{
  "absUrl": "https://arxiv.org/abs/2611.19840v1",
  "arxivId": "2611.19840",
  "arxivVersion": 1,
  "arxivVersionedId": "2611.19840v1",
  "authors": [
    {
      "affiliations": [
        "Independent Researcher",
        "Code and data: RA-Eval v0.1, Zenodo DOI 10.5281/zenodo.24237533"
      ],
      "name": "Dongwu Qian"
    }
  ],
  "contract": "researcher-sidecars-v1",
  "id": "2611.19840v1",
  "originalTitle": "Grounding Language Models in Speech Does Not Fix \"AI Flavor\": Calibrated LLM Judges, Goodhart Traps, and a Preference-Tuning Fix",
  "pdfUrl": "https://arxiv.org/pdf/2611.19840v1.pdf",
  "schemaVersion": 1,
  "type": "preprint"
}
