Metraipolis › archive › dc_6cd67b95d7f7

HARDEN: Constrained Evolutionary Search for Harder, Answer-Preserving Evaluation Cases — arXiv:2609.30571

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.30571
License (as archived)
CC0-1.0
Archived
2026-10-08T01:38:12.743Z
SHA-256 of the archived text
94e21ff49d4087d75dfa24c253944969e6b3f3d118d47412f79eb612c888b40e
Stable URL of this page
https://metraipolis.com/documento/dc_6cd67b95d7f7

How to cite

HARDEN: Constrained Evolutionary Search for Harder, Answer-Preserving Evaluation Cases — arXiv:2609.30571 — arXiv (en), https://arxiv.org/abs/2609.30571 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_6cd67b95d7f7 (Metraipolis archive dc_6cd67b95d7f7, sha256 94e21ff49d4087d75dfa24c253944969e6b3f3d118d47412f79eb612c888b40e)
BibTeX
@misc{dc6cd67b95d7f7,
  title = {HARDEN: Constrained Evolutionary Search for Harder, Answer-Preserving Evaluation Cases — arXiv:2609.30571},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.30571},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_6cd67b95d7f7; sha256 94e21ff49d4087d75dfa24c253944969e6b3f3d118d47412f79eb612c888b40e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.30571. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Language models are often evaluated on curated benchmarks that underrepresent the complexity of enterprise deployments. We introduce HARDEN, a constrained evolutionary search method to adapt the input of existing evaluation cases into more challenging variants while keeping their expected outputs fixed. HARDEN searches along generated domain-specific complexity axes while enforcing feasibility constraints such as preserving task semantics, realism, and execution validity. Across FinQA, PubMedQA, and ContractNLI and three Qwen3.5 model scales (35B-A3B, 122B-A10B, and 397B-A17B), HARDEN reduces task-model accuracy by 22.7% on average and by up to 49.9% relative to single-pass baselines using the same feasibility checks. These results show that evolutionary search can produce substantially harder valid evaluation cases.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-28 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.30571 — 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 2026-10-08T01:12:14.253Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).