{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2609.24983v1",
  "slug": "2609-24983v1-0wyopxo",
  "url": "https://feed7.dev/p/2609-24983v1-0wyopxo",
  "title": "onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction",
  "why_included": "onPanda turns the first bad token into an annotation boundary, regenerating from each correction to create on-policy supervision with less manual rewriting.",
  "summary": "onPanda asks an annotator to find the **first inappropriate token**, replace it from candidates or free-form text, discard the remaining output and regenerate from the corrected prefix. A small controlled study reports **52% lower median annotation time** than manual post-editing.",
  "practical_implication": "For agent tuning, capture corrections at the moment a trajectory first diverges instead of rewriting the final answer wholesale. The resulting positions and paired alternatives can support on-policy SFT and preference data, including tool-connected trajectories.",
  "agent_context": "onPanda asks an annotator to find the **first inappropriate token**, replace it from candidates or free-form text, discard the remaining output and regenerate from the corrected prefix. A small controlled study reports **52% lower median annotation time** than manual post-editing.\n\nFor agent tuning, capture corrections at the moment a trajectory first diverges instead of rewriting the final answer wholesale. The resulting positions and paired alternatives can support on-policy SFT and preference data, including tool-connected trajectories.\n\nThe time result comes from a small study, and the supplied material gives no breakdown by task or annotator. Whether the workflow scales to long coding trajectories without adding review overhead remains open.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2609.24983v1",
    "published_at": "2026-09-21T17:56:53.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "benchmark",
  "domains": [
    "data"
  ],
  "topics": [
    "agent-evals",
    "harness-engineering"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The time result comes from a small study, and the supplied material gives no breakdown by task or annotator. Whether the workflow scales to long coding trajectories without adding review overhead remains open."
  ],
  "connected_context": {
    "meaning": "This turns the prior call to mine real agent trajectories into a specific annotation primitive: intervene at the first bad token and preserve the corrected prefix as on-policy training evidence. It may make expert review cheaper than full rewrites, but the small timing study does not establish that first-token correction remains efficient or sufficient for long, branching tool-use failures.",
    "corpus_size": 843,
    "generated_at": "2026-09-22T09:08:51.561Z",
    "connections": [
      {
        "title": "Improving Agents is a Data Mining Problem — Vivek Trivedy, LangChain",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=CvRngaQZQ3Y",
        "feed7_url": "https://feed7.dev/p/improving-agents-is-a-data-mining-problem-vivek-trivedy-langchain-1cgibr8",
        "reason": "Trace mining identifies recurring production failures; onPanda supplies a concrete way to annotate the first divergence in those traces for subsequent training."
      },
      {
        "title": "State of Data — Sean Cai, Independent / State of Data",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=ZyIoTOAbRfs",
        "feed7_url": "https://feed7.dev/p/state-of-data-sean-cai-independent-state-of-data-0v9fy69",
        "reason": "It operationalizes the preference for real workflow trajectories by extracting correction data from generated paths rather than relying only on manufactured tasks or saved final answers."
      },
      {
        "title": "Trading Desks to Clinical Trials: Parallels in Applied Vertical AI — Ayush Bhardwaj, Allos AI",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=Yphdry8ttAQ",
        "feed7_url": "https://feed7.dev/p/trading-desks-to-clinical-trials-parallels-in-applied-vertical-ai-ayush-1hwvsg3",
        "reason": "Token-level intervention could reduce the cost of expert labeling in vertical workflows, but it does not remove the candidate’s requirement for domain experts when usefulness lacks an objective answer key."
      },
      {
        "title": "TraceML: An Empirical Analysis of Human-Agent Planning in Machine Learning Development",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.26086v1",
        "feed7_url": "https://feed7.dev/p/2608-26086v1-0k35b2d",
        "reason": "TraceML’s long-horizon planning failures suggest an important boundary: correcting the first inappropriate token may capture a local divergence without representing broader deficiencies in effort allocation or plan revision."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-21T17:56:53.000Z",
  "modified_at": "2026-09-21T17:56:53.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2609-24983v1-0wyopxo",
    "json": "https://feed7.dev/p/2609-24983v1-0wyopxo.json",
    "markdown": "https://feed7.dev/p/2609-24983v1-0wyopxo.md"
  }
}