{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-16-001",
    "source_story_id": "tmp-lead-contract-grade-gpu-kernel-verifier",
    "edition_id": "mp-2026-08-16-morning-0038",
    "edition_url": "https://themachinepress.com/edition/2026-08-16",
    "position": 1,
    "story_type": "lead",
    "section": "chips-infrastructure",
    "editorial_classification": "editorial",
    "headline": "The Loose Test Passed. The Contract Found 62 Percent Broken",
    "slug": "the-loose-test-passed-the-contract-found-62-percent-broken",
    "dek": "Twelve adversarial gates found at least one violation in 62.1 percent of 2,638 machine-generated GPU kernels that a public harness had accepted.",
    "summary": "Twelve adversarial gates found at least one violation in 62.1 percent of 2,638 machine-generated GPU kernels that a public harness had accepted.",
    "body_text": "A loose kernel test can compare a few random outputs at one shape and still miss nondeterminism, shape failures, NaN and infinity handling, or low-precision accumulation. The authors built a twelve-gate verifier, including several tolerance-free checks, and applied it to 2,638 machine-generated kernels already accepted by a public system. Their audit found 39.5 percent broken beyond any tolerance dispute and 62.1 percent carrying at least one violation; the standard test accepted 1,487 kernels that the stricter verifier rejected.\n\nThe paper backs the audit with positive controls, a tolerance sweep, comparison to the reference benchmark's own correctness code and a stratified hand review. The same verifier also checked the authors' own native Blackwell training-backward kernel for the gated-linear-recurrence family against a double-precision oracle. The result is a preprint audit of one corpus and one contract design, not proof that every generated kernel is unsafe, but it sharply weakens headline correctness rates built on a single permissive test.",
    "why_it_matters": "Twelve adversarial gates found at least one violation in 62.1 percent of 2,638 machine-generated GPU kernels that a public harness had accepted.",
    "limitations": [
      "The result is a preprint audit of one corpus and one contract design, not proof that every generated kernel is unsafe, but it sharply weakens headline correctness rates built on a single permissive test."
    ],
    "importance": 10,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-16-001/the-loose-test-passed-the-contract-found-62-percent-broken",
    "json_url": "https://themachinepress.com/story/mp-2026-08-16-001.json",
    "first_published_at": "2026-08-16T09:00:00.000-04:00",
    "modified_at": "2026-08-16T15:43:36.912-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-16-001-001",
        "text": "Twelve adversarial gates found at least one violation in 62.1 percent of 2,638 machine-generated GPU kernels that a public harness had accepted.",
        "source_ids": [
          "source-2026-08-16-001"
        ],
        "qualification": "The result is a preprint audit of one corpus and one contract design, not proof that every generated kernel is unsafe, but it sharply weakens headline correctness rates built on a single permissive test."
      }
    ],
    "source_ids": [
      "source-2026-08-16-001"
    ],
    "tags": [
      "GPU kernels",
      "verification",
      "code generation",
      "Blackwell"
    ],
    "image_url": "https://themachinepress.com/issues/2026-08-16/lead-contract-grade-gpu-kernel-verifier.png",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-16-001",
      "title": "arXiv preprint 2608.12700",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.12700",
      "canonical_url": "https://arxiv.org/abs/2608.12700",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-12T21:25:56.000-04:00",
      "accessed_at": "2026-08-16T15:41:08.010-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-16-001-001"
      ]
    }
  ],
  "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 Loose Test Passed. The Contract Found 62 Percent Broken",
    "publisher": "The Machine Press",
    "published_at": "2026-08-16T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-16-001/the-loose-test-passed-the-contract-found-62-percent-broken"
  }
}