{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-26-018",
    "source_story_id": "tmp-sidebar-cross-language-equivalence",
    "edition_id": "mp-2026-08-26-morning-0048",
    "edition_url": "https://themachinepress.com/edition/2026-08-26",
    "position": 20,
    "story_type": "ticker",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "Harder Programs Looked Equivalent When They Were Not",
    "slug": "harder-programs-looked-equivalent-when-they-were-not",
    "dek": "PolyHuman tests functional equivalence across human-written C++, Java and Python.",
    "summary": "PolyHuman tests functional equivalence across human-written C++, Java and Python.",
    "body_text": "Models increasingly mislabeled non-equivalent programs as equivalent as difficulty rose, and the strongest tested model showed Python sensitivity plus run-to-run instability under identical settings. Manual analysis of 81 systematic disagreements found partial reliance on similarity cues rather than reliable semantic judgment.",
    "why_it_matters": "PolyHuman tests functional equivalence across human-written C++, Java and Python.",
    "limitations": [],
    "importance": 6,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-26-018/harder-programs-looked-equivalent-when-they-were-not",
    "json_url": "https://themachinepress.com/story/mp-2026-08-26-018.json",
    "first_published_at": "2026-08-26T09:00:00.000-04:00",
    "modified_at": "2026-08-26T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-26-018-001",
        "text": "PolyHuman tests functional equivalence across human-written C++, Java and Python.",
        "source_ids": [
          "source-2026-08-26-020"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-08-26-020"
    ],
    "tags": [
      "code models",
      "functional equivalence",
      "multilingual programming"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-26-020",
      "title": "arXiv preprint 2608.23961",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.23961",
      "canonical_url": "https://arxiv.org/abs/2608.23961",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-24T21:41:00.000-04:00",
      "accessed_at": "2026-08-26T08:17:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-26-018-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Harder Programs Looked Equivalent When They Were Not",
    "publisher": "The Machine Press",
    "published_at": "2026-08-26T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-26-018/harder-programs-looked-equivalent-when-they-were-not"
  }
}