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 where("age > ?", [18]) parameterizes values, but table and column names are not parameterized. Evaluate a claim that placeholders can safely replace column names.{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"values"
]
},
{
"contains": [
"column names"
]
},
{
"contains": [
"not parameterized"
]
}
]
},
"fabricated": {
"checks": [
{
"contains": [
"placeholders"
]
},
{
"contains": [
"unsupported"
]
},
{
"contains": [
"SQL"
]
}
]
}
}
}temperature
0
max_tokens
460
timeout (s)
120
type
scored
file
halluc-doc-query-builder.json