加权记忆树：按成败打分并衰减记忆，提升长任务Agent推理
Giving an agent a memory hierarchy is not what makes it reason better over long…
Giving an agent a memory hierarchy is not what makes it reason better over long tasks. What helps is scoring each memory and letting unused ones decay out of the prompt.
给代理一个记忆层级结构并不是让它在长任务中推理得更好的原因。真正有帮助的是对每条记忆进行评分，并让那些未被使用的记忆从提示中逐渐淡出。
Agents that keep everything in context lose accuracy to their own stale records. So build a rule that demotes memories the selector keeps skipping, rather than a bigger store.
那些把所有内容都保留在上下文中的代理，会因自身过时的记录而失去准确性。因此，应建立一条规则，将选择器反复跳过的记忆降级，而不是扩大存储空间。
For a long-horizon agent, a demotion rule buys more than a better retriever: lower a memory's standing each time the selector passes it over.
对于长时程代理而言，降级规则比更好的检索器更有价值：每当选择器跳过一条记忆时，就降低其地位。
This paper separates hierarchy from scoring, and only the scored version improves accuracy and prompt cost in every model tested.
这篇论文将层级结构与评分分开，只有经过评分的版本在测试的每个模型中都能提高准确性和提示成本效率。
The Weighted Memory Tree scores each memory by whether its action succeeded, decays it whenever the memory is eligible but unselected, and is set against linear history and an unscored tree.
加权记忆树通过其动作是否成功来对每条记忆进行评分，每当记忆符合条件但未被选中时便使其衰减，并与线性历史和未评分的树进行对比。
– arxiv. org/abs/2608.20631
– arxiv. org/abs/2608.20631
Title: "Weighted Memory Tree: Remembering What Matters for Long-Horizon LLM Agents"
标题：“加权记忆树：记住对长时程LLM代理重要的事情”
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力