NVDA / H11 / OPERATING GROWTH PERSISTENCE
从H09继续优化：H11增长持续率实验Continuing H09: H11 growth-persistence experiment從H09繼續優化：H11增長持續率實驗
沿用2026-10-06输入与市场收盘截面；2026-10-07纽约09:27再次核验收盘$239.24。财报与指引沿用冻结资料，并非全部输入刷新至10月7日。Use the 2026-10-06 input and market-close cutoff. The $239.24 close was rechecked on 2026-10-07 at 09:27 New York. Financials and guidance remain frozen; this is not a complete October 7 input refresh.沿用2026-10-06輸入與市場收盤截面；2026-10-07紐約09:27再次核驗收盤$239.24。財報與指引沿用凍結資料，並非全部輸入刷新至10月7日。
H11未通过三年股价与共识联合筛选。按用户选择，本轮主表以H09作为继续研究的参考；H11仅保留实验诊断。参考身份由此前H07明确切换至H09，不代表正式Champion晋升。H11 fails the joint three-year price and consensus screen. Following the user’s selection, H09 is this round’s headline research reference; H11 remains experimental diagnostics. This explicitly changes the earlier H07 reference to H09 without promoting a formal Champion.H11未通過三年股價與共識聯合篩選。按用户選擇，本輪主表以H09作為繼續研究的參考；H11僅保留實驗診斷。參考身份由此前H07明確切換至H09，不代表正式Champion晉升。
6M股价平均误差从H09的14.33%扩大至H11的19.03%，12M从14.91%扩大至27.91%。分析师平均绝对百分比差异有所缩小，但对数误差恶化，不能据此采用H11。Mean realized-price error increases from H09’s 14.33% to H11’s 19.03% at 6M and from 14.91% to 27.91% at 12M. The mean absolute percentage gap to analysts improves, but log error worsens; this does not justify adopting H11.6M股價平均誤差從H09的14.33%擴大至H11的19.03%，12M從14.91%擴大至27.91%。分析師平均絕對百分比差異有所縮小，但對數誤差惡化，不能據此採用H11。
当前最终研究参考：H09Current final research reference: H09當前最終研究參考：H09
沿用2026-10-06输入与市场收盘截面；2026-10-07纽约09:27再次核验收盘$239.24。财报与指引沿用冻结资料，并非全部输入刷新至10月7日。Use the 2026-10-06 input and market-close cutoff. The $239.24 close was rechecked on 2026-10-07 at 09:27 New York. Financials and guidance remain frozen; this is not a complete October 7 input refresh.沿用2026-10-06輸入與市場收盤截面；2026-10-07紐約09:27再次核驗收盤$239.24。財報與指引沿用凍結資料，並非全部輸入刷新至10月7日。
| 期限Horizon期限 | 目标日 → 盈利窗口末日Target → earnings-window end目標日 → 盈利窗口末日 | Core EPS | P/E | 最终计算输出 USD/股Calculated output USD/share最終計算輸出 USD/股 |
|---|---|---|---|---|
| 0M | 2026-10-06 → 2027-10-06 | 13.86619 | 17.25348× | $239.24 |
| 6M | 2027-04-06 → 2028-04-06 | 16.62886 | 17.25348× | $286.91 |
| 12M | 2027-10-06 → 2028-10-06 | 18.90702 | 17.25348× | $326.21 |
版本RA-NVDA-HORIZON-09-v0.1，固定增长半衰期4季度，市场PE不变分支。每个盈利窗口从目标日开始，覆盖未来12个月；价格＝本分支Core EPS×17.25348189。0M=$239.24是市场价格锚，不是独立基本面公允价值。Version RA-NVDA-HORIZON-09-v0.1, four-quarter growth half-life and held-market-P/E branch. Each earnings window starts on its target date and covers the next twelve months. Price equals branch Core EPS × 17.25348189. The $239.24 0M value is a market-price anchor, not independent fundamental fair value.版本RA-NVDA-HORIZON-09-v0.1，固定增長半衰期4季度，市場PE不變分支。每個盈利窗口從目標日開始，覆蓋未來12個月；價格＝本分支Core EPS×17.25348189。0M=$239.24是市場價格錨，不是獨立基本面公允價值。
H11未通过的实验输出，独立列示Separate H11 outputs from the failed experimentH11未通過的實驗輸出，獨立列示
| 期限Horizon期限 | 目标日 → 盈利窗口末日Target → earnings-window end目標日 → 盈利窗口末日 | Core EPS | P/E | 最终计算输出 USD/股Calculated output USD/share最終計算輸出 USD/股 |
|---|---|---|---|---|
| 0M | 2026-10-06 → 2027-10-06 | 15.57848 | 15.35709× | $239.24 |
| 6M | 2027-04-06 → 2028-04-06 | 22.01598 | 15.35709× | $338.10 |
| 12M | 2027-10-06 → 2028-10-06 | 31.00449 | 15.35709× | $476.14 |
H11-v0.1当前ρ=1，PE₀=15.35708570。6M=$338.10、12M=$476.14反映这一分支更强的远期增长假设，未获历史比较支持，因此不作为本轮最终参考。当前输出较高与历史预测偏低可以同时发生：每个历史起点使用自己的ρ，不能拿当前ρ回写历史。H11-v0.1 currently has rho=1 and PE₀=15.35708570. Its $338.10 6M and $476.14 12M outputs reflect stronger distant growth assumptions that are not supported by the historical comparison, so they are not the final reference. High current outputs and low historical forecasts can coexist because each past origin has its own rho; today’s rho cannot be written back into history.H11-v0.1當前ρ=1，PE₀=15.35708570。6M=$338.10、12M=$476.14反映這一分支更強的遠期增長假設，未獲歷史比較支持，因此不作為本輪最終參考。當前輸出較高與歷史預測偏低可以同時發生：每個歷史起點使用自己的ρ，不能拿當前ρ回寫歷史。
优化了什么：让增长衰减速度从经营数据学习What changes: learn growth decay from operating data優化了什麼：讓增長衰減速度從經營數據學習
H09以同季度收入同比信息估计初始增长g₀，并固定每4个季度将增长动量减半。H11只替换这一衰减速度ρ；首个指引季度、季节性方法、毛利率、费用率、股数和税率规则保持H09口径。ρ越接近1，后续增长持续越久。H09 derives initial growth g₀ from same-quarter year-over-year revenue information and halves growth momentum every four quarters. H11 replaces only that decay factor rho. The first guided quarter, seasonality method, gross margin, expense ratio, shares and tax rules retain the H09 specification. Rho closer to one sustains subsequent growth longer.H09以同季度收入同比信息估計初始增長g₀，並固定每4個季度將增長動量減半。H11只替換這一衰減速度ρ；首個指引季度、季節性方法、毛利率、費用率、股數和税率規則保持H09口徑。ρ越接近1，後續增長持續越久。
| 项目Item項目 | 定义Definition定義 |
|---|---|
| H09 | ρ₀ = 2^(−1/4) = 0.840896 |
| H11 training | x = frozen H09 g₀; y = log(actual Q2 daily revenue / actual Q1 daily revenue) − frozen seasonal difference |
| Fit | ρ = clip((Sxy + λρ₀) / (Sxx + λ), 0, 1); λ = 4Sxx/N |
| Forecast | j ≥ 2: daily revenue_j = daily revenue_(j−1) × exp(seasonal difference + g₀ρ^(j−1)) |
| Price | PE₀ = P₀ / own EPS₀; P_h = own EPS_h × PE₀ |
每个预测日只用此前原始季度起点中，连续两季实际收入均已公布的最近8个标签。x和季节差按旧起点当时的资料重建；y的分子与分母都是真实收入，不用指引充当实际收入。向H09固定ρ收缩，相当于4个先验起点；不足8个或信息量不足时明确回退H09。At each origin, use the latest eight earlier original quarterly origins whose two consecutive actual revenue quarters have both been published. Reconstruct x and seasonal differences using information available at each earlier origin. Both the numerator and denominator of y are actual revenue, not guidance. Shrink toward H09’s fixed rho with four prior-equivalent origins; fewer than eight valid labels or insufficient information explicitly falls back to H09.每個預測日只用此前原始季度起點中，連續兩季實際收入均已公佈的最近8個標籤。x和季節差按舊起點當時的資料重建；y的分子與分母都是真實收入，不用指引充當實際收入。向H09固定ρ收縮，相當於4個先驗起點；不足8個或信息量不足時明確回退H09。
价格、分析师目标价、未来实际EPS均不进入ρ拟合。历史实际收入按记录的可用日筛选，不能把当前才看到的财报送回旧起点。首季度与H09精确一致，所有后续季度及第10季度后的延展使用同一个ρ。Prices, analyst targets and future realized EPS do not enter the rho fit. Filter actual revenue by recorded availability dates; information released later cannot enter earlier origins. The first quarter exactly matches H09, and every later quarter, including extension beyond quarter ten, uses the same rho.價格、分析師目標價、未來實際EPS均不進入ρ擬合。歷史實際收入按記錄的可用日篩選，不能把當前才看到的財報送回舊起點。首季度與H09精確一致，所有後續季度及第10季度後的延展使用同一個ρ。
这是一项受阻尼趋势思想启发的自定义实验，并非标准Holt模型，也没有增加国债或美联储利率折价。每个分支用自己的EPS₀重新计算市场PE锚，不能把H09的PE乘上H11的EPS。This is a custom experiment inspired by damped trends, not a standard Holt model. It adds no Treasury or Federal Reserve discount. Each branch recalculates its market P/E anchor from its own EPS₀; H09 P/E must not be multiplied by H11 EPS.這是一項受阻尼趨勢思想啓發的自定義實驗，並非標準Holt模型，也沒有增加國債或美聯儲利率折價。每個分支用自己的EPS₀重新計算市場PE錨，不能把H09的PE乘上H11的EPS。
阻尼趋势方法背景Damped-trend background阻尼趨勢方法背景 · 滚动预测时点验证Rolling-origin validation滾動預測時點驗證
当前持续率为何达到上限Why current persistence reaches its upper bound當前持續率為何達到上限
| 项目Item項目 | 当前记录Current record當前記錄 |
|---|---|
| 有效训练标签Training labels有效訓練標籤 | 8 |
| 最早／最新保留起点Earliest / latest retained origin最早／最新保留起點 | 2024-05-30 / 2026-02-26 |
| 最新标签可用日Latest label availability最新標籤可用日 | 2026-08-27 |
| Sxx / Sxy | 0.19102340 / 0.21049642 |
| λ | 0.09551170 |
| ρ raw → effective | 1.01492580 → 1.000000 |
| g₀ | 0.17157598 |
当前未约束估计ρ=1.014926，经事先规定的[0,1]约束后为1。模型因此不再衰减趋势增长；这不是“没有增长”，也不是验证了长期高增长。最新可训练起点停在2026-02-26，其第二个实际季度于2026-08-27可用；5月和8月起点尚缺完整两季标签。The current unconstrained rho estimate is 1.014926, clipped to one under the prespecified [0,1] constraint. Trend growth therefore no longer decays; this is neither zero growth nor evidence establishing sustained long-term growth. The latest trainable origin is 2026-02-26, with its second actual quarter available on 2026-08-27. May and August origins still lack complete two-quarter labels.當前未約束估計ρ=1.014926，經事先規定的[0,1]約束後為1。模型因此不再衰減趨勢增長；這不是“沒有增長”，也不是驗證了長期高增長。最新可訓練起點停在2026-02-26，其第二個實際季度於2026-08-27可用；5月和8月起點尚缺完整兩季標籤。
三年12个历史起点均可拟合，无回退；ρ范围为0.2534至1.0000，其中1个历史起点触及上限。这里暴露的结构局限是：一个相邻季度的增长关系，被反复用于远期盈利窗口；同比推算的g₀、季节估计误差及经营波动也可能被ρ吸收。All twelve historical origins can be fitted without fallback. Rho ranges from 0.2534 to 1.0000, with one historical origin at the upper bound. A structural limitation is that a one-quarter transition relationship is repeatedly applied to distant earnings windows. Rho may also absorb differences in the year-over-year-derived g₀, seasonal estimation error and operating noise.三年12個歷史起點均可擬合，無回退；ρ範圍為0.2534至1.0000，其中1個歷史起點觸及上限。這裏暴露的結構侷限是：一個相鄰季度的增長關係，被反覆用於遠期盈利窗口；同比推算的g₀、季節估計誤差及經營波動也可能被ρ吸收。
ρ=1首次出现于2026年8月起点，其6M、12M价格结果尚未到期，当前高持续率没有成熟价格表现可验证。滚动样本的退出与进入会改变估计，不能将这次触顶归因为最新一季突然加速。Rho first reaches one at the August 2026 origin, whose 6M and 12M prices have not matured. There are no mature price outcomes validating this high-persistence state. Rolling observations entering and leaving the fit can change the estimate; the bound hit cannot be attributed to sudden acceleration in the latest quarter.ρ=1首次出現於2026年8月起點，其6M、12M價格結果尚未到期，當前高持續率沒有成熟價格表現可驗證。滾動樣本的退出與進入會改變估計，不能將這次觸頂歸因為最新一季突然加速。
这些是模型结构与结果所提示的解释，不是已被因果验证的失败归因。最近8个标签与4个先验的设定在评分前冻结，本轮没有看到结果后再调窗口、上限或参数。These explanations are suggested by the model structure and results, not causally established failure attribution. The eight-label window and four-origin prior were frozen before scoring. No window, bound or parameter is retuned after observing the results.這些是模型結構與結果所提示的解釋，不是已被因果驗證的失敗歸因。最近8個標籤與4個先驗的設定在評分前凍結，本輪沒有看到結果後再調窗口、上限或參數。
只评价过去三年Evaluate only the past three years只評價過去三年
按预测发起日筛选2023-10-06至2026-10-06，保留原12个季度起点，实际首尾为2023-11-22与2026-08-27。6M价格已到期10个、待到期2个；12M已到期8个、待到期4个。当前10月6日截面单列，不加入历史评分。旧于三年的资料仅可用于当时的经营训练，不给十年表现打分或设门槛。Filter forecast origins from 2023-10-06 through 2026-10-06, retaining the original twelve quarterly dates from 2023-11-22 to 2026-08-27. Six-month prices have ten mature and two pending observations; twelve-month prices have eight mature and four pending observations. The current October 6 snapshot is separate. Earlier records may support contemporaneous operating training; there is no ten-year score or gate.按預測發起日篩選2023-10-06至2026-10-06，保留原12個季度起點，實際首尾為2023-11-22與2026-08-27。6M價格已到期10個、待到期2個；12M已到期8個、待到期4個。當前10月6日截面單列，不加入歷史評分。舊於三年的資料僅可用於當時的經營訓練，不給十年表現打分或設門檻。
实际价格标签为目标日当日或之后首个交易日收盘。EPS标签需要整个未来12个月盈利窗口公布完毕，因此0M／6M／12M仅成熟7／5／3个，单列诊断，不与价格成熟样本混用。0M是市场锚的恒等式，不计算独立价格准确率。Actual-price labels use the first trading close on or after the target date. EPS labels require publication of the entire next-twelve-month earnings window, leaving only 7 / 5 / 3 mature observations for 0M / 6M / 12M. EPS is diagnostic and is not confused with price maturity. The 0M market-anchor identity is not scored as independent price accuracy.實際價格標籤為目標日當日或之後首個交易日收盤。EPS標籤需要整個未來12個月盈利窗口公佈完畢，因此0M／6M／12M僅成熟7／5／3個，單列診斷，不與價格成熟樣本混用。0M是市場錨的恆等式，不計算獨立價格準確率。
H11未能改善H09的实际股价预测H11 does not improve H09 realized-price forecastsH11未能改善H09的實際股價預測
6M股价平均误差从H09的14.33%扩大至H11的19.03%，12M从14.91%扩大至27.91%。分析师平均绝对百分比差异有所缩小，但对数误差恶化，不能据此采用H11。Mean realized-price error increases from H09’s 14.33% to H11’s 19.03% at 6M and from 14.91% to 27.91% at 12M. The mean absolute percentage gap to analysts improves, but log error worsens; this does not justify adopting H11.6M股價平均誤差從H09的14.33%擴大至H11的19.03%，12M從14.91%擴大至27.91%。分析師平均絕對百分比差異有所縮小，但對數誤差惡化，不能據此採用H11。
| 期限Horizon期限 | 模型Model模型 | N | MAPE | MALE | 中位误差Median APE中位誤差 | P90误差P90 APEP90誤差 | 平均带符号误差Mean signed error平均帶符號誤差 |
|---|---|---|---|---|---|---|---|
| 6M | H11 | 10 | 19.03% | 0.224568 | 17.46% | 35.13% | -14.35% |
| 6M | H09 | 10 | 14.33% | 0.147441 | 12.21% | 30.89% | -1.72% |
| 6M | H07 / H03 held PE | 10 | 20.83% | 0.226953 | 20.03% | 34.86% | -4.87% |
| 6M | 价格不变Unchanged price價格不變 | 10 | 21.14% | 0.253708 | 19.73% | 37.75% | -18.88% |
| 12M | H11 | 8 | 27.91% | 0.363928 | 27.96% | 45.10% | -27.91% |
| 12M | H09 | 8 | 14.91% | 0.162365 | 10.46% | 26.31% | -3.65% |
| 12M | H07 / H03 held PE | 8 | 18.95% | 0.240083 | 7.67% | 41.13% | -16.85% |
| 12M | 价格不变Unchanged price價格不變 | 8 | 32.20% | 0.420237 | 32.34% | 47.51% | -32.20% |
MAPE＝平均｜预测÷实际−1｜；MALE＝平均｜log(预测÷实际)｜，均越低越好。H11不仅平均误差更大，中位与P90误差也恶化。相同样本上，6M和12M均未通过预设筛选。不能把1−MAPE称为准确率。MAPE is mean |forecast / actual − 1|; MALE is mean |log(forecast / actual)|. Lower is better. H11 worsens mean, median and P90 errors and fails the prespecified screen at both horizons on identical observations. One minus MAPE is not an accuracy rate.MAPE＝平均｜預測÷實際−1｜；MALE＝平均｜log(預測÷實際)｜，均越低越好。H11不僅平均誤差更大，中位與P90誤差也惡化。相同樣本上，6M和12M均未通過預設篩選。不能把1−MAPE稱為準確率。
配对误差差区间与方向诊断Paired error-difference intervals and direction diagnostics配對誤差差區間與方向診斷
| 期限Horizon期限 | MALE相对H09改善MALE improvement vs H09MALE相對H09改善 | 95% CI | H11方向命中H11 direction accuracyH11方向命中 | 一直预测上涨Always-up baseline一直預測上漲 |
|---|---|---|---|---|
| 6M | -52.31% | [-0.01280, 0.14506] | 80.00% | 90.00% |
| 12M | -124.14% | [0.10692, 0.29621] | 87.50% | 100.00% |
四起点移动区块、5000次抽样；仅探索性诊断，未校正反复使用历史、版本或窗口选择。样本少且期限重叠，上涨阶段占主导。Use four-origin moving blocks and 5,000 draws. These are exploratory diagnostics without adjustment for repeated historical use or model/window selection. Samples are small, horizons overlap and rising markets dominate.四起點移動區塊、5000次抽樣；僅探索性診斷，未校正反覆使用歷史、版本或窗口選擇。樣本少且期限重疊，上漲階段佔主導。
三年分析师比较：改善并不一致Three-year analyst comparison: mixed results三年分析師比較：改善並不一致
模型12M(t)与同日StockAnalysis All Analysts平均一年目标价(t)配对，共12点。这里只衡量与分析师预期的差异；不是未来股价误差，也不把分析师共识视为真实公允价值。Compare model 12M(t) with the same-date StockAnalysis All Analysts average one-year target(t), for twelve matched observations. This measures distance from analyst expectations, not realized-price error, and does not treat consensus as true fair value.模型12M(t)與同日StockAnalysis All Analysts平均一年目標價(t)配對，共12點。這裏只衡量與分析師預期的差異；不是未來股價誤差，也不把分析師共識視為真實公允價值。
| 模型Model模型 | N | 平均绝对差异Mean absolute gap平均絕對差異 | MALE | 中位差异Median gap中位差異 | 平均带符号差异Mean signed gap平均帶符號差異 | ±10% | ±20% |
|---|---|---|---|---|---|---|---|
| H11 | 12 | 14.75% | 0.154534 | 14.54% | -7.52% | 5/12 | 9/12 |
| H09 | 12 | 16.86% | 0.146112 | 8.81% | 13.24% | 7/12 | 9/12 |
| H07Reference | 12 | 22.64% | 0.222992 | 20.79% | 9.76% | 3/12 | 5/12 |
H11平均绝对百分比差异从16.86%降至14.75%，但MALE从0.146112升至0.154534，中位差异从8.81%升至14.54%。平均带符号差异从+13.24%变为−7.52%。不能只挑改善的一项来宣布胜出。H11 reduces the mean absolute percentage gap from 16.86% to 14.75%, but MALE increases from 0.146112 to 0.154534 and the median gap from 8.81% to 14.54%. The mean signed gap shifts from +13.24% to −7.52%. Selecting only the improving metric would not establish success.H11平均絕對百分比差異從16.86%降至14.75%，但MALE從0.146112升至0.154534，中位差異從8.81%升至14.54%。平均帶符號差異從+13.24%變為−7.52%。不能只挑改善的一項來宣佈勝出。
沿用2026-10-07图表悬停记录；原始历史发布时间、修订和拆股沿革尚未独立核验。目标价未用于拟合，仅公开聚合比较，原始Pro数据保留内部。Use the chart-tooltip records captured on 2026-10-07. Original historical publication times, revisions and split lineage remain unverified. Targets are not used in fitting; only aggregate comparisons are public and raw Pro data remains internal.沿用2026-10-07圖表懸停記錄；原始歷史發佈時間、修訂和拆股沿革尚未獨立核驗。目標價未用於擬合，僅公開聚合比較，原始Pro數據保留內部。
StockAnalysis平均目标价图表StockAnalysis average-target chartStockAnalysis平均目標價圖表
EPS诊断同样不支持替换EPS diagnostics also do not support replacementEPS診斷同樣不支持替換
| 期限Horizon期限 | 模型Model模型 | N | MAPE | MALE |
|---|---|---|---|---|
| 0M | H11EPS | 7 | 18.53% | 0.213064 |
| 0M | H09EPS | 7 | 10.99% | 0.117832 |
| 6M | H11EPS | 5 | 33.43% | 0.410963 |
| 6M | H09EPS | 5 | 16.61% | 0.157820 |
| 12M | H11EPS | 3 | 51.66% | 0.731910 |
| 12M | H09EPS | 3 | 20.43% | 0.200379 |
这些是同一三年起点中的已成熟盈利窗口，不是十年EPS结果。样本只有7／5／3个，只作诊断；各期限H11误差均高于H09。These are mature earnings windows within the same three-year origin cohort, not ten-year EPS results. With only 7 / 5 / 3 observations, they are diagnostic; H11 errors exceed H09 at every horizon.這些是同一三年起點中的已成熟盈利窗口，不是十年EPS結果。樣本只有7／5／3個，只作診斷；各期限H11誤差均高於H09。
保留H09，并明确下一轮问题Retain H09 and define the next research question保留H09，並明確下一輪問題
H11未通过三年股价与共识联合筛选。按用户选择，本轮主表以H09作为继续研究的参考；H11仅保留实验诊断。参考身份由此前H07明确切换至H09，不代表正式Champion晋升。H11 fails the joint three-year price and consensus screen. Following the user’s selection, H09 is this round’s headline research reference; H11 remains experimental diagnostics. This explicitly changes the earlier H07 reference to H09 without promoting a formal Champion.H11未通過三年股價與共識聯合篩選。按用户選擇，本輪主表以H09作為繼續研究的參考；H11僅保留實驗診斷。參考身份由此前H07明確切換至H09，不代表正式Champion晉升。
评分前固定的实际价格门槛为：6M和12M各自MALE至少改善5%、MAPE下降、中位和P90恶化不超过10%；共识的MALE、平均绝对百分比差异和绝对平均带符号差异均不恶化。H11的两项价格筛选和共识筛选均失败。Before scoring, each price horizon must improve MALE by at least 5%, reduce MAPE and limit median/P90 worsening to 10%. Consensus MALE, mean absolute percentage gap and absolute mean signed gap must all be nonworsening. H11 fails both price-horizon screens and the consensus screen.評分前固定的實際價格門檻為：6M和12M各自MALE至少改善5%、MAPE下降、中位和P90惡化不超過10%；共識的MALE、平均絕對百分比差異和絕對平均帶符號差異均不惡化。H11的兩項價格篩選和共識篩選均失敗。
下一轮值得研究的是：用与6M、12M盈利窗口相匹配的累计经营结果评价增长衰减，而不是把一个相邻季度关系反复外推。先确认成熟标签数量，再冻结单一方案及H09对照；如样本不足，应保留H09并积累新预测。该方向尚未实施，也不能保证提高准确率。A useful next question is whether growth decay can be assessed against cumulative operating outcomes aligned with the 6M and 12M earnings windows, instead of repeatedly extrapolating one adjacent-quarter relationship. Check mature-label coverage before freezing one candidate and its H09 control. If data is insufficient, retain H09 and accumulate new forecasts. This direction is not implemented and offers no guarantee of improved accuracy.下一輪值得研究的是：用與6M、12M盈利窗口相匹配的累計經營結果評價增長衰減，而不是把一個相鄰季度關係反覆外推。先確認成熟標籤數量，再凍結單一方案及H09對照；如樣本不足，應保留H09並積累新預測。該方向尚未實施，也不能保證提高準確率。
独立计算检查2189项与信息隔离检查649项通过；审计未读取评分表。旧研究和分析师原始记录8985个文件保持原样。通过这些检查证明实现与约束一致，不证明模型更准确。Independent calculation checks (2189) and information-isolation checks (649) pass; the audit did not read score tables. All 8985 older research and benchmark files remain unchanged. These checks establish implementation consistency, not improved predictive performance.獨立計算檢查2189項與信息隔離檢查649項通過；審計未讀取評分表。舊研究和分析師原始記錄8985個文件保持原樣。通過這些檢查證明實現與約束一致，不證明模型更準確。
审计还修复了一处无效标签检查顺序：先排除缺失、零或负收入等无效实际季度，再重建旧预测。429项逐值核验确认有效数据的训练、路径和价格与评分前结果完全一致；没有改变参数或重跑评分。The audit also repaired invalid-label check ordering: exclude missing, zero or negative actual revenue before reconstructing older forecasts. 429 exact comparisons confirm that valid-data training, paths and prices remain identical to the scored freeze. Parameters and scores are unchanged.審計還修復了一處無效標籤檢查順序：先排除缺失、零或負收入等無效實際季度，再重建舊預測。429項逐值核驗確認有效數據的訓練、路徑和價格與評分前結果完全一致；沒有改變參數或重跑評分。
本次公开研究页面，不部署H11为正式模型。三年窗口已反复用于开发，季度样本少且相互重叠；财务历史按可用日重建，原始不可变PIT版本尚不完整。独立样本外验证仍未完成。This publication records research and does not deploy H11 as a formal model. The three-year window has been repeatedly used for development, quarterly samples are small and overlapping, and financial history is reconstructed by availability date without complete immutable original PIT vintages. Independent OOS validation remains incomplete.本次公開研究頁面，不部署H11為正式模型。三年窗口已反覆用於開發，季度樣本少且相互重疊；財務歷史按可用日重建，原始不可變PIT版本尚不完整。獨立樣本外驗證仍未完成。
Protocol frozen UTC 2026-10-07T13:27:35.101760+00:00 · SHA256 215de0f083a37d2973dc63de47765536ab84c9e5938e2153c05fe5c4e004b59a
Record SHA256 c565c74a1b03e6ee5da167d52a64654d84182e2e280d19b929ab462617746ec0
H07–H10三年比较H07–H10 three-year comparisonH07–H10三年比較 · H10原研究H10 original studyH10原研究 · H09原研究H09 original studyH09原研究 · 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