Evidence Selection

Ask about this topic

Overview

Evidence selection asks which retrieved documents, passages, claims, or multimodal observations are sufficient to support a decision or answer. It matters in retrieval-augmented generation, long-context and video reasoning, clinical decision support, and systematic-review screening, where missing or irrelevant evidence can distort results. Methods include learned retrievers and rerankers, utility and uncertainty scoring, graph or discrete optimization, and agents that acquire, verify, or compress evidence under a cost or token budget. A central challenge is that topical relevance and answer accuracy do not guarantee complete, traceable support: evidence quality, model variability, and downstream effects need separate evaluation, and more or better-filtered evidence does not always improve outcomes.

Latest papers

CardsList

61 papers15 papers in the last four weeks, up 150% on the four weeks before. 0.1% of all new papers.

Feb 23, 2026

Controllable Evidence Selection in Retrieval-Augmented Question Answering via Deterministic Utility Gating

Many modern AI question-answering systems convert text into vectors and retrieve the closest matches to a user question. While effective for topical similarity, similarity scores alone do not explain why some retrieved text can serve as evidence while other equally similar text cannot. When many candidates receive similar scores, systems may select sentences that are redundant, incomplete, or address different conditions than the question requires. This paper presents a deterministic evidence selection framework for retrieval-augmented question answering. The approach introduces Meaning-Utility Estimation (MUE) and Diversity-Utility Estimation (DUE), fixed scoring and redundancy-control procedures that determine evidence admissibility prior to answer generation. Each sentence or record is evaluated independently using explicit signals for semantic relatedness, term coverage, conceptual distinctiveness, and redundancy. No training or fine-tuning is required. In the prototype, a unit is accepted only if it explicitly states the fact, rule, or condition required by the task. Units are not merged or expanded. If no unit independently satisfies the requirement, the system returns no answer. This deterministic gating produces compact, auditable evidence sets and establishes a clear boundary between relevant text and usable evidence.

Victor P. Unda

Independent Researcher, Applied AI Systems

Evidence SelectionRetrieval-Augmented GenerationKnowledge-Intensive QA
AskPDF