NVDA / H16 / 2026-10-13
H16：近期残差权重与分期限验证H16: recent residual weights and horizon-specific validationH16：近期殘差權重與分期限驗證
研究日期2026-10-13；输入与实际完整常规收盘2026-10-12，参考价$226.32。最新财报FY2027Q2期末2026-07-26、发布2026-08-26、可得2026-08-27；国债观察2026-10-09。本轮不是10月13日收盘刷新。Research date 2026-10-13; inputs and actual completed regular close 2026-10-12, reference $226.32. Latest FY2027Q2: ended 2026-07-26, released 2026-08-26, available 2026-08-27; Treasury observation 2026-10-09. This is not an October 13 close refresh.研究日期2026-10-13；輸入與實際完整常規收盤2026-10-12，參考價$226.32。最新財報FY2027Q2期末2026-07-26、發佈2026-08-26、可得2026-08-27；國債觀察2026-10-09。本輪不是10月13日收盤刷新。
H16对6M、12M分别测试唯一固定时间衰减残差规则，两期都未通过原采用门槛，主表保持不变。0M没有新候选、没有新会计来源证据，继续使用H15 Fair0；6M/12M继续使用H09。分期限独立采用不等于每轮都必须改模型。H16 independently tests one fixed time-decay residual rule at 6M and 12M. Both fail the original adoption gates and all selected values stay unchanged. There is no new 0M candidate or accounting-source evidence: retain H15 Fair0 at 0M and H09 at 6M/12M. Independent horizon adoption does not require a model change every study.H16對6M、12M分別測試唯一固定時間衰減殘差規則，兩期都未通過原採用門檻，主表保持不變。0M沒有新候選、沒有新會計來源證據，繼續使用H15 Fair0；6M/12M繼續使用H09。分期限獨立採用不等於每輪都必須改模型。
最终保留主表：H15 Fair0与H09Final retained headline: H15 Fair0 and H09最終保留主表：H15 Fair0與H09
| 期限／版本Horizon / version期限／版本 | 目标日 → EPS窗口末日Target → EPS-window end目標日 → EPS窗口末日 | Core EPS | Core PE | 税率Tax税率 | USD/股USD/shareUSD/股 | 身份Role身份 |
|---|---|---|---|---|---|---|
| 0M RA-NVDA-H15-FAIR0-v0.1 | 2026-10-12 → 2027-10-12 | 13.63099 | 18.79810× | 17.00% | $256.24 | 沿用H15 Fair0；无新0M候选Retain H15 Fair0; no new 0M candidate沿用H15 Fair0；無新0M候選 |
| 6M RA-NVDA-HORIZON-09-v0.1 | 2027-04-12 → 2028-04-12 | 16.71457 | 16.21268× | 15.00% | $270.99 | 保留原H09价格预测Retain original H09 price forecast保留原H09價格預測 |
| 12M RA-NVDA-HORIZON-09-v0.1 | 2027-10-12 → 2028-10-12 | 18.97455 | 16.21268× | 15.00% | $307.63 | 保留原H09价格预测Retain original H09 price forecast保留原H09價格預測 |
EPS窗均为目标日起未来12个日历月，起含止不含；同一24.285B已报告稀释加权平均股数代理。0M税17%与独立现金FairPE来自H15；未来税15%的EPS与市场PE来自H09。Core EPS不是GAAP净利润EPS或分析师调整后EPS。EPS windows cover twelve calendar months from target, start inclusive and end exclusive, using the same 24.285B reported diluted weighted-average share proxy. 0M 17% tax and independent cash FairPE come from H15; future 15%-tax EPS and held-market PE come from H09. Core EPS is separate from GAAP net-income EPS and analyst-adjusted EPS.EPS窗均為目標日起未來12個日曆月，起含止不含；同一24.285B已報告稀釋加權平均股數代理。0M税17%與獨立現金FairPE來自H15；未來税15%的EPS與市場PE來自H09。Core EPS不是GAAP淨利潤EPS或分析師調整後EPS。
未来PE仍用$226.32÷旧税15%的EPS0 13.95945＝16.21268。不能使用税17%的新0M EPS重置未来倍数；当前6M和12M尚未兑现。Future PE remains $226.32 / old 15%-tax EPS0 13.95945 = 16.21268. The 17%-tax 0M EPS must not reset future multiples. Current 6M and 12M forecasts remain pending.未來PE仍用$226.32÷舊税15%的EPS0 13.95945＝16.21268。不能使用税17%的新0M EPS重置未來倍數；當前6M和12M尚未兑現。
唯一冻结规则：标签可用日计龄与固定先验Sole frozen rule: label-availability age and fixed priors唯一凍結規則：標籤可用日計齡與固定先驗
同期限残差r_i=ln(实际目标收盘/H09预测)。ageDays=查询日−标签可用日，w_i=2^(-ageDays/365.25)，W=sumw，alpha_h=sum(w_i×r_i)/(W+4)。至少12个时间合格原始训练起点才启用；原EPS与PE起点不变，目标价=H09价×exp(alpha_h)。四个单位权重、零残差先验固定，没有权重归一化、rolling窗口、剪尾、alpha上下界、网格或结果后调参。Same-horizon residual r_i=ln(actual target close/H09 forecast). ageDays=query date−label availability, w_i=2^(-ageDays/365.25), W=sumw, alpha_h=sum(w_i×r_i)/(W+4). At least twelve time-eligible raw training origins are required. Original EPS and PE anchor stay unchanged, and target=H09 price×exp(alpha_h). Four unit-weight zero-residual priors are fixed, without normalization, rolling windows, trimming, alpha bounds, grids or post-result tuning.同期限殘差r_i=ln(實際目標收盤/H09預測)。ageDays=查詢日−標籤可用日，w_i=2^(-ageDays/365.25)，W=sumw，alpha_h=sum(w_i×r_i)/(W+4)。至少12個時間合格原始訓練起點才啓用；原EPS與PE起點不變，目標價=H09價×exp(alpha_h)。四個單位權重、零殘差先驗固定，沒有權重歸一化、rolling窗口、剪尾、alpha上下界、網格或結果後調參。
365.25天半衰期约为12个月／四个季度，是计算前声明的研究假设，不是数据估计的最优长度，也不是NIST建议的NVDA参数。每个标签在目标日后首交易日常规收盘后下一日可用，并要求可用日严格早于查询起点。原40季度起点只作当时训练；评价只做原三年12起点。The 365.25-day half-life, roughly twelve months/four quarters, is a predeclared research assumption, not an estimated optimum or a NIST-recommended NVDA parameter. Each label is the first target-trading-day regular close, available the next day and strictly before the query origin. Original forty quarterly origins supply as-of training only; evaluation covers the twelve original origins within three years.365.25天半衰期約為12個月／四個季度，是計算前聲明的研究假設，不是數據估計的最優長度，也不是NIST建議的NVDA參數。每個標籤在目標日後首交易日常規收盤後下一日可用，並要求可用日嚴格早於查詢起點。原40季度起點只作當時訓練；評價只做原三年12起點。
已到期应可用行缺H09预测、首交易日收盘或来源，该期限不可估计；即使w很小也必须保留和核验，不跳过、插值、改交易日或缩短训练。真实标签时间不足12才可alpha=0精确H09身份。不同期限标签不共享，分析师和实际未来EPS不拟合。Missing required eligible H09 forecasts, first-target-trading-day closes or sources make the horizon not estimable. Even very small-weight rows must remain verified, without skipping, interpolation, trading-date substitution or shortened training. Only genuine time insufficiency below twelve labels permits alpha=0 exact H09 identity. Horizon labels are not shared; analysts and actual future EPS do not enter fitting.已到期應可用行缺H09預測、首交易日收盤或來源，該期限不可估計；即使w很小也必須保留和核驗，不跳過、插值、改交易日或縮短訓練。真實標籤時間不足12才可alpha=0精確H09身份。不同期限標籤不共享，分析師和實際未來EPS不擬合。
总训练N、权重总和与信息年龄分别披露Disclose raw N, weight mass and information ages separately總訓練N、權重總和與信息年齡分別披露
| 期限Horizon期限 | 原始合格NRaw eligible N原始合格N | W=sumw | ESS | 加权标签龄天Weighted label age days加權標籤齡天 | 加权预测起点龄天Weighted forecast-origin age days加權預測起點齡天 | 最近标签距查询天Latest-label gap days最近標籤距查詢天 |
|---|---|---|---|---|---|---|
| 6M | 38 | 5.76044 | 11.56339 | 523.52 | 707.37 | 46 |
| 12M | 36 | 5.76456 | 11.52058 | 520.38 | 886.71 | 44 |
| 期限Horizon期限 | 最近标签可用日Latest label available最近標籤可用日 | 最早标签可用日Oldest label available最早標籤可用日 | 标签信息龄范围天Label age range days標籤信息齡範圍天 |
|---|---|---|---|
| 6M | 2026-08-27 | 2017-05-24 | 46–3428 |
| 12M | 2026-08-29 | 2017-11-25 | 44–3243 |
| 期限Horizon期限 | 加权残差均值μWeighted residual mean μ加權殘差均值μ | 先验收缩W/(W+4)Prior shrink W/(W+4)先驗收縮W/(W+4) | alpha | 残差倍率Residual multiplier殘差倍率 | 三年W范围Three-year W range三年W範圍 | 三年ESS范围Three-year ESS range三年ESS範圍 |
|---|---|---|---|---|---|---|
| 6M | 0.04482874 | 0.59018232 | 0.02645713 | 1.02681023× | 5.259–6.257 | 11.295–11.560 |
| 12M | 0.17995808 | 0.59035518 | 0.10623918 | 1.11208784× | 5.216–6.243 | 11.099–11.519 |
ESS=(sumw)²/sum(w²)只描述权重集中度，不处理时间相关和窗口重叠，不是独立样本数或自由度。rawN38/36也不是38/36个独立信息单位。加权标签年龄仍约520天，预测起点更早；12M最近兑现误差仍来自一年或更早的原始预测，无法消除成熟延迟。ESS=(sumw)²/sum(w²) describes weight concentration only. It does not address serial dependence or overlapping windows and is not an independent sample count or degrees of freedom. Raw N38/36 does not represent thirty-eight/thirty-six independent information units. Weighted labels are still about 520 days old and forecast origins older; recently realized 12M errors still originate a year or more earlier, so maturation delay remains.ESS=(sumw)²/sum(w²)只描述權重集中度，不處理時間相關和窗口重疊，不是獨立樣本數或自由度。rawN38/36也不是38/36個獨立信息單位。加權標籤年齡仍約520天，預測起點更早；12M最近兑現誤差仍來自一年或更早的原始預測，無法消除成熟延遲。
alpha=μweighted×W/(W+4)。衰减同时改变残差平均与相对先验收缩，因此即使误差改善也不能全部归因于近期PE状态。固定标签但时间经过时，全部w同倍下降，固定prior使alpha继续向0回归；这是预先声明的遗忘机制，不是新增经营或宏观消息。alpha=μweighted×W/(W+4). Decay changes both residual averages and relative prior shrinkage, so gains cannot all be attributed to recent PE states. As time passes without new labels, all weights decay together and the fixed prior moves alpha toward zero. This is predeclared forgetting, not new operating or macro news.alpha=μweighted×W/(W+4)。衰減同時改變殘差平均與相對先驗收縮，因此即使誤差改善也不能全部歸因於近期PE狀態。固定標籤但時間經過時，全部w同倍下降，固定prior使alpha繼續向0迴歸；這是預先聲明的遺忘機制，不是新增經營或宏觀消息。
未采用候选仅供诊断Rejected candidates are diagnostic only未採用候選僅供診斷
| 期限／版本Horizon / version期限／版本 | 目标日 → EPS窗口末日Target → EPS-window end目標日 → EPS窗口末日 | Core EPS | Core PE | 税率Tax税率 | USD/股USD/shareUSD/股 | 身份Role身份 |
|---|---|---|---|---|---|---|
| 6M RA-NVDA-H16-RECENCY-v0.1 | 2027-04-12 → 2028-04-12 | 16.71457 | 16.64734× | 15.00% | $278.25 | 未采用H16时间衰减候选Rejected H16 time-decay candidate未採用H16時間衰減候選 |
| 12M RA-NVDA-H16-RECENCY-v0.1 | 2027-10-12 → 2028-10-12 | 18.97455 | 18.02992× | 15.00% | $342.11 | 未采用H16时间衰减候选Rejected H16 time-decay candidate未採用H16時間衰減候選 |
0M没有本轮候选，原H15 Fair0只在保留主表显示。未来候选$278.25／$342.11仅列研究表，主卡仍$270.99／$307.63，不把更高目标误认为改善。There is no 0M candidate in this study; original H15 Fair0 appears only in the retained headline. Future $278.25 / $342.11 values appear only in study tables. Headline targets remain $270.99 / $307.63; higher targets do not establish improvement.0M沒有本輪候選，原H15 Fair0只在保留主表顯示。未來候選$278.25／$342.11僅列研究表，主卡仍$270.99／$307.63，不把更高目標誤認為改善。
三年价格误差：H09与冻结H15控制分开Three-year price errors: H09 and frozen H15 control三年價格誤差：H09與凍結H15控制分開
| 期限Horizon期限 | 模型Model模型 | 成熟／待定NMature / pending N成熟／待定N | MAPE | MALE | 中位APEMedian APE中位APE | P90 APE | 最大APEMaximum APE最大APE | 有符号偏差Signed bias有符號偏差 |
|---|---|---|---|---|---|---|---|---|
| 6M | H16 | 10 / 2 | 16.45% | 0.15304 | 14.06% | 30.71% | 49.28% | 7.55% |
| 6M | H09 | 10 / 2 | 14.33% | 0.14744 | 12.21% | 30.89% | 32.47% | -1.72% |
| 6M | H15Frozen | 10 / 2 | 17.42% | 0.15934 | 14.85% | 31.07% | 50.77% | 10.75% |
| 6M | PriceUnchanged | 10 / 2 | 21.14% | 0.25371 | 19.73% | 37.75% | 48.69% | -18.88% |
| 12M | H16 | 8 / 4 | 35.15% | 0.30421 | 30.83% | 58.80% | 67.47% | 27.80% |
| 12M | H09 | 8 / 4 | 14.91% | 0.16237 | 10.46% | 26.31% | 39.87% | -3.65% |
| 12M | H15Frozen | 8 / 4 | 34.13% | 0.29608 | 32.63% | 52.97% | 59.86% | 27.68% |
| 12M | PriceUnchanged | 8 / 4 | 32.20% | 0.42024 | 32.34% | 47.51% | 65.68% | -32.20% |
原12季度起点全部保留，6M成熟10/待2，12M成熟8/待4。每h十二起点全active、无fallback，full与active误差相同，并非两次独立验证。目标标签为首目标交易日常规收盘，不用最高价；MAPE是误差不是准确率。All twelve quarterly origins are retained: ten mature/two pending at 6M and eight mature/four pending at 12M. All twelve are active at each horizon, without fallback, so full and active errors agree and are not independent validations. Labels use first-target-trading-day regular closes, not highs. MAPE is error, not an accuracy rate.原12季度起點全部保留，6M成熟10/待2，12M成熟8/待4。每h十二起點全active、無fallback，full與active誤差相同，並非兩次獨立驗證。目標標籤為首目標交易日常規收盤，不用最高價；MAPE是誤差不是準確率。
H16的6M MAPE16.45%低于H15的17.42%，但仍高于H09的14.33%；12M35.15%反而高于H15的34.13%及H09的14.91%。不能以6M对失败控制略好，称通过对所保留模型的门槛。H15只作冻结失败控制，没有重新获得采用资格。H16 6M MAPE16.45% is below H15 17.42%, but still above H09 14.33%. Its 12M35.15% exceeds both H15 34.13% and H09 14.91%. A small 6M gain over a rejected control does not pass gates against the retained model. H15 remains a frozen rejected control without renewed adoption eligibility.H16的6M MAPE16.45%低於H15的17.42%，但仍高於H09的14.33%；12M35.15%反而高於H15的34.13%及H09的14.91%。不能以6M對失敗控制略好，稱通過對所保留模型的門檻。H15只作凍結失敗控制，沒有重新獲得採用資格。
历史评价截止2026-10-12，原价格档案截至2026-10-06，全部成熟目标更早，没有新增成熟样本；本轮是新权重规则重算既有面板。Historical evaluation ends 2026-10-12; the original price archive ends 2026-10-06. All mature targets are earlier and no new outcomes mature. This study evaluates a new weighting rule on the existing panel.歷史評價截止2026-10-12，原價格檔案截至2026-10-06，全部成熟目標更早，沒有新增成熟樣本；本輪是新權重規則重算既有面板。
原固定时间组门槛不放宽Original fixed-group gates remain unchanged原固定時間組門檻不放寬
| 期限Horizon期限 | 固定组Fixed group固定組 | 成熟NMature N成熟N | H16 MAPE | H09 MAPE | H16 MALE | H09 MALE |
|---|---|---|---|---|---|---|
| 6M | 1 | 4 | 16.59% | 13.79% | 0.16181 | 0.15248 |
| 6M | 2 | 4 | 21.89% | 19.58% | 0.19537 | 0.19109 |
| 6M | 3 | 2 | 5.29% | 4.93% | 0.05086 | 0.05006 |
| 12M | 1 | 4 | 42.20% | 19.11% | 0.36942 | 0.21364 |
| 12M | 2 | 4 | 28.11% | 10.72% | 0.23900 | 0.11109 |
| 12M | 3 | 0 | 待验证Pending validation待驗證 | 待验证Pending validation待驗證 | 待验证Pending validation待驗證 | 待验证Pending validation待驗證 |
沿用每四起点一组的原三组。成熟N≥2要求MALE/MAPE不恶化，不足N只记待验证；active分组与full一致。新算法不允许放宽组、尾部或偏差条件，时间组也不等于已识别的经济regime。Retain the original three four-origin groups. Groups with at least two mature outcomes must not worsen MALE/MAPE; smaller groups remain pending. Active groups agree with full. The new algorithm does not relax group, tail or bias rules, and time partitions are not identified economic regimes.沿用每四起點一組的原三組。成熟N≥2要求MALE/MAPE不惡化，不足N只記待驗證；active分組與full一致。新算法不允許放寬組、尾部或偏差條件，時間組也不等於已識別的經濟regime。
12M分析师外部比较External analyst 12M comparison12M分析師外部比較
| 比较Comparison比較 | 模型Model模型 | N | MAPE | MALE | 中位APEMedian APE中位APE | P90 APE |
|---|---|---|---|---|---|---|
| 同日平均12M目标差Same-date average 12M target disagreement同日平均12M目標差 | H16 | 12 | 43.70% | 0.34008 | 36.42% | 87.06% |
| 同日平均12M目标差Same-date average 12M target disagreement同日平均12M目標差 | H09 | 12 | 16.86% | 0.14611 | 8.81% | 35.93% |
| 共同成熟实际12M误差Common mature realized 12M price error共同成熟實際12M誤差 | H16 | 8 | 35.15% | 0.30421 | 30.83% | 58.80% |
| 共同成熟实际12M误差Common mature realized 12M price error共同成熟實際12M誤差 | H09 | 8 | 14.91% | 0.16237 | 10.46% | 26.31% |
| 共同成熟实际12M误差Common mature realized 12M price error共同成熟實際12M誤差 | analyst | 8 | 18.39% | 0.22916 | 13.01% | 40.70% |
同日12M平均目标N12：H16目标差MAPE43.70%，H09 16.86%。共同成熟N8/待4：实际12M误差H16 35.15%、H09 14.91%、分析师18.39%。分析师不拟合、不是要求必须贴共识的硬门槛；没有6M分析师历史标签，不折算12M目标。Same-date average 12M targets have N12: H16 disagreement MAPE43.70%, H09 16.86%. Common mature N8/four pending: actual 12M errors are H16 35.15%, H09 14.91%, analysts18.39%. Analysts do not train the model and consensus proximity is not a hard gate. Historical 6M analyst labels are unavailable; 12M targets are not converted.同日12M平均目標N12：H16目標差MAPE43.70%，H09 16.86%。共同成熟N8/待4：實際12M誤差H16 35.15%、H09 14.91%、分析師18.39%。分析師不擬合、不是要求必須貼共識的硬門檻；沒有6M分析師歷史標籤，不折算12M目標。
StockAnalysis分析师预测来源StockAnalysis analyst forecast sourceStockAnalysis分析師預測來源
两期独立拒绝；0M没有本轮采用测试Independent future rejection; no new 0M adoption test兩期獨立拒絕；0M沒有本輪採用測試
| 期限Horizon期限 | 本轮是否测试Tested this study本輪是否測試 | 失败／总门槛Failed / total gates失敗／總門檻 | 决定Decision決定 |
|---|---|---|---|
| 0 | 否；原样保留No; retained unchanged否；原樣保留 | 不适用Not applicable不適用 | 沿用H15 Fair0Retain H15 Fair0沿用H15 Fair0 |
| 6M | 是Yes是 | 22 / 27 | 拒绝，沿用H09Reject; retain H09拒絕，沿用H09 |
| 12M | 是Yes是 | 20 / 23 | 拒绝，沿用H09Reject; retain H09拒絕，沿用H09 |
沿用H15全部逐期门槛：完整及active MALE至少相对改善5%、MAPE下降，median/P90/max和绝对有符号偏差不恶化，固定成熟组不恶化、active成熟N≥4与当前active、完整覆盖和独立实现审查。6M失败22/27、12M失败20/23，包含相同full/active屏幕，不是独立实验数。Retain all H15 horizon gates: at least 5% relative MALE improvement and lower MAPE in full and active cohorts; nonworse median/P90/max and absolute signed bias; nonworse mature fixed groups; at least four mature active origins, active current output, full coverage and independent implementation review. 6M fails22/27 and12M20/23, including repeated full/active screens, not independent experiment counts.沿用H15全部逐期門檻：完整及active MALE至少相對改善5%、MAPE下降，median/P90/max和絕對有符號偏差不惡化，固定成熟組不惡化、active成熟N≥4與當前active、完整覆蓋和獨立實現審查。6M失敗22/27、12M失敗20/23，包含相同full/active屏幕，不是獨立實驗數。
0M没有新规则，只通过留存恒等检查确认H15输出不变。本轮不得称又一次0M来源采用、提高Fair Value准确度或更新三期模型。若以后某一期真正通过预设证据，可独立更换；本轮未来两期均保留。0M has no new rule; retained-identity checks confirm unchanged H15 output. This study does not claim renewed 0M source adoption, better Fair Value accuracy or three updated models. A later horizon may be replaced independently if it passes frozen evidence; both future horizons are retained here.0M沒有新規則，只通過留存恆等檢查確認H15輸出不變。本輪不得稱又一次0M來源採用、提高Fair Value準確度或更新三期模型。若以後某一期真正通過預設證據，可獨立更換；本輪未來兩期均保留。
留存0M：既有指导税与现金假设Retained 0M: existing guided-tax cash assumptions留存0M：既有指導税與現金假設
H15 Fair0继续用公司FY2027 GAAP归一化税率指导16%–18%中点17%，同步作用Core EPS、现金收益、债务税盾、WACC与终值。H09经营美元与24.285B份数、H13 Base资本现金规则不变；指导延续到60M和终值仍是研究假设。H15 Fair0 retains the 17% midpoint of FY2027 normalized GAAP tax guidance16%–18%, applied to Core EPS, cash returns, debt tax shields, WACC and terminal value. H09 operating dollars and24.285B shares, plus H13 Base capital/cash rules, remain unchanged. Persistence through sixty months and terminal remains a research assumption.H15 Fair0繼續用公司FY2027 GAAP歸一化税率指導16%–18%中點17%，同步作用Core EPS、現金收益、債務税盾、WACC與終值。H09經營美元與24.285B份數、H13 Base資本現金規則不變；指導延續到60M和終值仍是研究假設。
来源/会计一致性的既有改进不证明FV准确。经济现金为模型估计，资产融资、投资专属税负和期权/RSU索偿未全映射；它不是严格FCFF、完整清算值或已证明保守下限。The earlier source/accounting improvement does not establish FV accuracy. Economic cash is modeled; asset financing, investment-specific taxes and option/RSU claims remain incompletely mapped. This is not strict FCFF, complete liquidation value or a proven conservative floor.來源/會計一致性的既有改進不證明FV準確。經濟現金為模型估計，資產融資、投資專屬税負和期權/RSU索償未全映射；它不是嚴格FCFF、完整清算值或已證明保守下限。
H15原0M来源质量采用与敏感性Original H15 0M source-quality adoption and sensitivitiesH15原0M來源質量採用與敏感性
实现验证与反复研究边界Implementation verification and repeated-research limits實現驗證與反覆研究邊界
独立审计PASS，32,597项检查、0失败。核对资格/缺洞、时间权重和ESS、闭式拟合、未归一化先验、日期推进遗忘、期限与未来标签隔离、全部统计门槛、0M留存恒等和旧文件保护。Independent audit passed 32,597 checks with zero failures. It verifies eligibility/missing rows, time weights and ESS, closed-form fits, unnormalized priors, time-advance forgetting, horizon/future-label isolation, statistical gates, retained 0M identity and old-file preservation.獨立審計PASS，32,597項檢查、0失敗。核對資格/缺洞、時間權重和ESS、閉式擬合、未歸一化先驗、日期推進遺忘、期限與未來標籤隔離、全部統計門檻、0M留存恆等和舊文件保護。
实现复算不认证价格预测准确。H16在看过H15失败后提出，已存在研究选择乐观偏差；冻结本轮不把反复使用的三年面板变成独立OOS。季度起点及盈利窗重叠，ESS不解除相关性；原可用日期重建不是完整不可变PIT，没有正式Champion。Implementation reproduction does not certify price accuracy. H16 follows observed H15 failure and is exposed to research-selection optimism. Freezing this run does not make the reused three-year panel independent OOS. Quarterly origins and earnings windows overlap; ESS does not remove dependence. Availability-date reconstruction is not complete immutable PIT, and formal Champion status is absent.實現復算不認證價格預測準確。H16在看過H15失敗後提出，已存在研究選擇樂觀偏差；凍結本輪不把反覆使用的三年面板變成獨立OOS。季度起點及盈利窗重疊，ESS不解除相關性；原可用日期重建不是完整不可變PIT，沒有正式Champion。
网站仅公开自有派生估值、方法、聚合误差和审计摘要。逐起点价格、权重/残差训练记录、StockAnalysis Pro目标及现金日账留内部。所有旧模型和失败状态原样保存。The website publishes original derived values, methods, aggregate errors and audit summaries only. Origin-level prices, weight/residual training rows, StockAnalysis Pro targets and cash ledgers remain internal. All old models and rejection states are preserved.網站僅公開自有派生估值、方法、聚合誤差和審計摘要。逐起點價格、權重/殘差訓練記錄、StockAnalysis Pro目標及現金日賬留內部。所有舊模型和失敗狀態原樣保存。
来源与永久记录Sources and permanent records來源與永久記錄
NVIDIA FY2027Q2 GAAP tax guidanceNVIDIA FY2027Q2 GAAP tax guidanceNVIDIA FY2027Q2 GAAP tax guidance
NVIDIA original financial reportsNVIDIA original financial reportsNVIDIA original financial reports
Treasury daily 10Y ratesTreasury daily 10Y ratesTreasury daily 10Y rates
RA-NVDA-HORIZON-16-v0.1 · Protocol SHA256 a1c5387822d964d7b4aa088bb1ad45f22f7cb583da3908027e43dbb46171ddee
Public aggregate SHA256 e7bd9e23d911b5bce1f4cb9cc8de499aabed174ca6f3a98ca49c59b339f49170
NVDA当前保留主表Current retained NVDA valuesNVDA當前保留主表 · H15原分期采用与拒绝Original H15 horizon adoption and rejectionH15原分期採用與拒絕 · 所有NVDA版本All NVDA versions所有NVDA版本