{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2609.05370v1",
  "slug": "2609-05370v1-03wuqdv",
  "url": "https://feed7.dev/p/2609-05370v1-03wuqdv",
  "title": "When LLM Decompilers Recompile More and Preserve Less",
  "why_included": "LLM decompilers can produce compilable C that passes shipped tests while changing behavior or hiding crashes. Validate reconstructed code against the original with shared fuzzed inputs.",
  "summary": "Decompile-Diverge found that outputs passing every shipped test still differed from original binaries on **4.9% overall**, reaching **13% for one system**. Its oracle generates a driver, grows inputs from the reference, and compares both implementations on the same corpus.",
  "practical_implication": "For security work, treat recompilation and fixture tests as entry checks, not semantic proof. Differential fuzzing should gate agent-generated decompilation before using it for vulnerability analysis or downstream automation.",
  "agent_context": "Decompile-Diverge found that outputs passing every shipped test still differed from original binaries on **4.9% overall**, reaching **13% for one system**. Its oracle generates a driver, grows inputs from the reference, and compares both implementations on the same corpus.\n\nFor security work, treat recompilation and fixture tests as entry checks, not semantic proof. Differential fuzzing should gate agent-generated decompilation before using it for vulnerability analysis or downstream automation.\n\nThe strongest refinement LLM raised Ghidra's build rate from **75% to 90%** while matched behavior fell from **74% to 62%**. Up to one tenth of CVE-grounded functions lost the disclosed crash, but results cover the evaluated corpora and configurations.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2609.05370v1",
    "published_at": "2026-09-04T17:16:18.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "benchmark",
  "domains": [
    "coding",
    "security"
  ],
  "topics": [
    "agent-evals",
    "benchmark-integrity",
    "agent-reliability"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The strongest refinement LLM raised Ghidra's build rate from **75% to 90%** while matched behavior fell from **74% to 62%**. Up to one tenth of CVE-grounded functions lost the disclosed crash, but results cover the evaluated corpora and configurations."
  ],
  "connected_context": {
    "meaning": "This adds semantic equivalence as an independent acceptance gate for decompilation agents: recompilation and shipped tests can improve while fidelity to the binary worsens. It reinforces prior evidence that green functional checks miss consequential constraints, and supplies a reference-based oracle—shared fuzzed inputs and differential execution—suited to behavior drift and lost security-relevant crashes.",
    "corpus_size": 703,
    "generated_at": "2026-09-08T10:04:26.072Z",
    "connections": [
      {
        "title": "SWE-Gate: Passing Functional Tests Is Not Enough for Software Engineering Agents",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2609.04167v1",
        "feed7_url": "https://feed7.dev/p/2609-04167v1-1vvofgw",
        "reason": "Both show that functional test success is an incomplete acceptance criterion; SWE-Gate finds review-constraint violations, while this Signal finds divergence from original binary behavior."
      },
      {
        "title": "Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Prompts and Weights in Frozen Small Code Models",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.12962v1",
        "feed7_url": "https://feed7.dev/p/2607-12962v1-0q4i26c",
        "reason": "The placebo-controlled repair result makes refinement gains suspect without causal validation; here, refinement raises buildability while reducing behavioral preservation."
      },
      {
        "title": "Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2609.04198v1",
        "feed7_url": "https://feed7.dev/p/2609-04198v1-1mype86",
        "reason": "The unstable-judge study supports validating an evaluation instrument before gating releases; this Signal offers a deterministic reference comparison that avoids relying solely on black-box judgment."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-04T17:16:18.000Z",
  "modified_at": "2026-09-04T17:16:18.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2609-05370v1-03wuqdv",
    "json": "https://feed7.dev/p/2609-05370v1-03wuqdv.json",
    "markdown": "https://feed7.dev/p/2609-05370v1-03wuqdv.md"
  }
}