上下文分配法则：生成式搜索中的因果测量与闭环编排

摘要

随着检索增强生成（RAG）转向多样化组合生成，其发展受到两个关键瓶颈的阻碍：证据利用的测量存在缺陷，以及上下文预算分配欠优。我们依次解决这两个问题。为解决测量问题，我们揭示了一种普遍存在的“诊断幻觉”：标准相关性代理指标在难负样本上会灾难性地失效。我们将其替换为一种高效的因果留一探针，该探针能够精确分离生成依赖，并正式校准 LLM 注意力的结构性稀释。为解决分配问题，我们将该因果探针部署在一个去混杂因子网格中。我们证明，当前占主导地位的整体式上下文扩展策略是一种架构陷阱，会受到相关性衰减的惩罚。相反，将计算资源跨多次顺序生成进行迭代分配，可带来 16.7 至 20.5 个绝对百分点的变革性组合召回率提升，并且可稳健扩展至 32B 规模模型。最后，我们将这些解决方案统一为一个可部署的闭环子模调度器。借助一个由归因引导的对比解码器来克服 LLM 的注意力惯性，我们的架构系统地强制整合新证据。通过超越经典开环基线，我们确立了顺序的、反馈驱动的编排作为生成式搜索的决定性范式。我们的代码、数据和因果测量工具可在 https://github.com/PeiYangLiu/ascp 获取。

English

As Retrieval-Augmented Generation (RAG) shifts toward diverse portfolio generation, it is stymied by two critical bottlenecks: flawed measurement of evidence utilization, and suboptimal context budget allocation. We resolve both sequentially. To resolve measurement, we expose a pervasive ``diagnostic illusion'': standard relevance proxies fail catastrophically on hard negatives. We replace them with an efficient causal leave-one-out probe that accurately isolates generative reliance and formally calibrates the structural dilution of LLM attention. To resolve allocation, we deploy this causal probe in a deconfounded factorial grid. We prove that the prevailing strategy of monolithic context widening is an architectural trap penalized by relevance decay. Instead, allocating compute iteratively across multiple sequential generations drives transformative portfolio recall gains of 16.7--20.5 absolute percentage points, scaling robustly up to 32B models. Finally, we unify these solutions into a deployable closed-loop submodular scheduler. Augmented by an attribution-steered contrastive decoder to override LLM attention inertia, our architecture systematically forces fresh evidence integration. By dominating classical open-loop baselines, we establish sequential, feedback-driven orchestration as the definitive paradigm for generative search. Our code, data, and causal measurement instruments are available at https://github.com/PeiYangLiu/ascp.