Interface-Induced Trajectory Censoring
Quick summary
arXiv:2609.03966v1 Announce Type: new Abstract: 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 defecti
Key takeaways
- arXiv:2609.03966v1 Announce Type: new Abstract: 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.
Why it matters
“Interface-Induced Trajectory Censoring” should be evaluated beyond branding and benchmark scores. Its practical importance will emerge in task accuracy, latency, unit cost, safety and integration with real workflows.
Member comments