{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreietwchvbch5e4rfa4raa547ybcywjvn7chnosp3xfiwnqduzsyuae",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mjbsu3ef7wu2"
  },
  "path": "/t/operational-self-improvement-in-a-frozen-qwen3-14b-on-consumer-hardware-two-autonomous-rules-66-point-ablation-delta-reasoning-constraints-discovery/175184#post_1",
  "publishedAt": "2026-04-12T06:51:27.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "https://zenodo.org/records/19530926"
  ],
  "textContent": "Paper: https://zenodo.org/records/19530926\n\nA frozen Qwen3 14B (Q4_K_M, Ollama) running on an Apple Mac Mini M4 (24GB) autonomously produced two corrective reasoning constraints through a quality-gated pipeline (MERRCURR), verified against a 110-probe behavioural battery. No fine-tuning, no cloud, no weight modification.\n\n**Results:**\n\n  * Two autonomously promoted rules across independent error classes — one with operational error recurrence 3→0 (Poisson p=0.0498), one with clean validation delta +2 and no regressions\n\n  * Bare model ablation: 28% → 94% with full architecture (Fisher p<0.001, 66-point delta on 20-probe subset)\n\n  * Within-probe dissociation: reasoning accuracy 88–100%, labelling compliance 0% on the same probes (Fisher p<0.001). The model reasons correctly but does not label — reasoning constraints work, formatting instructions do not in any tested configuration\n\n  * Multi-principal transfer: 84%, 83%, 80% across three industries, all CIs overlapping\n\n  * 19 days continuous sovereign operation, zero cloud dependency verified in code\n\n\n\n\nFourth paper in the ATLAS research programme. Prior papers: positional restructuring (DOI: 10.5281/zenodo.19427878), calibrated self-assessment (DOI: 10.5281/zenodo.19435861), MERRCURR pipeline (DOI: 10.5281/zenodo.19448879).\n\nFull methodology, statistical tests, probe rubrics, and limitations in the paper.",
  "title": "Operational Self-Improvement in a Frozen Qwen3 14B on Consumer Hardware — Two Autonomous Rules, 66-Point Ablation Delta, Reasoning Constraints Discovery"
}