Preview LLM judge
Run the LLM judge once against a sample response (rule-editor sandbox). `binding_id` runs against a specific candidate instead of the active one.
Run the LLM judge once against a sample response (rule-editor sandbox). binding_id runs against a specific candidate instead of the active one.
ApiKeyAuthAuthorizationBearer <token>Bearer token authentication. Send an API key (a saai_api_-prefixed token) as Authorization: Bearer <token>.
Required permission
manage_data_protectionapplication/json- body
instruction*string1 <= length <= 2000sample*stringSample response text to judge; the server enforces the configured judge cap (default 128 KiB).
length <= 1048576binding_id?uuiduuidThe request has succeeded.
application/json- response
spans*array<string>Verbatim spans the judge flagged.
redacted*stringsampleDocument with every span replaced by [REDACTED].
judged*stringThe document the judge actually received: the sample serialized as a tool response is at runtime (a JSON sample is flattened to its values), then shaped for the prompt by dropping blank and repeated lines. Lets the tester show exactly what the model saw.
sample_document*stringThe same serialization without the prompt shaping, so blank and repeated lines survive. Lets the tester render the redaction preview in the shape of the admin's own sample rather than the reflowed prompt.
curl -X POST "https://example.com/dlp/judge/preview" \ -H "Content-Type: application/json" \ -d '{ "instruction": "string", "sample": "string" }'{ "spans": [ "string" ], "redacted": "string", "judged": "string", "sample_document": "string"}