Metraipolis › archive › dc_4c779183d581

Stable Answers, Unfinished Reasoning: Why Self-Consensus Is Not a Safe Early-Exit Signal — arXiv:2609.09989

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.09989
License (as archived)
CC0-1.0
Archived
2026-10-16T07:53:26.378Z
SHA-256 of the archived text
d736d9c383b97c84b666b98c7b82506fe4de946477cf632ba9386ccf680e6d16
Stable URL of this page
https://metraipolis.com/documento/dc_4c779183d581

How to cite

Stable Answers, Unfinished Reasoning: Why Self-Consensus Is Not a Safe Early-Exit Signal — arXiv:2609.09989 — arXiv (en), https://arxiv.org/abs/2609.09989 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4c779183d581 (Metraipolis archive dc_4c779183d581, sha256 d736d9c383b97c84b666b98c7b82506fe4de946477cf632ba9386ccf680e6d16)
BibTeX
@misc{dc4c779183d581,
  title = {Stable Answers, Unfinished Reasoning: Why Self-Consensus Is Not a Safe Early-Exit Signal — arXiv:2609.09989},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.09989},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4c779183d581; sha256 d736d9c383b97c84b666b98c7b82506fe4de946477cf632ba9386ccf680e6d16},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.09989. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

A natural way to cut reasoning-model inference cost is to repeatedly probe a single partial trajectory for its current answer and stop once probes agree -- self-consensus. We ask whether any such rule is both safe and token-saving, and whether one can be selected once and reused. A preregistered sweep of 3,520 consensus rules, replayed on frozen trajectories from two models and three benchmarks, clears none of three acceptance gates fixed in advance; the frontier reproduces on a held-out split and on two unseen models -- while a boundary-confidence control (DEER) swept through the same pipeline clears all three. The reason lies in the signal: agreement establishes that the current answer persists under a fixed probing procedure, not that the reasoning has terminated -- a consensus-termination gap. Stopping on it commits non-terminal answers. At a rule still saving 32% of the tokens, one stop in nine fires on an answer the trajectory itself later abandons, and most of those stops cut off a correction it would otherwise have made. Widening the agreement window does not remove them: the share levels off near 7%, and by then the saving has fallen to 8%. Probe re-wording and a hand-labelled error taxonomy show the agreed answer is often a placeholder the model had not settled on. Used on its own as the stop signal, agreement fails not because it is insufficiently strict, but because it repeatedly measures the wrong object.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-10 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.09989 — 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 2026-10-16T07:24:00.918Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-10?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).