Evidence ledger contract
The validator accepts:
{
"artifact_type": "dashboard_documentation",
"query_authorized": false,
"target": {"id": "view-1", "name": "Sales Overview", "kind": "view"},
"sources": [
{"id": "image-1", "type": "view_image", "scope": "default rendered state"}
],
"records": [
{"id": "r1", "subject": "dashboard", "attribute": "title", "value": "Sales Overview", "status": "observed", "sources": ["image-1"]},
{"id": "r2", "subject": "dashboard", "attribute": "audience", "value": "sales managers", "status": "inferred", "sources": ["image-1"], "note": "KPI and regional breakdown emphasis"},
{"id": "r3", "subject": "dashboard", "attribute": "refresh cadence", "value": null, "status": "unknown", "sources": [], "note": "not exposed by available metadata"}
]
}
artifact_type is dashboard_documentation or datasource_dictionary. Target kind is view, workbook, or datasource. Source types are user_provided, view_image, view_metadata, workbook_metadata, datasource_metadata, lineage, or queried_data.
Observed records require at least one source. Inferred records require sources and a note explaining the inference. Unknown records require null value, no sources, and a note. A queried_data source is rejected unless query_authorized is true.
Record IDs and source IDs must be unique. A subject/attribute pair may appear only once; contradictory duplicates are rejected rather than silently merged. Unknown keys, unreferenced sources, and invalid types are rejected.
Output returns the normalized target, sources, records, and counts by evidence status/source type. It validates ledger structure, not whether source content semantically supports a claim; independently check each claim against the actual source. It is an internal drafting aid, not a user-facing citation system.