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. Memory grows with every reconnect. Heap snapshots show many retained socket listeners bound to old sessions. Identify root cause and fix direction.
{
"metrics": {
"accuracy": {
"checks": [
{
"contains": [
"socket listeners"
]
},
{
"contains": [
"old sessions"
]
},
{
"contains": [
"memory"
]
}
]
},
"evidence": {
"checks": [
{
"contains": [
"reconnect"
]
},
{
"contains": [
"heap snapshot"
]
},
{
"contains": [
"retained"
]
}
]
},
"constraint": {
"checks": [
{
"contains": [
"root cause"
]
},
{
"contains": [
"remove listener"
]
},
{
"contains": [
"cleanup"
]
}
]
}
}
}temperature
0
max_tokens
500
timeout (s)
120
type
scored
file
reason-root-memory-listener.json