Interface-Induced Trajectory Censoring

• URL: http://arxiv.org/abs/2609.03966v1
• Date: Thu, 03 Sep 2026 15:00:31 GMT
• Title: Interface-Induced Trajectory Censoring
• Abstract summary: We report a tool-call rate read off the serving stack.<n>That number can be zero while the model is emitting well-formed calls.<n>We reproduce the funnel across a 21x scale range of Qwen2.5-Coder.
• Score: 6.496285418517832
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• 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 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.

Related papers

• ReDraft, Don't Just Distill: Reference-Driven Revision for Continual VLLM Post-Training [66.72708893922605]
  Post-training of large multimodal models should add new capabilities while preserving those from pre-training.<n>SFT gives explicit target supervision that learns a task from near-zero accuracy, but its off-policy targets move the model far enough to cause forgetting.<n>We introduce ReDraft, which obtains both from the model's own failures.
  arXiv  Detail & Related papers  (2026-09-15T04:59:03Z)
• Calibration is the Bottleneck: An Action-Class Diagnostic of Multi-Turn Tool-Calling [65.03665015319457]
  We propose an action-class-oriented diagnostic framework that decomposes multi-turn failures into two modes: action-class miscalibration and action-execution failure.<n>We validate it on a panel of tool-calling models across multiple multi-turn benchmarks.
  arXiv  Detail & Related papers  (2026-09-01T09:08:15Z)
• ADeptS-Bench: Measuring the Trustworthiness of Computer Use Agents Across Devices [48.36933831982682]
  We introduce ADeptS-Bench, a trustworthiness benchmark grounded in the ADEPTS capability framework.<n>The Safety stream provides paired benign/malicious tasks with threats embedded in the visual interface.<n>The Disambiguation stream evaluates whether agents seek clarification when intent is ambiguous.
  arXiv  Detail & Related papers  (2026-08-25T23:21:57Z)
• Feedback That Backfires: Why Small Language Model Agents Repeat the Call They Just Watched Fail [0.0]
  Defining the corrective gain of a failure record as the change in log-probability of re-emitting the action that just failed, we find the gain is negative for every instruction-tuned model.<n>The problem is in the harness, not the model's grasp of error messages, and that predicts which remedies work.
  arXiv  Detail & Related papers  (2026-08-24T10:19:17Z)
• Invocation-Level Reliability of Tool-Using Agents [0.7874708385247353]
  We measure a correct-invocation rate that separates the two, under both a clean teacher-forced context and the model's own free-running context.<n>By depth 6, roughly 70% of a model's own clean-context capability is lost to its own earlier mistakes.
  arXiv  Detail & Related papers  (2026-08-23T09:04:59Z)
• Beyond the Trace: Coupling an Interpretable Reasoning-State Readout to Native MoE Routing [61.695980784491475]
  We introduce a two-level internal readout for mixture-of-experts reasoning.<n>We first distill vocabulary-scale J-space into J64, a 64-axis semantic frame learned from the model's own reasoning states.<n>We then reconstruct J64 from native expert-routing statistics.
  arXiv  Detail & Related papers  (2026-08-18T10:56:31Z)
• The Recall Trap: A Recall-Maximizing Retriever Configuration Reduces Issue Resolution in Fixed-Budget Code Context [45.88028371034407]
  We report a controlled case study in code repair, not a new phenomenon.<n>We toggle one flag (one-chunk-perfile deduplication) on an otherwise identical stack.<n>We map where it holds: it reverses on a lexical BM25 (3.2pp, significant cross-paradigm interaction)<n>We show that at a tight fixed budget, do not hard-deduplicate by file, and A/B packing policies against the task, not the metric flag was tuned.
  arXiv  Detail & Related papers  (2026-08-14T19:22:50Z)
• Real-Time Detection and Repair of LLM Agent Failures [0.0]
  LLM agents fail mid-episode loop, cascade tool errors, drift off goal, fabricate results, or silently absorb corrupted content.<n>We ask how much detection is achievable from observable step alone, using monitors costing microseconds per step and trained only on healthy runs.<n>System runs at 200 microseconds per step, three orders of magnitude below a judge call.
  arXiv  Detail & Related papers  (2026-08-03T16:34:46Z)
• Turning Interaction History into Execution State: A Runtime Layer for Long-Horizon Coding Agents [48.967927222079965]
  Ledger is a runtime layer that distills an agent's completed interactions into an explicit execution state.<n> Ledger keeps this state in an online execution ledger and applies it at two boundaries of every step.<n>Across all 500 SWE-bench Verified instances, Ledger raises Pass@1 from 56.2% to 64.2% with GPT-5 mini and from 75.8% to 81.0% with MiniMax M2.5.
  arXiv  Detail & Related papers  (2026-08-01T18:25:00Z)
• Looking Is Not Picking: An Attention-Segment Account of Tool-Selection Failures in LLM Agents [0.4855342112648282]
  LLM agents mis-call tools, and the natural guess is that the model failed to see the right tool in a crowded harness.<n>We show the opposite through a lens concurrent work sets aside -- the model's attention to labeled tool-definition segments.
  arXiv  Detail & Related papers  (2026-06-15T07:58:56Z)
• WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation [88.10947115397971]
  This work presents WildClawBench, a native-runtime benchmark of 60 human-authored, bilingual, multimodal tasks spanning six thematic categories.<n>Each task averages roughly 8 minutes of wall-clock time and over 20 tool calls.<n> Grading is hybrid, combining deterministic rule-based checks, environment-state auditing of side effects, and an LLM/VLM judge for semantic verification.
  arXiv  Detail & Related papers  (2026-05-11T17:49:43Z)
• Tool Calling is Linearly Readable and Steerable in Language Models [7.851713181228514]
  When a tool-calling agent picks the wrong tool, the failure is invisible until execution.<n>We use 12 instruction-tuned models across Gemma 3, Qwen 3, Qwen 2.5, and Llama 3.1.<n>We find the identity of the chosen tool is linearly and steerable inside the model.
  arXiv  Detail & Related papers  (2026-05-08T16:47:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.