{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-03-014",
    "source_story_id": "tmp-story-assistive-contact-physics-benchmark",
    "edition_id": "mp-2026-09-03-morning-0056",
    "edition_url": "https://themachinepress.com/edition/2026-09-03",
    "position": 14,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "Task Success Fell When Safe Contact Counted",
    "slug": "task-success-fell-when-safe-contact-counted",
    "dek": "A bathing benchmark pairs a deformable simulated person with region and force screening calibrated against a care manikin.",
    "summary": "A bathing benchmark pairs a deformable simulated person with region and force screening calibrated against a care manikin.",
    "body_text": "The protocol freezes a vision-only scorer and adds physics-aware contact measures to ordinary completion, reducing opportunities for evaluator leakage. Across 140 runs per method, the LLM-augmented state machine completed 72.9% of tasks but only 56.4% survived correct-region and force-safety checks; VoxPoser completed 27.9%, and zero-shot pi0.5 completed 0.7%. The benchmark argues that task completion alone is an unsafe proxy for contact-rich assistive behavior.",
    "why_it_matters": "A bathing benchmark pairs a deformable simulated person with region and force screening calibrated against a care manikin.",
    "limitations": [
      "The protocol freezes a vision-only scorer and adds physics-aware contact measures to ordinary completion, reducing opportunities for evaluator leakage.",
      "Across 140 runs per method, the LLM-augmented state machine completed 72.9% of tasks but only 56.4% survived correct-region and force-safety checks; VoxPoser completed 27.9%, and zero-shot pi0.5 completed 0.7%."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-03-014/task-success-fell-when-safe-contact-counted",
    "json_url": "https://themachinepress.com/story/mp-2026-09-03-014.json",
    "first_published_at": "2026-09-03T09:00:00.000-04:00",
    "modified_at": "2026-09-03T14:20:28.218-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-03-014-001",
        "text": "A bathing benchmark pairs a deformable simulated person with region and force screening calibrated against a care manikin.",
        "source_ids": [
          "source-2026-09-03-014"
        ],
        "qualification": "The protocol freezes a vision-only scorer and adds physics-aware contact measures to ordinary completion, reducing opportunities for evaluator leakage."
      }
    ],
    "source_ids": [
      "source-2026-09-03-014"
    ],
    "tags": [
      "assistive robotics",
      "contact safety",
      "benchmark"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-03-014",
      "title": "arXiv preprint 2609.02402",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.02402",
      "canonical_url": "https://arxiv.org/abs/2609.02402",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-02T06:14:32.000-04:00",
      "accessed_at": "2026-09-03T14:20:28.218-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-03-014-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Task Success Fell When Safe Contact Counted",
    "publisher": "The Machine Press",
    "published_at": "2026-09-03T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-03-014/task-success-fell-when-safe-contact-counted"
  }
}