RCWT: Measuring Task-Budget Displacement from Coordination Content in LLM Calls

This paper introduces the Roundtable Context Window Test (RCWT), a controlled protocol for measuring how coordination content consumes task budget in LLM calls. It varies shared state, prior discussion, tool outputs, summaries, and role instructions while controlling total budget, position, task family, and scoring. In the main context-dependent recall task at W=4096, three commercial models remain near baseline until residual reference evidence falls to a few hundred tokens, then degrade sharply. An intact-task ablation keeps the full task and reference block present while expanding total prompt length; GPT-4.1-mini, Claude Haiku 4.5, and Gemini 2.5 Flash achieve every scored field correctly up to a 95% coordination ratio. The authors therefore frame the main cliff as task-budget displacement, not proof of coordination-induced semantic interference.