Triton Inference Server: a preemption metric - VERIFIED ABSENT

VERIFIED NEGATIVE. Triton Inference Server does not emit a preemption metric. nv_inference_preemptions is NOT a metric this engine registers - it is recorded here under the conventional name a reader would go looking for so the absence is queryable rather than implied by a gap in a list. Nothing in Triton core schedules sequences, so it cannot observe preemption; and the TRT-LLM backend's custom families (request_metrics, kv_cache_block_metrics, inflight_batcher_metrics, general_metrics, disaggregated_serving_metrics, runtime_memory_metrics) do not include one either. What was searched: grep for preempt across the docs metric tables and the tensorrtllm_backend README's sample exposition output returns no series.. Do not read this record's metric_name as something the engine exports; it is the name that is missing.