Metraipolis › archive › dc_bb1b736756f2

There Is No Neutral Harness: Modern LLM Leaderboards Are Manufactured by Config-Fragile Items — arXiv:2608.21382

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.21382
License (as archived)
CC0-1.0
Archived
2026-10-18T07:38:37.958Z
SHA-256 of the archived text
c547cc95c15322ceb2b0fcfd899a8de6e1c6b0d99481ca635a8637a719c4e8d4
Stable URL of this page
https://metraipolis.com/documento/dc_bb1b736756f2

How to cite

There Is No Neutral Harness: Modern LLM Leaderboards Are Manufactured by Config-Fragile Items — arXiv:2608.21382 — arXiv (en), https://arxiv.org/abs/2608.21382 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bb1b736756f2 (Metraipolis archive dc_bb1b736756f2, sha256 c547cc95c15322ceb2b0fcfd899a8de6e1c6b0d99481ca635a8637a719c4e8d4)
BibTeX
@misc{dcbb1b736756f2,
  title = {There Is No Neutral Harness: Modern LLM Leaderboards Are Manufactured by Config-Fragile Items — arXiv:2608.21382},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.21382},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bb1b736756f2; sha256 c547cc95c15322ceb2b0fcfd899a8de6e1c6b0d99481ca635a8637a719c4e8d4},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.21382. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Multiple-choice benchmarks fix the questions and the correct answers, but not the harness: the order of the options, the wording of the prompt, and whether a language model's answer is read from generated text or from per-option likelihoods. Work on this harness sensitivity reports it as aggregate score variance, leaving unexamined which items the variance falls on and whether they are the items that separate one model from the next. We treat the evaluation harness of large language models (LLMs) as an independent variable and resolve its effect to single items. We introduce the \textit{fragility grid}: 12 open-weight instruction-tuned LLMs from 4 families answer the same 3{,}679 items from 4 benchmarks (ARC, HellaSwag, MMLU, TruthfulQA) under 26 equally defensible harness configurations, recording one correctness bit for every model, item, and configuration. The comparison is matched, since the items, the weights, and the greedy decoding stay fixed while only the harness varies. Under the grid a model's score is a band rather than a point: gemma4-31b scores between 31 and 89 percent depending only on the harness. Three results follow. On the items that two adjacent models both answer stably the pair is tied, and config-fragile items carry 95.7 percent of a pair's gap on average. Four of the 12 models reach rank one under some configuration, so the harness selects the winner. Item discrimination, the property that benchmark-compression methods maximize, correlates with fragility at 0.28 (95 percent CI 0.25 to 0.30), so compression keeps the fragile items rather than removing them. The scoring choice, not the option order that protocols usually fix, is the load-bearing axis. We release the per-item records and the analysis script, from which every number regenerates on a CPU in seconds, and we position the fragility grid as a check a leaderboard can run before it reports an order.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-25 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.21382 — 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 2026-10-18T07:37:52.634Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).