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 candidate says topological sort with adjacency lists is O(V log V + E log E) because every edge must be sorted. Correct the complexity claim for standard DFS or Kahn traversal.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"O(V + E)"
]
},
{
"contains": [
"adjacency list"
]
},
{
"contains": [
"topological sort"
]
}
]
},
"fabricated": {
"checks": [
{
"contains": [
"must be sorted"
]
},
{
"contains": [
"O(V log V + E log E)"
]
},
{
"contains": [
"standard traversal"
]
}
]
}
}
}temperature
0
max_tokens
520
timeout (s)
120
type
scored
file
halluc-complexity-topological-sort.json