A Post-Mortem Without Excuses

Dear Shareholders,

致股東：

A post-mortem has value only if it changes future behavior. Explanation without control improvement is storytelling. This letter reviews where our process was incomplete even though the reported result remained slightly positive.

只有能改變未來行為的事後檢討才有價值。只有解釋、沒有控制改善，就只是說故事。本封信檢討即使報告結果仍略為正數，我們的流程在哪些地方仍不完整。

Our framework separates thesis error, translation error, sizing error, timing error, and execution error. Thesis error means the causal view was wrong. Translation error means the view may have been right but the selected instrument did not express it cleanly. Sizing error means the exposure was inconsistent with uncertainty. Timing error means the catalyst window was misunderstood. Execution error means the decision was not implemented at the intended price, speed, or hedge ratio.

我們的框架區分論點錯誤、轉譯錯誤、規模錯誤、時點錯誤與執行錯誤。論點錯誤代表因果判斷本身錯誤。轉譯錯誤代表觀點可能正確，但選擇的工具沒有乾淨表達。規模錯誤代表曝險與不確定性不一致。時點錯誤代表誤解催化窗口。執行錯誤則代表未能以預定價格、速度或避險比率完成決策。

Our market thinking was broadly cautious about unstable correlations and long-duration dependence. That caution was useful, but it also created a risk of underparticipation when risk appetite improved. We recognized that trend and convexity strategies could outperform in a volatile year, while credit and relative value could stabilize the base. Our weakness was not the classification. It was the speed with which we converted changing evidence into position changes.

本期市場思考對不穩定相關性與過度依賴長久期保持審慎。這種審慎有用，但當風險偏好改善時，也產生參與不足的風險。我們辨識到，在高波動年度中，趨勢與凸性策略可能領先，而信用與相對價值可穩定收益底座。我們的弱點不在分類，而在把新證據轉化為部位改變的速度。

Our trading action avoided a large drawdown but captured little upside. We held back while waiting for volatility to normalize, and in some cases normalization occurred through price appreciation rather than through a quiet consolidation. We also maintained some defensive assumptions after the active market driver had migrated. A rule can become stale if its update condition is too slow.

交易行動避免了大型回撤，但只捕捉到有限上行。我們等待波動正常化，但在部分情況下，正常化是透過價格上漲完成，而不是安靜盤整。我們也在主要市場驅動因素已經遷移後，仍保留部分防禦假設。如果更新條件太慢，規則本身也會過時。

The annual position remains +0.00005 unit profit. The positive sign does not excuse missed opportunity, and missed opportunity does not justify abandoning risk control. The correct response is to improve responsiveness without converting the portfolio into a prediction machine.

本年度部位仍為 +0.00005 單位正收益。正數不能成為錯失機會的藉口，而錯失機會也不能合理化放棄風險控制。正確回應是提高反應速度，同時避免把投資組合變成純粹預測機器。

The lesson learned is that reduction rules and addition rules need equal precision. Risk systems often specify exactly when to cut but describe re-entry with vague phrases such as “when conditions improve.” That asymmetry can protect during the first shock and then leave the portfolio stranded during recovery. We also learned that an update rule must identify not only threshold levels but persistence, speed, and market confirmation.

本期教訓是，減曝規則與加曝規則需要同等精確。風險系統常常很準確地規定何時削減風險，卻用「情況改善時」這類模糊語句描述重新進場。這種不對稱可以在第一波衝擊中提供保護，卻讓投資組合在復甦時被留在場外。我們也學到，更新規則不能只辨識閾值，還必須辨識持續時間、變化速度與市場確認。

Post-mortems are most useful when they identify the earliest observable point at which a different action became justified. We distinguish information that was genuinely unavailable from information that was available but ignored, delayed, or assigned insufficient weight. This prevents unfairly judging a decision with hindsight while still holding the process accountable for evidence it should have used.

事後檢討最有價值之處，是找出最早在哪個可觀察時點，不同行動已具有合理依據。我們區分真正無法取得的資訊，以及當時已可取得但被忽略、延遲處理或權重不足的資訊。這能避免使用事後眼光不公平地評判決策，同時仍要求流程對原本應使用的證據負責。

Every lesson must end in one of four outcomes: a monitoring change, a sizing change, an execution change, or no change because the event was not reasonably actionable. Adding a new rule for every loss would create an overfitted system that performs perfectly against history and poorly against the future. Controls are added only when the failure mechanism is repeatable and the proposed remedy would have improved decisions across more than one scenario.

每項教訓都必須導向四種結果之一：改變監控、改變規模、改變執行，或因事件不具合理可操作性而不作改變。如果每次損失都新增一條規則，系統會過度配適，只能完美解釋歷史，卻無法面對未來。只有當失效機制可以重複，而且補救措施能改善不只一種情境下的決策時，我們才會加入控制。

Decision quality is assessed using information available at the time, but process quality includes whether the organization sought the right information soon enough. A team cannot excuse a late response by saying the final data had not arrived if faster market indicators, management behavior, funding conditions, or price relationships were already contradicting the thesis. The review therefore examines both the decision and the design of the information pipeline.

決策品質應使用當時可得資訊評估，但流程品質也包括組織是否及時尋找正確資訊。如果更快速的市場指標、管理層行為、融資條件或價格關係已經反駁論點，團隊不能只以最終資料尚未公布為理由解釋反應過慢。因此，檢討不只檢查決策，也檢查資訊管線設計。

Organizational learning requires preserving disagreement. Minority views, rejected alternatives, and uncertainty ranges are recorded before the outcome is known. Afterward, we compare which assumptions were informative, which were noise, and whether group confidence became detached from evidence. The objective is not to reward dissent automatically, but to prevent consensus from erasing useful warnings.

組織學習需要保留分歧。在結果尚未揭曉前，我們會記錄少數觀點、被拒絕的替代方案與不確定性區間。事後再比較哪些假設具有資訊價值、哪些只是雜訊，以及群體信心是否與證據脫離。目的不是自動獎勵反對意見，而是避免共識抹除有用警告。

Control changes receive an owner, implementation date, testing method, and retirement condition. A rule that no longer addresses a live risk should be revised or removed rather than accumulated indefinitely. This keeps the process rigorous without allowing governance complexity to become a source of slower or less transparent decisions.

控制變更會被指定負責人、執行日期、測試方法與退場條件。若規則不再處理現存風險，就應修訂或移除，而不是無限累積。這能保持流程嚴謹，同時避免治理複雜度本身造成較慢或較不透明的決策。

Our next action is to implement a three-stage re-entry protocol and maintain an error budget alongside the drawdown budget. Stage one is a small diagnostic position after the stress variable stops worsening. Stage two requires improved breadth, liquidity, or correlation. Stage three requires thesis confirmation and permits meaningful risk, still within the drawdown budget. Every stage will have a reversal condition. The drawdown budget limits capital loss, while the error budget tracks recurring process failures such as late reduction, premature entry, hedge mismatch, or documentation gaps. Repeated process errors will trigger review even when aggregate P&L remains positive. Together, these changes are designed to make prudence responsive rather than static.

下一步，我們將實施三階段重新進場程序，並在回撤預算之外建立錯誤預算。第一階段是在壓力變數停止惡化後建立小型診斷部位。第二階段要求市場廣度、流動性或相關性改善。第三階段要求論點得到確認，並允許在回撤預算內配置較實質風險。每個階段都會設定反轉條件。回撤預算限制資本損失；錯誤預算則追蹤反覆出現的流程失誤，例如減曝過晚、進場過早、避險不匹配或文件缺口。即使總體損益仍為正數，只要流程錯誤反覆出現，也會觸發檢討。這些改變共同旨在讓審慎具有反應能力，而不是停留在靜態。