mssense-eval-benchmark / schema /evaluation_sample.v1_1.schema.json
Novylab's picture
Upload 13 files
b6a6d0d verified
Raw
History Blame Contribute Delete
4.27 kB
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"$id": "https://intentia.io/schemas/mssense/evaluation_sample.v1_1.schema.json",
"title": "mssense Evaluation Sample v1.1",
"description": "Schema for one row of mssense_eval_benchmark_v1_1.jsonl. V5.15.h delta vs v1: business_rules.items accepts string or object; record_hash + template_hash + modality_role added as required; augmentation_type/lineage_parent_id added as optional for paraphrastic augmentations.",
"type": "object",
"required": [
"sample_id",
"benchmark_version",
"evaluation_only",
"task_family",
"channel_family",
"input_modality",
"modality_role",
"user_intent",
"lat_candidate",
"expected_decision",
"oracle",
"template_hash",
"record_hash"
],
"properties": {
"sample_id": {"type": "string"},
"benchmark_version": {"type": "string"},
"evaluation_only": {"const": true},
"task_family": {"type": "string"},
"channel_family": {"type": "string"},
"input_modality": {
"type": "string",
"enum": ["text", "audio", "demo", "capture", "mixed"]
},
"modality_role": {
"type": "string",
"enum": ["real_artefact", "meta_tagged"],
"description": "real_artefact = input_payload carries an audio/capture/image/recording reference for the declared modality; meta_tagged = the modality label is meta-information (e.g. 'the user typed they want to record audio') but the payload is text-only. Introduced in V5.15.h to remove the ambiguity surfaced by the v1.0 audit."
},
"user_intent": {"type": "string"},
"input_payload": {"type": "object"},
"lat_candidate": {"type": "array"},
"business_rules": {
"type": "array",
"items": {
"oneOf": [
{
"type": "object",
"properties": {
"id": {"type": "string"},
"type": {"type": "string"},
"description": {"type": "string"},
"lat_step_ids": {"type": "array", "items": {"type": "string"}},
"iris_control_type": {"type": ["string", "null"]}
}
},
{
"type": "string",
"description": "DEPRECATED — raw natural-language business rule. Will be wrapped into the canonical object shape by the V5.15.b adapter if encountered."
}
]
}
},
"expected_decision": {
"type": "string",
"enum": ["INVALID", "VALID", "UNDERSTOOD", "QUESTION", "THINKING", "DONE"]
},
"expected_issue_types": {"type": "array"},
"expected_question": {"type": ["string", "null"]},
"expected_choices": {"type": "array"},
"difficulty": {"type": "string", "enum": ["easy", "medium", "hard"]},
"oracle": {"type": "object"},
"template_hash": {
"type": "string",
"description": "SHA-256 of the underlying TEMPLATE — collisions are expected across paraphrastic variants of the same case template. Renamed from content_hash in V5.15.h to match the observed semantics."
},
"record_hash": {
"type": "string",
"description": "SHA-256 of the canonical sorted-key serialisation of THIS record (excluding record_hash, template_hash, content_hash). Use this for record-level uniqueness checks, not template_hash."
},
"content_hash": {
"type": "string",
"description": "DEPRECATED alias of template_hash kept for tooling backward compatibility. Will be removed in v2.0."
},
"_legacy_fields": {
"type": "array",
"items": {"type": "string"},
"description": "List of field names retained from v1.0 for backward compatibility but deprecated."
},
"augmentation_type": {
"type": "string",
"enum": ["paraphrase", "domain_swap", "channel_swap", "modality_swap"],
"description": "Present only on derived (non-seed) samples. Allows downstream benchmarks to filter for seed-only evaluation or to weight metrics."
},
"lineage_parent_id": {
"type": "string",
"description": "Sample_id of the seed this augmented record was derived from. Present only on augmented samples."
},
"augmentation_index": {"type": "integer"},
"augmentation_method": {"type": "string"}
},
"additionalProperties": true
}