{
  "abstract": "Medical multimodal reasoning models are typically trained on teacher-generated rationales, discarding the valid reasoning in students' own failed trajectories, while general-domain correction methods offer little guidance for errors in image reading, anatomical context, or differential diagnosis. We propose MedEvo, which uses medical knowledge, namely a tag taxonomy and error-type diagnosis, to navigate both trajectory repair and data selection. For each failed rollout, a gold-blind auditor locates the first consequential error and provides a minimal, answer-free hint; the same student retains the valid prefix and regenerates the erroneous suffix. After each round, MedEvo identifies capability gaps across medical tags on a held-out feedback set and prioritizes related questions for new rollouts and repairs. With supervised fine-tuning (SFT) alone, MedEvo raises Qwen3.5-9B mean accuracy across seven benchmarks from 65.57% to 69.12% (47.40% to 54.50% on the multimodal track of MedXpertQA (MedXpertQA-MM)), surpassing teacher-distillation SFT (66.93%). It also raises Qwen3.5-35B and Qwen3.5-397B mean accuracy from 67.91% to 73.04% and from 71.83% to 74.95%, respectively. Our analyses show that failed trajectories contain reasoning worth preserving, that repair success does not imply learning value, and that independent feedback is key to sustained improvement.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.17690",
  "author_total": 14,
  "authors": [
    {
      "name": "Xiaoliang Wang"
    },
    {
      "name": "Zhuo Chen"
    },
    {
      "name": "Chuanyang Zheng"
    },
    {
      "name": "Quanbin Wang"
    },
    {
      "name": "Jiangwei Lao"
    },
    {
      "name": "Yingao Liu"
    },
    {
      "name": "Jiajia Liu"
    },
    {
      "name": "Jinming Cao"
    },
    {
      "name": "Honglin Lin"
    },
    {
      "name": "Zhihong Lu"
    },
    {
      "name": "Hao Jin"
    },
    {
      "name": "Jian Wang"
    },
    {
      "name": "Jinjie Gu"
    },
    {
      "name": "Lijun Wu"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/5wre6evkpfhz3.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.17690",
  "fulltext_url": null,
  "id": "5wre6evkpfhz3",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.17690",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.AI"
    ],
    [
      "arxiv",
      "cs.CV"
    ],
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "q-bio.QM"
    ],
    [
      "openalex",
      "2718|Artificial Intelligence in Healthcare and Education"
    ]
  ],
  "openalex": "W7221807860",
  "pdf_url": "https://arxiv.org/pdf/2610.17690",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-26",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/5wre6evkpfhz3.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:35:44Z",
      "native_id": "2610.17690",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.17690"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "arxiv:q-bio.QM",
      "subfield": "bioinformatics"
    }
  ],
  "title": "MedEvo: Medical Reasoning Evolution through Knowledge Navigation",
  "title_text": "MedEvo: Medical Reasoning Evolution through Knowledge Navigation",
  "type": "preprint",
  "updated_at": "2026-10-30T12:35:44Z",
  "url": "https://research.draconitic.com/paper/5wre6evkpfhz3",
  "venue": "arXiv",
  "versions": []
}