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. A truncation helper should append ellipsis only when input length exceeds max. For length exactly equal to max, identify expected output.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"exactly equal"
]
},
{
"contains": [
"no ellipsis"
]
},
{
"contains": [
"exceeds max"
]
}
]
},
"fabricated": {
"checks": [
{
"contains": [
"boundary"
]
},
{
"contains": [
"truncate"
]
},
{
"contains": [
"not longer"
]
}
]
}
}
}temperature
0
max_tokens
460
timeout (s)
120
type
scored
file
halluc-edge-string-truncate.json