{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiblpfxxycmqgxdermpo7v6hqjj6dr6iy4j4xc4lvoj5hw2wozzt2i",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mh2rmf3lagq2"
  },
  "path": "/t/a-persistent-8-15-failure-rate-across-domains-evidence-for-an-epistemic-boundary/174264#post_1",
  "publishedAt": "2026-03-14T18:19:09.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "https://doi.org/10.5281/zenodo.19020581",
    "GitHub - elly99-AI/MarCognity-AI: A research framework for structured LLM evaluation, claim verification and reflective reasoning mechanisms. · GitHub"
  ],
  "textContent": "**LLMs can’t fix their own epistemic blind spots.**\n\nNot with better prompting.\nNot with stricter verification.\nNot with more sophisticated agents.\n\nThe **MarCognity benchmark** reveals a persistent **8–15% failure rate across 8 domains** ,\neven when the system is equipped with a **full metacognitive layer**.\n\nThis residual error is **not noise**.\n\nIt may be an **emergent property of autoregressive generation itself** —\na structural boundary where the epistemic space accessible to a probabilistic model\nis narrower than the space of claims requiring justification.\n\nI call this the **Epistemic Boundary**.\n\n\n\n**Resources:**\n\n• Zenodo : https://doi.org/10.5281/zenodo.19020581\n\n• GitHub : GitHub - elly99-AI/MarCognity-AI: A research framework for structured LLM evaluation, claim verification and reflective reasoning mechanisms. · GitHub",
  "title": "A Persistent 8–15% Failure Rate Across Domains: Evidence for an Epistemic Boundary"
}