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 verdict, corrections, evidence. A note claims stream.finished from Node.js automatically consumes a readable stream and drains all data before resolving. Correct what finished observes versus what it consumes.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"stream.finished"
]
},
{
"contains": [
"observes completion"
]
},
{
"contains": [
"does not consume"
]
}
]
},
"fabricated": {
"checks": [
{
"contains": [
"automatically consumes"
]
},
{
"contains": [
"drains all data"
]
},
{
"contains": [
"false"
]
}
]
}
}
}temperature
0
max_tokens
520
timeout (s)
120
type
scored
file
halluc-doc-node-stream-finished.json