Decoupling Evidence Interpretation from Decision Aggregation in LLM Systems
August 17, 2026
This method separates source interpretation from decision aggregation using a four-field evidence tuple: hypothesis, reliability, rationale, and provenance. This separation identifies count-scale drift, where thresholding unnormalized weights causes posterior operating points to shift based on the number of sources consulted.
HOW THIS AFFECTS YOU
●
builderYou should implement structured evidence tuples instead of simple prompt concatenation to avoid aggregation errors.
●
researcherThis framework provides a more rigorous way to study error propagation in multi-source reasoning.