{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-23-004",
    "source_story_id": "tmp-story-fleetsieve-llm-profiling",
    "edition_id": "mp-2026-08-23-morning-0045",
    "edition_url": "https://themachinepress.com/edition/2026-08-23",
    "position": 4,
    "story_type": "dispatch",
    "section": "chips-infrastructure",
    "editorial_classification": "editorial",
    "headline": "The Fleet Stopped Profiling When the Decision Stopped Moving",
    "slug": "the-fleet-stopped-profiling-when-the-decision-stopped-moving",
    "dek": "FleetSieve measures only the LLM-serving configurations likely to change a resource-coupled allocation.",
    "summary": "FleetSieve measures only the LLM-serving configurations likely to change a resource-coupled allocation.",
    "body_text": "On a fixed H100 grid for a 31-billion-parameter open model, the method matched the oracle aggregate choice with 6.9 percent fewer GPU-seconds than uniform random profiling in the fixed comparison. Its joint capacity and tail-latency model also avoided a configuration whose 46.4-second p99 violated a 30-second service objective, although the paper reports that FleetSieve did not use the fewest GPU-seconds for every workload.",
    "why_it_matters": "FleetSieve measures only the LLM-serving configurations likely to change a resource-coupled allocation.",
    "limitations": [
      "Its joint capacity and tail-latency model also avoided a configuration whose 46.4-second p99 violated a 30-second service objective, although the paper reports that FleetSieve did not use the fewest GPU-seconds for every workload."
    ],
    "importance": 7,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-23-004/the-fleet-stopped-profiling-when-the-decision-stopped-moving",
    "json_url": "https://themachinepress.com/story/mp-2026-08-23-004.json",
    "first_published_at": "2026-08-23T09:00:00.000-04:00",
    "modified_at": "2026-08-23T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-23-004-001",
        "text": "FleetSieve measures only the LLM-serving configurations likely to change a resource-coupled allocation.",
        "source_ids": [
          "source-2026-08-23-004"
        ],
        "qualification": "Its joint capacity and tail-latency model also avoided a configuration whose 46.4-second p99 violated a 30-second service objective, although the paper reports that FleetSieve did not use the fewest GPU-seconds for every workload."
      }
    ],
    "source_ids": [
      "source-2026-08-23-004"
    ],
    "tags": [
      "LLM serving",
      "GPU fleets",
      "latency"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-23-004",
      "title": "arXiv preprint 2608.19659",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.19659",
      "canonical_url": "https://arxiv.org/abs/2608.19659",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-20T01:54:29.000-04:00",
      "accessed_at": "2026-08-23T08:27:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-23-004-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Fleet Stopped Profiling When the Decision Stopped Moving",
    "publisher": "The Machine Press",
    "published_at": "2026-08-23T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-23-004/the-fleet-stopped-profiling-when-the-decision-stopped-moving"
  }
}