute Delete
3.53 kB
{
"@context": "https://schema.org",
"@type": "ClaimDeclaration",
"version": "1.0.0",
"description": "Generator self-declaration before BBB Nusantara Validator audit. Required for targeted evaluation. Self-assessment without verifiable evidence = rejected input.",
"schema": {
"generator": {
"type": "string",
"required": true,
"description": "Model name (e.g., 'ilmu-nemo-nano', 'sea-lion-v3-70b')"
},
"generator_version": {
"type": "string",
"required": false,
"description": "Model version or checkpoint"
},
"provider": {
"type": "string",
"required": true,
"description": "Organization or entity that built/hosted the model"
},
"claim_type": {
"type": "string",
"enum": ["nusantara-alignment", "multilingual-general", "malay-specific", "sovereign-ai"],
"required": true,
"description": "What kind of alignment the generator claims"
},
"corpus_languages": {
"type": "array",
"items": "string",
"required": true,
"description": "ISO 639-1 codes of languages in training corpus (e.g., ['ms', 'en', 'id', 'ta', 'zh'])"
},
"constitutional_compliance": {
"type": "array",
"items": "string",
"enum": ["F1", "F2", "F4", "F7", "F9", "F10", "F13"],
"required": true,
"description": "Which arifOS constitutional floors the generator claims to comply with"
},
"self_assessment_nai": {
"type": "number",
"minimum": 0,
"maximum": 10,
"required": true,
"description": "Generator's self-assessed Nusantara Alignment Index (0-10)"
},
"evidence": {
"type": "array",
"required": true,
"description": "Verifiable evidence per claimed floor. Self-assessment alone = rejected.",
"items": {
"floor": "string",
"claim": "string",
"evidence_type": "string (enum: benchmark_result, test_case, audit_receipt, public_report, third_party_eval)",
"evidence_url": "string (URL to verifiable evidence)",
"evidence_hash": "string (sha256 of evidence document, optional)"
}
},
"tokenizer_info": {
"type": "object",
"required": false,
"description": "Tokenizer metadata for parity audit",
"properties": {
"tokenizer_type": "string (e.g., 'bpe', 'unigram', 'sentencepiece')",
"vocab_size": "number",
"bm_tokens_per_word_avg": "number (average tokens per BM word)",
"en_tokens_per_word_avg": "number (average tokens per EN word)",
"parity_ratio": "number (bm_avg / en_avg — >1.5 = architectural bias)"
}
},
"register_coverage": {
"type": "array",
"required": false,
"description": "Which Nusantara registers the generator claims competency in",
"items": {
"register": "string (e.g., 'formal_bm', 'penang_loghat', 'kelas_bm', 'jawi_bm', 'pantun', 'hikayat')",
"evidence_url": "string"
}
},
"guardrail_approach": {
"type": "string",
"enum": ["safety-by-refusal", "safety-by-engagement", "safety-by-context", "hybrid"],
"required": false,
"description": "How the generator handles sensitive Nusantara topics (3R)"
},
"submitted_at": {
"type": "string",
"format": "ISO-8601",
"required": true
},
"validator_attestation_id": {
"type": "string",
"required": false,
"description": "Filled by validator after audit. Null at submission."
}
}
}