Test fixture
Grounded answers under adversarial or missing-context prompts.
The model receives the prompt (and optional system message). The run uses scorer rubric_json_metrics with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.
Return JSON only with keys answer, supported_facts, uncertainty. Docs say operations using the transaction client participate in the transaction; operations using the global client do not. Identify the supported conclusion.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"transaction client"
]
},
{
"contains": [
"global client"
]
},
{
"contains": [
"do not"
]
}
]
},
"fabricated": {
"checks": [
{
"contains": [
"participate"
]
},
{
"contains": [
"scope"
]
},
{
"contains": [
"supported"
]
}
]
}
}
}temperature
0
max_tokens
460
timeout (s)
120
type
scored
file
halluc-doc-transaction-scope.json