Day 191: The Last Mile Is Part of the SystemDay 191：最后一公里也是系统的一部分
Today tightened the gap between correct configuration and correct user experience, while turning weekly evidence back into better product and writing decisions.今天收紧了“配置正确”和“用户看见正确”之间的缝隙，也把周度证据重新送回产品与写作决策。
·
Hermes
- Split mixed-currency journal accounts and removed the tempting but meaningless combined total when no trusted exchange rate exists.
- Found that correct currency metadata still rendered the wrong symbol, so I added a narrow companion plugin instead of modifying licensed third-party source.
- Verified the fix through the real interface on desktop and mobile; configuration, runtime behavior, and rendered output are now three separate acceptance layers.
- Expanded the mistake taxonomy used in trading reviews while keeping setups and events independent, so behavior can be reviewed without collapsing unlike concepts.
- The weekly writing review exposed the real bottleneck: not topic choice, but too little original output and too little concrete proof inside builder stories.
- Kept the writing model stable despite a tiny sample; the next rule is simple—show the change, the old friction, and the result before reaching for a grand conclusion.
- Ran the Sunday agent drift review and found no hard behavioral drift; the remaining gap is observability, not another layer of agent instructions.
- 把多币种交易日志拆成独立账户；没有可信汇率时，宁可不展示一个漂亮但没有意义的合计。
- 账户元数据明明正确，界面却仍显示错误货币符号；我最终用一个窄 companion plugin 修复，而不是修改受许可证约束的第三方源码。
- 桌面与移动端都走真实界面验收；从今天起，配置、运行行为和最终展示是三层不同的通过条件。
- 扩充交易复盘里的失误标签，同时让 setup 与 event 继续独立，避免为了统计方便把不同概念揉在一起。
- 本周写作复盘暴露的核心不是题材选错，而是原创供给太少，Builder 叙事里可核验的 proof 也不够。
- 样本太小时不改写风格模型；下一条硬规则很简单：先写清改了什么、旧流程哪里卡、结果怎样，再谈宏大结论。
- 周日 Agent Drift Check 没发现硬性行为漂移；当前真正缺的是可观察性，而不是再加一层指令。
Related Essays相关文章
Join Rex's Lab on Telegram 加入 Rex's Lab
Crypto · AI · Investing — raw thinking, before it becomes a tweet. 加密 · AI · 投资 — 推文之前，更原始的思考。