When Learned Context Planning Fails to Beat Strong Retrieval: A Controlled Study of Planning, Routing, and Reranking for Long-Context QA

当学习型上下文规划无法胜过强检索：针对长上下文问答中规划、路由与重排序的对照研究

Abstract: Learned context planning selects evidence atoms before an answer model reasons over them. We test whether this learned selection improves long-context multiple-choice QA after strong retrieval, routing, budgeted-selector, and reranking controls. 摘要： 学习型上下文规划（Learned context planning）旨在让答案模型进行推理之前先选择证据原子。我们测试了在强检索、路由、预算选择器和重排序等对照条件下，这种学习型选择是否能提升长上下文多项选择问答（MCQ）的效果。

Our primary diagnostic uses all 503 LongBench-v2 MCQ questions with Qwen2.5-7B-Instruct. The planner is SFT-trained on outcome-selected traces from 140 training and 28 development questions; because the 503-question analysis includes those questions, it is partly transductive. 我们的主要诊断使用了 LongBench-v2 中的全部 503 个 MCQ 问题，并采用 Qwen2.5-7B-Instruct 模型。规划器通过 140 个训练问题和 28 个开发问题的结果筛选轨迹进行监督微调（SFT）；由于 503 个问题的分析包含了这些问题，因此该分析在一定程度上属于直推式（transductive）。

At an 18k-character budget, anchored hybrid retrieval reaches 36.18% accuracy and BM25 reaches 35.98%, while the best direct planner-guided method reaches 34.19%. On the untouched 152-question test split, anchored hybrid remains higher (42.11% versus 36.84%). Leakage-safe routers cannot convert a large oracle gap. 在 1.8 万字符的预算下，锚定混合检索（anchored hybrid retrieval）达到了 36.18% 的准确率，BM25 达到了 35.98%，而最佳的直接规划器引导方法仅达到 34.19%。在未触及的 152 个问题测试集上，锚定混合检索依然表现更高（42.11% 对比 36.84%）。防泄漏路由器无法弥补巨大的预言机（oracle）差距。

Under tight budgets, the best planner is ahead by only 0.40 points at 6k and loses at 9k; planner-guided reranking has a +1.79-point estimate at 6k with a paired interval crossing zero and ties the control at 9k. Packing-order and score-flatness analyses did not identify a stable mechanism. Under this setup, learned planning is a weak relevance signal rather than a replacement for strong retrieval. 在严格的预算限制下，最佳规划器在 6k 预算时仅领先 0.40 个百分点，而在 9k 时则表现落后；规划器引导的重排序在 6k 时估计提升了 1.79 个百分点，但配对区间跨越了零点，且在 9k 时与对照组持平。打包顺序和分数平坦度分析未能确定一种稳定的机制。在此设置下，学习型规划仅是一种微弱的相关性信号，而非强检索的替代方案。