Metraipolis › archive › dc_cabbd30fba2d

VACE: Validation-Gated Alternating Co-Evolution of Agent Models and Harnesses — arXiv:2609.37105

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.37105
License (as archived): CC0-1.0
Archived: 2026-10-07T16:26:53.362Z
SHA-256 of the archived text: a850cc2ed82728594e81f4673fe6fd91da21d660096cf2c73e322b76821429fe
Stable URL of this page: https://metraipolis.com/documento/dc_cabbd30fba2d

How to cite

VACE: Validation-Gated Alternating Co-Evolution of Agent Models and Harnesses — arXiv:2609.37105 — arXiv (en), https://arxiv.org/abs/2609.37105 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cabbd30fba2d (Metraipolis archive dc_cabbd30fba2d, sha256 a850cc2ed82728594e81f4673fe6fd91da21d660096cf2c73e322b76821429fe)

BibTeX

@misc{dccabbd30fba2d,
  title = {VACE: Validation-Gated Alternating Co-Evolution of Agent Models and Harnesses — arXiv:2609.37105},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.37105},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cabbd30fba2d; sha256 a850cc2ed82728594e81f4673fe6fd91da21d660096cf2c73e322b76821429fe},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.37105. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Language model agents can be improved by updating their model weights or refining the harness that guides task execution. These components are coupled: weight updates change how the model uses the harness, while harness updates change the trajectories used for training. We propose VACE, Validation-Gated Alternating CoEvolution, which alternates agentic reinforcement learning with trajectory-driven harness refinement. After each RL stage, VACE reuses the collected trajectories to propose a harness revision and evaluates the incumbent and candidate with the updated model held fixed. The candidate guides subsequent training only if it improves validation performance. With Qwen3.5-9B, VACE achieves 45.26% test accuracy on OfficeQA and a mean partial-credit score of 75.19% on AutomationBench, exceeding weight-only RL by 6.43 and 9.09 percentage points and ungated alternation by 4.59 and 6.95 points, respectively. Across 44 harness proposals, 17 reduce validation performance at the updated checkpoint and are rejected before subsequent RL training, highlighting the importance of validation gating.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.37105 — 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 2026-10-07T15:54:53.943Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).