摘要：学习式上下文规划会在回答模型进行推理之前选择证据原子。我们测试了在纳入强检索、路由、预算选择器和重排序对照后，这种学习式选择是否能够提升长上下文多项选择问答（QA）的表现。我们的主要诊断使用了 LongBench-v2 的全部 503 道 MCQ 题目以及 Qwen2.5-7B-Instruct。规划器基于从 140 道训练题和 28 道开发题中按结果筛选的轨迹进行监督微调（SFT）训练；由于这 503 题分析包含了这些题目，因此该分析部分具有传导式（transductive）特征。在 1.8 万字符预算下，锚定混合检索的准确率达到 36.18%，BM25 达到 35.98%，而最佳的直接规划引导方法仅达到 34.19%。在未接触的 152 道测试集上，锚定混合检索依然更高（42.11% 对 36.84%）。在防止信息泄漏的条件下，路由器无法将较大的预言机（oracle）差距转化为实际收益。在紧预算下，最佳规划器仅在 6000 字符预算时领先 0.40 个百分点，并在 9000 字符预算时落后；规划器引导的重排序在 6000 字符预算下估计有 +1.79 个百分点的提升，但其配对区间跨越零点，并且在 9000 字符预算下与对照方法持平。打包顺序和分数平坦度分析均未发现稳定的作用机制。在这一实验设置下，学习式规划是一种较弱的相关性信号，而非强检索的替代方案。

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. 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. 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. 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.