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. Input 03/04/2026 may mean March 4 or April 3 depending on locale. State the ambiguity and avoid choosing a date without format metadata.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"03/04/2026"
]
},
{
"contains": [
"locale"
]
},
{
"contains": [
"ambiguous"
]
}
]
},
"fabricated": {
"checks": [
{
"contains": [
"without format"
]
},
{
"contains": [
"do not choose"
]
},
{
"contains": [
"date"
]
}
]
}
}
}temperature
0
max_tokens
460
timeout (s)
120
type
scored
file
halluc-edge-date-parser.json