{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-18-011",
    "source_story_id": "tmp-story-traceflow-failure-guidance",
    "edition_id": "mp-2026-09-18-morning-0071",
    "edition_url": "https://themachinepress.com/edition/2026-09-18",
    "position": 11,
    "story_type": "dispatch",
    "section": "robotics",
    "editorial_classification": "editorial",
    "headline": "Failed Rollouts Became a Steering Field",
    "slug": "failed-rollouts-became-a-steering-field",
    "dek": "TraceFlow lifted ordered packing from 21 to 39 successes in 50 trials without changing policy weights.",
    "summary": "TraceFlow lifted ordered packing from 21 to 39 successes in 50 trials without changing policy weights.",
    "body_text": "TraceFlow stores time-ordered robot states and actions with only a terminal success bit, then turns the densities of retrieved successes and failures into a bounded correction for a frozen flow-matching policy. On a physical ordered-packing task, completions rose from 21 of 50 to 39; one stacking round reached 47 and eliminated wrong-sequence episodes in that test. Simulation gains were selective rather than universal, with two task subsets declining slightly and stacking branches peaking before round ten. The result makes failure traces useful without claiming unlimited self-improvement.",
    "why_it_matters": "TraceFlow lifted ordered packing from 21 to 39 successes in 50 trials without changing policy weights.",
    "limitations": [
      "TraceFlow stores time-ordered robot states and actions with only a terminal success bit, then turns the densities of retrieved successes and failures into a bounded correction for a frozen flow-matching policy."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-18-011/failed-rollouts-became-a-steering-field",
    "json_url": "https://themachinepress.com/story/mp-2026-09-18-011.json",
    "first_published_at": "2026-09-18T09:00:00.000-04:00",
    "modified_at": "2026-09-18T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-18-011-001",
        "text": "TraceFlow lifted ordered packing from 21 to 39 successes in 50 trials without changing policy weights.",
        "source_ids": [
          "source-2026-09-18-011"
        ],
        "qualification": "TraceFlow stores time-ordered robot states and actions with only a terminal success bit, then turns the densities of retrieved successes and failures into a bounded correction for a frozen flow-matching policy."
      }
    ],
    "source_ids": [
      "source-2026-09-18-011"
    ],
    "tags": [
      "robot learning",
      "failure traces",
      "flow matching"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-18-011",
      "title": "arXiv preprint 2609.20646",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.20646",
      "canonical_url": "https://arxiv.org/abs/2609.20646",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-17T12:26:28.000-04:00",
      "accessed_at": "2026-09-18T08:26:40.599-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-18-011-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Failed Rollouts Became a Steering Field",
    "publisher": "The Machine Press",
    "published_at": "2026-09-18T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-18-011/failed-rollouts-became-a-steering-field"
  }
}