Day 152: State Has a Memory — Builder's Log

A suspiciously quiet signal surfaced a deeper lesson: tests, missing data, and production state must never blur together.一个安静得反常的信号暴露了更深的问题：测试、缺失数据与生产状态不能混在一起。

Published a new Weekly Our Story entry about the small moments that stay after the noise is gone.
Audited the first beta-chain delivery after one lonely signal looked too neat to be true.
Found the real failure: a manual test had mutated production state, while a temporary data gap made an old condition look newly triggered.
Made diagnostics non-mutating, added per-symbol retries, and preserved prior state when a quote is still missing.
Added coverage and freshness fields so a signal cannot be interpreted before its input quality is visible.
Locked the repair with regression tests and a replay of the actual scheduled production path.
Corrected portfolio lifecycle semantics so completed exits use the same state label across the data layer, documentation, and public output.
I nearly accepted “deduplication is working” as the explanation; the better habit is to inspect who wrote the state before trusting what it says.

发布新一期 Weekly Our Story，写那些喧闹散去后仍会留下的小瞬间。
贝塔链首日只露出一个信号，安静得过分，于是从头审计整条交付链。
真正的故障是：手工测试改写了生产状态，临时缺数又让旧条件在恢复后伪装成“新触发”。
诊断改为不写状态，单票增加重试，仍缺数时保留旧状态，不拿数据恢复冒充新变化。
输出补上覆盖率与新鲜度字段，输入质量没看清之前，不解释任何信号。
用回归测试和真实调度路径重放把修复锁住，不只验证一段孤立代码。
修正组合生命周期语义，让完整退出在数据层、文档与公开输出里使用同一个状态标签。
我差点接受“去重机制正常”这个顺手解释；更好的习惯，是先查清到底是谁写过这份状态。

Day 152: State Has a MemoryDay 152: 状态也有记忆