{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-10-010",
    "source_story_id": "tmp-story-omnicapbench-audiovisual-captioning",
    "edition_id": "mp-2026-10-10-morning-0092",
    "edition_url": "https://themachinepress.com/edition/2026-10-10",
    "position": 10,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The Caption Looked Fluent While Identities Drifted",
    "slug": "the-caption-looked-fluent-while-identities-drifted",
    "dek": "OmniCapBench replaces one whole-caption score with atomic checks for entities, shots and audio events.",
    "summary": "OmniCapBench replaces one whole-caption score with atomic checks for entities, shots and audio events.",
    "body_text": "OmniCapBench decomposes audio-visual captions into verifiable units across entity references, visual shots and audio events. Its 786 densely annotated videos support deterministic constraint checks and localized semantic comparisons, exposing where a polished caption loses temporal grounding, swaps identities, misaligns sound and image, or invents details. Testing frontier multimodal models found strong local perception but weaker long-horizon reasoning, especially for identity continuity and cross-modal alignment. The benchmark is intended as a diagnostic map rather than a single leaderboard number. Its judgments still depend partly on model-based semantic comparison, but the atomic structure makes errors easier to locate and audit.",
    "why_it_matters": "OmniCapBench replaces one whole-caption score with atomic checks for entities, shots and audio events.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-10-010/the-caption-looked-fluent-while-identities-drifted",
    "json_url": "https://themachinepress.com/story/mp-2026-10-10-010.json",
    "first_published_at": "2026-10-10T09:00:00.000-04:00",
    "modified_at": "2026-10-10T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-10-010-001",
        "text": "OmniCapBench replaces one whole-caption score with atomic checks for entities, shots and audio events.",
        "source_ids": [
          "source-2026-10-10-010"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-10-010"
    ],
    "tags": [
      "audio-visual captioning",
      "multimodal evaluation",
      "identity drift"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-10-010",
      "title": "arXiv preprint 2610.12458, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12458v1",
      "canonical_url": "https://arxiv.org/abs/2610.12458v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:59:21.000-04:00",
      "accessed_at": "2026-10-10T08:22:13.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-10-010-001"
      ]
    }
  ],
  "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 Caption Looked Fluent While Identities Drifted",
    "publisher": "The Machine Press",
    "published_at": "2026-10-10T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-10-010/the-caption-looked-fluent-while-identities-drifted"
  }
}