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 function pushes into options.tags where options defaults to a module-level object. Identify shared default mutation across calls.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"module-level object"
]
},
{
"contains": [
"options.tags"
]
},
{
"contains": [
"across calls"
]
}
]
},
"fabricated": {
"checks": [
{
"contains": [
"shared default"
]
},
{
"contains": [
"mutation"
]
},
{
"contains": [
"push"
]
}
]
}
}
}temperature
0
max_tokens
460
timeout (s)
120
type
scored
file
halluc-bug-shared-defaults.json