Test fixture
Arithmetic, symbolic steps, and structured problem solving.
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, evidence, constraints. A test passes 90% of the time in CI. The reason for the 10% failures is unknown. What uncertainty exists and what should be investigated to diagnose the flakiness?
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"flaky"
]
},
{
"contains": [
"investigate"
]
},
{
"contains": [
"non-deterministic"
]
}
]
},
"evidence": {
"checks": [
{
"contains": [
"90%"
]
},
{
"contains": [
"unknown"
]
},
{
"contains": [
"CI"
]
}
]
},
"constraint": {
"checks": [
{
"contains": [
"root cause"
]
},
{
"contains": [
"timing"
]
},
{
"contains": [
"environment"
]
}
]
}
}
}temperature
0
max_tokens
500
timeout (s)
120
type
scored
file
reason-unc-network-flakiness.json