Metraipolis › archive › dc_86557689423e
Interface-Induced Trajectory Censoring — arXiv:2609.03966
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.03966
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T22:59:35.242Z
- SHA-256 of the archived text
fe6d97a6b4d460c05f7bd8c5d382277bb1d93375b41139ff3c3e2c74c84f3b62- Stable URL of this page
https://metraipolis.com/documento/dc_86557689423e
How to cite
Interface-Induced Trajectory Censoring — arXiv:2609.03966 — arXiv (en), https://arxiv.org/abs/2609.03966 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_86557689423e (Metraipolis archive dc_86557689423e, sha256 fe6d97a6b4d460c05f7bd8c5d382277bb1d93375b41139ff3c3e2c74c84f3b62)
BibTeX
@misc{dc86557689423e,
title = {Interface-Induced Trajectory Censoring — arXiv:2609.03966},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.03966},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_86557689423e; sha256 fe6d97a6b4d460c05f7bd8c5d382277bb1d93375b41139ff3c3e2c74c84f3b62},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.03966. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Agent evaluations report a tool-call rate read off the serving stack. That number can be zero while the model is emitting well-formed calls: the interface censors the trajectory before anything downstream sees it. On BFCL v4's own data, executor and scorer, holding weights, cases, decoding and seeds fixed and changing only the serving adapter, the same model scores 0.00 or 0.96 / 0.19. A 2x2 over chat template and parser locates the effect exactly: both main effects are exactly zero and all of it sits in the interaction -- no component is defective, and repairing one side of the contract buys precisely nothing. On tau-bench's 115 interactive retail tasks the same swap moves server-parsed calls from 0 to 636 and tasks reaching any tool execution from 0 to 103. Our probe reproduces the funnel across a 21x scale range of Qwen2.5-Coder: the server parses 0/100 at every size while well-formed emitted calls rise to 80/100 at 32B (~72 after calibration against an adjudicated gold standard). Under a matched envelope, across a comparable scale span, the silent fraction stays at 0-2, a prediction committed to the repository before the run. Llama-3.1-8B's 23% rate of calling the task function itself as a tool falls to 0 under one strict:true flag. The mismatch reaches inside the training loop, and its consequence is scale-dependent: in verl's AgentLoop at 7B, 45 of 115 generations carry a complete call; 0 are accepted, 0 execute, 0 return an observation. At 1.5B the same zero is over-determined, so we report the two scales separately. At evaluation time, repairing the adapter restores the mechanism but not a significant outcome gain: parsing 0->84, rescues 0->9, pass rate 53->62 (n.s.). We release a 98-line preflight check that catches every silent failure here. The observed tool-call rate is not a property of the model alone; it is a property of the model-interface stack that measures it. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-07 in cs.LG). Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.03966 — 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 2026-10-16T22:31:32.222Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-07?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).