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 record must be kept at least 30 days, deleted no later than 45 days, and was created on May 1. Find the earliest valid deletion date.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"May 31"
]
},
{
"contains": [
"30 days"
]
},
{
"contains": [
"earliest"
]
}
]
},
"evidence": {
"checks": [
{
"contains": [
"no later than 45"
]
},
{
"contains": [
"created May 1"
]
},
{
"contains": [
"retention"
]
}
]
},
"constraint": {
"checks": [
{
"contains": [
"constraint"
]
},
{
"contains": [
"valid window"
]
},
{
"contains": [
"not May 1"
]
}
]
}
}
}temperature
0
max_tokens
500
timeout (s)
120
type
scored
file
reason-constraint-retention-date.json