{
  "abstract": "Large language model decision support unfolds across turns, so a final context may contain both\n\nexternally supplied facts and the assistant's earlier interpretations. Prior work shows that multi-\n\nturn presentation, prior model outputs, failed attempts, and visible initial choices can alter later\n\nbehavior; conversational path dependence itself is not claimed as new. We present a controlled\n\nthree-phase diagnostic decomposition of same-fact conversational path dependence in\n\nambiguous LLM judgment. Across synthetic interpersonal cases, Phase 1A held the final external\n\nrecord constant while varying fact order and prior assistant commitment. Its largest matched\n\neffect occurred in one apparently ambiguous case: relative to the order-matched control,\n\ncommit-to-A shifted the final category from B to genuine indeterminacy. Phase 1B crossed three\n\nex ante evidence-diagnosticity levels with two intended polarities and estimated prior-\n\ncommitment, order, and salience sensitivity separately. In the tested case family, pooled absolute\n\ncommitment sensitivity was 0.0280, 0.0335, and 0.1150 at low, medium, and high ambiguity;\n\norder and salience did not follow the same ordering. The largest effect was a High-A commit-to-B\n\ncontrast of -0.342 in directional belief, but it was strongly asymmetric, and High-A failed its\n\nneutral-condition polarity check. Phase 1C tested the stronger hypothesis of recursive self-\n\nreinforcement over three no-new-evidence continuation turns. No-new-evidence and self-\n\nsummary trajectories showed zero mean directional drift, zero mean confidence growth, and no\n\nconclusion or recommendation changes. Fresh evidence produced systematic movement,\n\nconfirming that the pipeline could register longitudinal updating when the external record\n\nchanged. Together, the results show immediate prior-commitment sensitivity under some tested\n\nconditions but no recursive amplification under the Phase 1C protocol.",
  "abstract_shown": true,
  "affiliations": [
    "University of Southern California"
  ],
  "article_type": null,
  "arxiv": null,
  "author_total": 1,
  "authors": [
    {
      "affs": [
        0
      ],
      "name": "Jiaying Wang"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/dqm0xybtxk9ns.bib",
  "cited_by": 0,
  "doi": "10.5281/zenodo.23660349",
  "fulltext_url": null,
  "id": "dqm0xybtxk9ns",
  "issns": [],
  "landing_url": "https://zenodo.org/records/23660349",
  "language": "eng",
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [],
  "openalex": null,
  "pdf_url": "https://zenodo.org/api/records/23660349/files/Conversational_Path_Dependence_arXiv_v1.pdf/content",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-26",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/dqm0xybtxk9ns.ris",
  "sources": [
    {
      "first_seen": "2026-11-09T12:20:33Z",
      "native_id": "23660349",
      "source": "zenodo",
      "url": "https://zenodo.org/records/23660349"
    }
  ],
  "subjects": [],
  "title": "Conversational Path Dependence in Ambiguous LLM Judgment",
  "title_text": "Conversational Path Dependence in Ambiguous LLM Judgment",
  "type": "preprint",
  "updated_at": "2026-11-09T12:20:33Z",
  "url": "https://research.draconitic.com/paper/dqm0xybtxk9ns",
  "venue": "Zenodo",
  "versions": []
}