RESEARCH CATALOG / 2026-10-06
验证、失败记录与研究资产Validation, failures & research assets驗證、失敗記錄與研究資產
旧对话档案说明：整理日期2026-10-06。以下旧对话数值是历史研究记录，不是今日行情或交易指令。原始工作簿与完整回测文件尚未取得，对原模型没有独立重跑；“冻结”指源对话的记录状态，不证明本站当时已公开。新的HORIZON-02／03／04公开数据实现及分别注明截面的估值另有具名研究记录，不能与旧样本合并。Historical conversation archive, compiled 2026-10-06. The earlier conversation figures are historical records, not today’s quotes or trade instructions. Original workbooks and complete backtest files were not obtained; for those original models, no independent rerun occurred. “Frozen” describes the source record, not proof of publication here at that time. The new HORIZON-02/03/04 public-data implementations and individually dated valuations have a separate named record and must not be merged with earlier samples.舊對話檔案説明：整理日期2026-10-06。以下舊對話數值是歷史研究記錄，不是今日行情或交易指令。原始工作簿與完整回測文件尚未取得，對原模型沒有獨立重跑；“凍結”指源對話的記錄狀態，不證明本站當時已公開。新的HORIZON-02／03／04公開數據實現及分別註明截面的估值另有具名研究記錄，不能與舊樣本合併。
本页提供横贯全部模型层的验证基础。各实验按自己的任务、期限、标签和样本比较，不把开发成绩称为独立验证。This page is the validation foundation across all model layers. Compare experiments within their task, horizon, label and sample; development scores are not independent validation.本頁提供橫貫全部模型層的驗證基礎。各實驗按自己的任務、期限、標籤和樣本比較，不把開發成績稱為獨立驗證。
H13：经营现金流、资产桥接与独立估值H13: operating cash flows, asset bridge and independent valuesH13：經營現金流、資產橋接與獨立估值
市场与输入截面为2026-10-06收盘 $239.24，行情于2026-10-07 11:07纽约重新核验；不是10月7日收盘估值。经营/EPS路径保留冻结H09资料，资本现金字段本轮重新提取。最新资本报表截至2026-07-26，申报2026-08-26，股数封面日2026-08-21；历史资料按原始申报+1日重建可用性，完整不可变PIT档案仍不齐全。The market/input cutoff is the 2026-10-06 close of $239.24, rechecked on October 7 at 11:07 New York; this is not an October 7 closing valuation. Operating/EPS paths retain frozen H09 inputs; capital/cash fields were extracted for this study. The latest capital report ends July 26, filed August 26, with cover shares dated August 21. Historical availability is reconstructed from original filings plus one day; a complete immutable PIT archive is still absent.市場與輸入截面為2026-10-06收盤 $239.24，行情於2026-10-07 11:07紐約重新核驗；不是10月7日收盤估值。經營/EPS路徑保留凍結H09資料，資本現金字段本輪重新提取。最新資本報表截至2026-07-26，申報2026-08-26，股數封面日2026-08-21；歷史資料按原始申報+1日重建可用性，完整不可變PIT檔案仍不齊全。
| 期限Horizon期限 | H13基准独立计算值H13 Base independent valueH13基準獨立計算值 | H09价格预测参考H09 price-forecast referenceH09價格預測參考 |
|---|---|---|
| 0M | $260.11 | $239.24 |
| 6M | $272.95 | $286.91 |
| 12M | $286.36 | $326.21 |
H13 已完成独立现金流估值，但本轮三年股价误差及分析师平均目标价差异均大于H09。预设联合门槛未通过；保留H09价格预测研究参考，H13只作为条件基本面与现金折现解释层。它没有提升本次样本的预测表现，未晋升正式模型。H13 completes an independent cash valuation, but three-year price errors and analyst-target gaps are both larger than H09 in this study. The prespecified joint screen does not pass. Retain H09 as the research price-forecast reference; H13 remains a conditional fundamental and cash-discounting explanation layer. It does not improve this sample’s forecast performance and is not promoted to a formal model.H13 已完成獨立現金流估值，但本輪三年股價誤差及分析師平均目標價差異均大於H09。預設聯合門檻未通過；保留H09價格預測研究參考，H13只作為條件基本面與現金折現解釋層。它沒有提升本次樣本的預測表現，未晉升正式模型。
H13的0M、6M、12M由经营价值和对应期限的股权桥接独立计算。H09继续单列为市场锚定的条件价格预测参考；两者回答的问题不同。H13 independently calculates 0M, 6M and 12M through operating value and the equity bridge at each horizon. H09 remains a separate market-anchored conditional price forecast; the two measures answer different questions.H13的0M、6M、12M由經營價值和對應期限的股權橋接獨立計算。H09繼續單列為市場錨定的條件價格預測參考；兩者回答的問題不同。
H07–H10三年比较H07–H10 three-year comparisonH07–H10三年比較 · H12档案H12 archiveH12檔案 · H11档案H11 archiveH11檔案 · H10档案H10 archiveH10檔案 · H09档案H09 archiveH09檔案 · H08档案H08 archiveH08檔案 · H07档案H07 archiveH07檔案 · H06档案H06 archiveH06檔案 · H05档案H05 archiveH05檔案 · H04档案H04 archiveH04檔案 · H03档案H03 archiveH03檔案 · H02档案H02 archiveH02檔案
什么算通过What counts as passing什麼算通過
先比较相同origin、相同期限、相同标签的基准，再看主误差、尾部、面板稳定性与独立样本。开发期walk-forward不等于未被反复调参的独立留出集；不同17样本实验也未必是同一组17个origin。Compare baselines on identical origins, horizons and labels, then main error, tails, panel stability and independent samples. Development walk-forward is not an untouched holdout; different17-observation experiments need not contain the same17 origins.先比較相同origin、相同期限、相同標籤的基準，再看主誤差、尾部、面板穩定性與獨立樣本。開發期walk-forward不等於未被反覆調參的獨立留出集；不同17樣本實驗也未必是同一組17個origin。
MALE = mean(|ln(prediction / actual)|), only positive inputs APE = |prediction − actual| / |actual| P90 APE = 90th-percentile absolute percentage error Touch ≠ endpoint success; retrospective reconstruction ≠ contemporaneous public PIT
MALE显示为对数误差，不写成百分比。源对话同时使用相对预测值的误差定义；那类比例单独标明，不与本页APE混用。零或非正输入按指标定义排除并记录。MALE is a log error, not a percentage. Source records also use prediction-denominator errors; label those separately rather than mix them withAPE here. Record exclusions for zero/nonpositive inputs.MALE顯示為對數誤差，不寫成百分比。源對話同時使用相對預測值的誤差定義；那類比例單獨標明，不與本頁APE混用。零或非正輸入按指標定義排除並記錄。
VALUATION-GATE-02 · 同17个成熟originVALUATION-GATE-02 · same17 matured originsVALUATION-GATE-02 · 同17個成熟origin
| 候选Candidate候選 | MALE | Median APE | P90 APE |
|---|---|---|---|
| Unchanged price | .678 | 54.2% | 70.0% |
| Raw EPS × MRate | .776 | 60.7% | 89.4% |
| 03A EPS × MRate | .644 | 52.6% | 87.4% |
| CAL EPS × MRate | .580 | 24.1% | 92.9% |
| 03A EPS × MBE | .553 | 30.8% | 86.1% |
| CAL EPS × MBE | .514 | 25.1% | 84.2% |
03A×MBE主误差改善18.5%，但尾部86.1%对基准70.0%，相对恶化约23%，超过≤10%门槛。13/17样本EPS与倍数误差方向相反、相关−.55：最终误差下降可能是抵消，不能证明两层都准确。没有最终价格Champion。03A×MBE improves main error18.5%, but tail86.1% versus70.0% worsens relatively~23%, beyond the≤10% gate. EPS/multiple errors have opposite signs in13/17, correlation−.55: a smaller final error may reflect cancellation rather than two accurate components. No final price champion.03A×MBE主誤差改善18.5%，但尾部86.1%對基準70.0%，相對惡化約23%，超過≤10%門檻。13/17樣本EPS與倍數誤差方向相反、相關−.55：最終誤差下降可能是抵消，不能證明兩層都準確。沒有最終價格Champion。
经营与倍数实验账本Operating & multiple experiment ledger經營與倍數實驗賬本
| 实验Experiment實驗 | 源成绩Source result源成績 | 结论Conclusion結論 |
|---|---|---|
| OPERATING-03A | 收入MALE .559→.434；中位38.5→35.3%、P90 74.1→67.3%；4/4面板改善。EPS .829→.665。RevenueMALE .559→.434; median38.5→35.3%, P90 74.1→67.3%;4/4 panels improve. EPS .829→.665.收入MALE .559→.434；中位38.5→35.3%、P90 74.1→67.3%；4/4面板改善。EPS .829→.665。 | 经营影子；完整价格尾部仍失败。Operating shadow; complete price tail still fails.經營影子；完整價格尾部仍失敗。 |
| MULTIPLE-01 | GAAP主标签M3 .382 vs MRate .427；换成trailing标签M3 .827 vs current .433。GAAP main-labelM3 .382 vsMRate .427; trailing-labelM3 .827 vs current .433.GAAP主標籤M3 .382 vs MRate .427；換成trailing標籤M3 .827 vs current .433。 | 标签敏感；不转用Core EPS。Label-sensitive; do not transfer toCore EPS.標籤敏感；不轉用Core EPS。 |
| MULTIPLE-02A | Core标签MRate .423、中位34.3%、P90 161.4%；β17/17为负。Trailing .517 vs persistence .452。Core-labelMRate .423, median34.3%, P90 161.4%;β negative17/17. Trailing .517 vs persistence .452.Core標籤MRate .423、中位34.3%、P90 161.4%；β17/17為負。Trailing .517 vs persistence .452。 | 暂定宏观层；非稳健forward PE。Provisional macro layer; not robust forwardPE.暫定宏觀層；非穩健forward PE。 |
| MBE2.0 | 倍数MALE .423→.396，改善6.4%、2/4面板。MultipleMALE .423→.396, improvement6.4%,2/4 panels.倍數MALE .423→.396，改善6.4%、2/4面板。 | 未晋升，保留失败／限制。Not promoted; retain limitations.未晉升，保留失敗／限制。 |
| HORIZON growth+rate | 另一实验12M .364 vs rate .423，P90 116.5 vs133.9%；增长符号8正9负。Different12M experiment .364 vs rate .423, P90 116.5 vs133.9%; growth sign8 positive/9 negative.另一實驗12M .364 vs rate .423，P90 116.5 vs133.9%；增長符號8正9負。 | 不能与上行161.4%表合并；系数不稳，未晋升。Do not merge with161.4% above; unstable coefficient, not promoted.不能與上行161.4%表合併；係數不穩，未晉升。 |
期限价格模型没有打败基准Horizon price models did not beat baseline期限價格模型沒有打敗基準
| Horizon | MTS price MALE | Unchanged MALE | Oracle EPS MALE |
|---|---|---|---|
| 3M | .483 | .235 | .522 |
| 6M | .537 | .385 | .515 |
| 9M | .585 | .519 | .423 |
| 12M | .950 | .656 | .417 |
四个可实现期限价格模型均落后基准。9M oracle虽主误差改善，尾部89.6%仍差于基准60.2%；使用未来EPS不能替代实盘验证。上述期限实验与GATE-02分开，不混算成绩。All four feasible horizon price models lag baseline.9M oracle improves main error but its89.6% tail remains worse than baseline60.2%; futureEPS does not replace live validation. Keep this experiment separate fromGATE-02.四個可實現期限價格模型均落後基準。9M oracle雖主誤差改善，尾部89.6%仍差於基準60.2%；使用未來EPS不能替代實盤驗證。上述期限實驗與GATE-02分開，不混算成績。
停止规则与下一阶段Stop rule & next phase停止規則與下一階段
原研究已停止在同一47季度／17估值origin上继续搜索λ、权重、交互和校准器。后续期限与期权实验是独立问题，仍须记录开发集与新样本边界，不能把重复利用同一历史称作新的独立验证。Original research stops searchingλ, weights, interactions and calibrators on the same47 quarters/17 valuation origins. Subsequent horizon/options experiments are separate questions but must record development/new-sample boundaries; reusing history is not new independent validation.原研究已停止在同一47季度／17估值origin上繼續搜索λ、權重、交互和校準器。後續期限與期權實驗是獨立問題，仍須記錄開發集與新樣本邊界，不能把重複利用同一歷史稱作新的獨立驗證。
| 下一步Next step下一步 | 完成标准Completion criterion完成標準 |
|---|---|
| 经营Operating經營 | 新origin冻结；9M收入／GM／Opex预测，成熟后分层评分。Freeze new origins;9M revenue/GM/Opex forecasts, score components at maturity.新origin凍結；9M收入／GM／Opex預測，成熟後分層評分。 |
| BuyZone | 02B按预定顺序跨股，保留负例、成交规则和成本。02B prespecified cross-stock order, retain negatives, fills and costs.02B按預定順序跨股，保留負例、成交規則和成本。 |
| CC / PMCC | κ与IV→RV校准、完整合约时点、稳健Edge；再独立验证PMCC结构。Calibrateκ andIV→RV, complete contract timestamps, robust edge; separately validatePMCC.κ與IV→RV校準、完整合約時點、穩健Edge；再獨立驗證PMCC結構。 |
| 价格ChampionPrice champion價格Champion | 主误差、尾部和稳定性共同过门槛；新增未触碰数据。Pass main-error, tail and stability gates together on untouched data.主誤差、尾部和穩定性共同過門檻；新增未觸碰數據。 |
公开摘要与内部证据Public summary & internal evidence公開摘要與內部證據
网站资产是可追溯的方法、版本演进、错误发现、停止规则与预测兑现账本。公开研究摘要和必要的聚合指标；供应商原始数据、完整Pro记录、截图、账户信息和凭证留在内部。内部原文归档不进入网站构建或源仓库。The website asset is traceable methods, version evolution, error discovery, stop rules and prediction outcomes. Publish summaries and necessary aggregate metrics; vendor raw data, fullPro records, screenshots, account details and credentials remain internal. Internal source archives are excluded from site builds and the source repository.網站資產是可追溯的方法、版本演進、錯誤發現、停止規則與預測兑現賬本。公開研究摘要和必要的聚合指標；供應商原始數據、完整Pro記錄、截圖、賬户信息和憑證留在內部。內部原文歸檔不進入網站構建或源倉庫。
PIT必须同时保留资料发布日期、当时可见时间与抓取时间。当前网页快照不能证明历史共识；从现在归档的live PIT不能倒填成过去可得数据。PIT requires publication, historical availability and retrieval timestamps. A current page snapshot cannot prove historical consensus; live PIT archived now cannot be backfilled as past availability.PIT必須同時保留資料發佈日期、當時可見時間與抓取時間。當前網頁快照不能證明歷史共識；從現在歸檔的live PIT不能倒填成過去可得數據。
查看冻结账本与源数据冲突Frozen ledger & source discrepancies查看凍結賬本與源數據衝突 · 查看V1历史验证边界V1 historical evidence boundaries查看V1歷史驗證邊界