Metraipolis › archive › dc_d35864be30cd

LatentSift: Policy-State Filtering for Token-Efficient Verification of Software Engineering Agents — arXiv:2609.36371

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.36371
License (as archived): CC0-1.0
Archived: 2026-10-07T16:25:44.248Z
SHA-256 of the archived text: 08c89bf7929f61de6741ebeac53b624ea4515d76861a1954d2b8595b74ed22df
Stable URL of this page: https://metraipolis.com/documento/dc_d35864be30cd

How to cite

LatentSift: Policy-State Filtering for Token-Efficient Verification of Software Engineering Agents — arXiv:2609.36371 — arXiv (en), https://arxiv.org/abs/2609.36371 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_d35864be30cd (Metraipolis archive dc_d35864be30cd, sha256 08c89bf7929f61de6741ebeac53b624ea4515d76861a1954d2b8595b74ed22df)

BibTeX

@misc{dcd35864be30cd,
  title = {LatentSift: Policy-State Filtering for Token-Efficient Verification of Software Engineering Agents — arXiv:2609.36371},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36371},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_d35864be30cd; sha256 08c89bf7929f61de6741ebeac53b624ea4515d76861a1954d2b8595b74ed22df},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36371. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Test-time scaling improves software engineering agents by generating multiple candidate trajectories and selecting the best one. Verifying and selecting among these long interactions can consume as many tokens as generation itself. Existing hybrid workflows first apply an LLM-based execution-free (EF) verifier to filter candidates before running tests, which adds another model pass over every trajectory. We introduce LatentSift, a token-free and execution-free filter that replaces this first stage with hidden states the policy already produces while generating the candidates. It represents each candidate through its reasoning, observation, and function-call states, compares them with positive and negative banks of such states collected from successful and unsuccessful trajectories during policy training, and fuses the resulting distance scores with a learned linear score to retain promising candidates for the execution-based stages. On SWE-bench Verified, across three agents and two policy sizes, LatentSift cuts EF-verifier tokens by 66.6--81.0% and total verification tokens, which include test generation, by 49.1--62.1% at K=16, while hybrid Best@16 matches or improves on each agent's reference workflow, rising from 59.26% to 60.06% on DeepSWE-Preview.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Software Engineering (cs.SE) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.36371 — 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 2026-10-07T15:54:53.826Z 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).