CounterRefine: Answer-Conditioned Counterevidence Retrieval for Inference-Time Knowledge Repair in Factual Question Answering
This addresses the issue of improving answer accuracy in factual question answering for users of retrieval-grounded systems, representing an incremental advancement by adding a repair mechanism.
The paper tackles the problem of factual question answering errors due to failures of commitment, where systems retrieve relevant evidence but still produce wrong answers, and introduces CounterRefine, a lightweight inference-time repair layer that improves a GPT-5 Baseline-RAG by 5.8 points and achieves a 73.1% correct rate on the SimpleQA benchmark.
In factual question answering, many errors are not failures of access but failures of commitment: the system retrieves relevant evidence, yet still settles on the wrong answer. We present CounterRefine, a lightweight inference-time repair layer for retrieval-grounded question answering. CounterRefine first produces a short answer from retrieved evidence, then gathers additional support and conflicting evidence with follow-up queries conditioned on that draft answer, and finally applies a restricted refinement step that outputs either KEEP or REVISE, with proposed revisions accepted only if they pass deterministic validation. In effect, CounterRefine turns retrieval into a mechanism for testing a provisional answer rather than merely collecting more context. On the full SimpleQA benchmark, CounterRefine improves a matched GPT-5 Baseline-RAG by 5.8 points and reaches a 73.1 percent correct rate, while exceeding the reported one-shot GPT-5.4 score by roughly 40 points. These findings suggest a simple but important direction for knowledgeable foundation models: beyond accessing evidence, they should also be able to use that evidence to reconsider and, when necessary, repair their own answers.