PHASE 2 / RESEARCH ARCHIVE

The archive is part of the argument. It shows which tempting improvements were tested, why they were not promoted, and how the failed lines made the final strategy cleaner.

归档是论证的一部分。它说明哪些看起来诱人的改进被测试过，为什么没有晋级，以及这些失败如何让最终策略更干净。

🟥 00 / Why Archive Matters

如果只展示最终版本，读者很难知道策略是研究出来的，还是从一堆参数里挑出来的。归档页的作用，是把研究者自由度暴露出来：哪些路试过，哪些路停下，哪些结果有观察价值但不能进入主策略。

If only the final version is shown, the reader cannot tell whether the strategy was researched or merely selected from a large parameter table. The archive exposes research degrees of freedom: what was tested, what was stopped, and which results remain useful as observations but not as core rules.

🧭 01 / Archive Map

Research line Original intuition Decision 中文说明
MA / breadth beta switch A market-state filter might avoid bad convertible-bond beta. Dashboard only; not sealed. 有尾部保护价值，但不能成为硬开关。
maturity365_price75 Stricter bond-side safety may improve robustness. Observation; not promoted. 账面改善明显，但机制和归因不够干净。
Liquidity hard filter Better liquidity may improve execution and reduce noise. Not promoted. 单点有效但平台不稳，适合人工检查，不适合硬规则。
Defensive sleeve A defense module may reduce drawdown. Archived. 会改变策略性质，偏离截面轮动主线。
top12_keep40/41 Wider retention may improve path and costs further. Observation; not sealed. 指标更好但 rank drift 更宽，封板选择更克制版本。

🟥 02 / MA / Breadth Switch: Useful Observation, Not A Hard Rule

这条线最有价值的地方，是它没有被简单写成“失败”。MA / breadth 开关确实能在部分尾部阶段降低回撤，说明 bad beta 识别方向不是空穴来风。但它没有强到可以进入封板策略。

The value of this line is that it was not simply labeled “failed.” MA / breadth switches did reduce drawdown in some tail periods, which means the bad-beta identification idea was not empty. But the evidence was not strong enough for a sealed trading rule.

Test stage What was learned 中文说明
Broad market proxy + MA40 Drawdown reduction looked strong, but proxy choice had hindsight risk. 广谱代理下结果好，但代理池选择有后验嫌疑。
Locked Layer1 price80 proxy MA40 still helped, but effect was weaker and more realistic. 固定成策略同源代理后仍有帮助，但打折明显。
Trend / breadth identifiers Breadth20 was the closest candidate; trend alone was weak. breadth20 最接近，但 trend 单独不够。
Platform scan Useful area was narrow, not a broad robust platform. 可用区域偏窄，不是宽平台。
Walk-forward Could not reliably select an effective rule out-of-sample. 事前训练无法稳定选出有效开关。

最终判断是：MA40 更像 tail/path insurance，而不是 timing alpha。它平时会付出机会成本，大回撤时可能赔付；这可以成为风险仪表盘，但不应该自动把组合切成 100% / 0%。

The final reading is that MA40 behaves more like tail/path insurance than timing alpha. It pays an opportunity cost during ordinary periods and may pay off during large drawdowns. That can be a risk dashboard input, but should not become an automatic 100% / 0% switch.

🟨 03 / maturity365_price75: Attractive Result, Messier Mechanism

Figure 1 / 图 1. Layer1 scan: attractive result vs promotable rule.