{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-28-011",
    "source_story_id": "tmp-story-naive-prompt-optimization",
    "edition_id": "mp-2026-08-28-morning-0050",
    "edition_url": "https://themachinepress.com/edition/2026-08-28",
    "position": 11,
    "story_type": "dispatch",
    "section": "research",
    "editorial_classification": "editorial",
    "headline": "The Prompt Optimizer Walked One Line",
    "slug": "the-prompt-optimizer-walked-one-line",
    "dek": "A single-lineage optimizer revised prompts from rollout feedback without maintaining a complex search population.",
    "summary": "A single-lineage optimizer revised prompts from rollout feedback without maintaining a complex search population.",
    "body_text": "Naive Prompt Optimization matched or outperformed GEPA in the reported evaluations with fewer rollouts, and its advantage increased with stronger teacher models. Optimized prompts also transferred to other student models, especially within a model family. The authors call the results preliminary and note that reinforcement learning performed better on some interactive tasks.",
    "why_it_matters": "A single-lineage optimizer revised prompts from rollout feedback without maintaining a complex search population.",
    "limitations": [],
    "importance": 7,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-28-011/the-prompt-optimizer-walked-one-line",
    "json_url": "https://themachinepress.com/story/mp-2026-08-28-011.json",
    "first_published_at": "2026-08-28T09:00:00.000-04:00",
    "modified_at": "2026-08-28T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-28-011-001",
        "text": "A single-lineage optimizer revised prompts from rollout feedback without maintaining a complex search population.",
        "source_ids": [
          "source-2026-08-28-011"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-08-28-011"
    ],
    "tags": [
      "prompt optimization",
      "agents",
      "evaluation"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-28-011",
      "title": "arXiv preprint 2608.27266",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.27266",
      "canonical_url": "https://arxiv.org/abs/2608.27266",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-27T11:47:58.000-04:00",
      "accessed_at": "2026-08-28T08:15:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-28-011-001"
      ]
    }
  ],
  "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 Prompt Optimizer Walked One Line",
    "publisher": "The Machine Press",
    "published_at": "2026-08-28T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-28-011/the-prompt-optimizer-walked-one-line"
  }
}