Metraipolis › archive › dc_fc93b6f85018
Harness Evolution Hits a Ceiling: When Weight Training Should Begin — arXiv:2610.11655
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.11655
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-09T05:48:14.816Z
- SHA-256 of the archived text
8b834edf4bd2feebf7749e96ab2c50e87ffc753ff6f0349602a14e675c18ae8c- Stable URL of this page
https://metraipolis.com/documento/dc_fc93b6f85018
How to cite
Harness Evolution Hits a Ceiling: When Weight Training Should Begin — arXiv:2610.11655 — arXiv (en), https://arxiv.org/abs/2610.11655 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fc93b6f85018 (Metraipolis archive dc_fc93b6f85018, sha256 8b834edf4bd2feebf7749e96ab2c50e87ffc753ff6f0349602a14e675c18ae8c)
BibTeX
@misc{dcfc93b6f85018,
title = {Harness Evolution Hits a Ceiling: When Weight Training Should Begin — arXiv:2610.11655},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.11655},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fc93b6f85018; sha256 8b834edf4bd2feebf7749e96ab2c50e87ffc753ff6f0349602a14e675c18ae8c},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.11655. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Improving a long-horizon LLM agent means evolving the harness around a frozen model or training its weights. We let a self-evolving harness make the system stronger first, then cross seed and evolved harnesses with base and trained weights to learn which gains the trained model keeps and which still need the runtime. We show that the right lever can be read off the agent's failure composition: labelling failed trajectories by the first signal that fires separates process failures (blocked calls, loops, exhausted step budgets) from content failures (a delivered plan that is poor). Harness evolution repairs the former, the behaviour it instils can be trained into the weights, and content failures are what weight training is for. On DeepPlanning, a self-evolving harness loop lifts the held-out score of Qwen3.5-4B from 0.16 to 0.30 and of Qwen3.5-9B from 0.32 to 0.44; for 4B, held-out delivery rises from 55% to 90% while content failures are left for the weights. LoRA adapters trained on evolved-harness trajectories internalise the gain: under the original harness they add +0.13 on held-out tasks for both sizes; on 4B they stack with the harness to more than double the held-out score, and on 9B the adapter alone matches the full evolution line, cutting content failures from a quarter of trajectories to one in twenty. A placebo adapter trained on answer-shuffled trajectories falls below the base model. The loop transfers to WebArena-Lite (+0.09 on 117 unseen tasks), where the gain lives in what the model sees and adapters do not add to it. The result is a diagnose-then-intervene rule applied twice: read the failure composition to choose between harness and weights, then read what the accepted edits changed to decide which gains to train in. Scores are four-rollout means against fresh anchors, same-night except where marked, across eight models from six families and two benchmarks. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.11655 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-09T05:47:00.816Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).