GPU Util — % utilisation
GPU Temp — °C die
Unified — GB of 128 · 8 GB guard
Throughput — tok / second
TTFT — ms · first token
throughput & first-token from the active lane
Active Lane idle no warm brain
OpenRouter $0.00 spend · session
Settings cloud-eval guardrails · cost cap · stall window · live — next run, no restart

Cloud-eval guardrails

The bounds on a metered cloud eval lane (a non-loopback model endpoint) — the fix for an OpenRouter eval that once hung ~2.5 h holding the lane and accruing uncapped spend. A local Spark lane runs unguarded and is unaffected. Edits write a private config file the guardrail reads at dispatch time, so the next cloud eval picks them up live.