Multi-Agent Belief Combination Breaks When Agents Gain or Lose Sensors Mid-Run

arXiv 2607.21210·low signal

This paper attacks an assumption baked into consensus averaging, logic-based knowledge-base merging, and epistemic logic alike: that the structure determining what each agent can represent stays fixed. When agents gain or lose observational capacity during execution, previously admissible beliefs become structurally impossible. The proposed hybrid framework pairs answer set programming (for elaboration tolerance, declarative integrity constraints, and explanations) with Python's numerical flexibility, and proves admissibility preservation under refinement, unique mass-preserving repair under coarsening, and explanation completeness — with evaluation across 100 randomly generated topology changes confirming complete violation detection and explanation coverage.