MODEL FAMILY / v0.1.0 / 2026-09-26
AI 基础设施与半导体AI infrastructure & semiconductorsAI 基礎設施與半導體
分项盈利桥 + 目标日 NTM EPS × Exit P/E；软件业务、并购调整与净债务另设适配项。Segment earnings bridge + target-date NTM EPS × exit P/E, with separate adapters for software, acquisition adjustments and net debt.分項盈利橋 + 目標日 NTM EPS × Exit P/E；軟件業務、併購調整與淨債務另設適配項。
需求与交付、客户资本开支兑现、产品周期、毛利率、供给与出口限制。Demand and delivery, customer capex execution, product cycles, gross margins, supply and export restrictions.需求與交付、客户資本開支兑現、產品週期、毛利率、供給與出口限制。
同类模型，不是同一套数字Shared models, not identical numbers同類模型，不是同一套數字
先按商业模式归类，再按增长加速、正常化或压力阶段选择假设。业务适配层保存分部映射、会计口径、资本强度与特有风险；同一模型族不共用同一个增长率、利润率或 P/E。阶段必须由带日期的证据判定，当前尚未给五家公司贴上已验证的阶段标签。Group by business economics, then select assumptions for acceleration, normalization or stress. Company adapters retain segment mappings, accounting, capital intensity and specific risks. A shared family does not imply identical growth, margins or P/E. Stages require dated evidence; no validated stage labels have yet been assigned to these five companies.先按商業模式歸類，再按增長加速、正常化或壓力階段選擇假設。業務適配層保存分部映射、會計口徑、資本強度與特有風險；同一模型族不共用同一個增長率、利潤率或 P/E。階段必須由帶日期的證據判定，當前尚未給五家公司貼上已驗證的階段標籤。
统一计算链与口径护栏Common calculation chain and basis controls統一計算鏈與口徑護欄
Revenueᵢ = Base revenueᵢ × (1 + growthᵢ) Operating profit = Σ(Revenueᵢ × operating marginᵢ) − corporate costs Pretax profit = Operating profit + recurring non-operating income − net interest Net income attributable = Pretax profit − tax − non-controlling interests Diluted EPS = Net income attributable ÷ diluted weighted-average shares Target price = EPS for the 12 months AFTER target date × Exit P/E
这是框架公式，不是已经填妥的预测。税额优先按可核验税率／税项建模；亏损税盾不自动计入。费用已在分部利润中包含时不能再减一次。当前NTM与目标日NTM不得混用；缺少季度预测时，财年按日历重叠加权仅是有季节性局限的近似。These are framework equations, not completed forecasts. Model tax from verified rates/items; do not automatically assume tax shields on losses. Avoid deducting costs already included in segment profit twice. Current NTM and target-date NTM are distinct. Without quarterly forecasts, calendar-overlap fiscal-year blending is only an approximation with seasonality limits.這是框架公式，不是已經填妥的預測。税額優先按可核驗税率／税項建模；虧損税盾不自動計入。費用已在分部利潤中包含時不能再減一次。當前NTM與目標日NTM不得混用；缺少季度預測時，財年按日曆重疊加權僅是有季節性侷限的近似。
P/E 先对齐盈利会计口径、SBC 与预测期间，再研究可比阶段和自身历史分布；不机械使用十五年均值，不把 PEG 当作定律。没有可靠可比数据时，倍数保持为明确的情景假设。EPS≤0 时停止普通 P/E 估值，另行研究现金流或分部资产价值；不强行给出负目标价。Align earnings accounting, SBC and forecast horizon before studying comparable stages and own-history multiples. Do not mechanically apply a 15-year average or treat PEG as a law. Without reliable comparables, multiples remain explicit scenario assumptions. When EPS≤0, stop ordinary P/E valuation and separately investigate cash-flow or segment asset value rather than issue a negative target.P/E 先對齊盈利會計口徑、SBC 與預測期間，再研究可比階段和自身歷史分佈；不機械使用十五年均值，不把 PEG 當作定律。沒有可靠可比數據時，倍數保持為明確的情景假設。EPS≤0 時停止普通 P/E 估值，另行研究現金流或分部資產價值；不強行給出負目標價。
发展阶段适配Development-stage adaptation發展階段適配
| 研究阶段Research stage研究階段 | 需要的证据与调整Evidence and adjustments需要的證據與調整 |
|---|---|
| 增长加速Acceleration增長加速 | 最近2–4次财报同口径趋势、指引与交付兑现；高增速明确持续期，不永久外推。Same-basis trends over 2–4 reports, guidance and delivery execution. Specify high-growth duration; never extrapolate indefinitely.最近2–4次財報同口徑趨勢、指引與交付兑現；高增速明確持續期，不永久外推。 |
| 增长正常化Normalization增長正常化 | 收入增速、利润转化与再投资逐步稳定；倍数与成熟度匹配，保留过渡路径。Revenue growth, profit conversion and reinvestment stabilize; match multiples to maturity and retain a transition path.收入增速、利潤轉化與再投資逐步穩定；倍數與成熟度匹配，保留過渡路徑。 |
| 压力／周期回撤Stress / cyclical contraction壓力／週期回撤 | 需求下修、毛利与现金流承压、融资约束；盈利与倍数可同时下降，不将Bear定义为固定跌幅。Demand revisions, margin/cash-flow pressure and financing constraints; earnings and multiples can fall together. Bear is not a fixed stock-return bucket.需求下修、毛利與現金流承壓、融資約束；盈利與倍數可同時下降，不將Bear定義為固定跌幅。 |
从证据到可复核估值From evidence to an auditable estimate從證據到可複核估值
| 环节Step環節 | 必须保留Required record必須保留 |
|---|---|
| 01 · Evidence | 申报与预测分别记录来源、发布日期、读取时间、对应期间；缺失不补造。Separate filings and forecasts, recording source, publication/retrieval time and covered period; leave missing values missing.申報與預測分別記錄來源、發佈日期、讀取時間、對應期間；缺失不補造。 |
| 02 · Earnings | 分项收入 × 分项营业利润率，减公司费用；再处理利息、税、归属和稀释股数，得到同口径 EPS。Segment revenue × segment operating margin, less corporate expense; reconcile interest, taxes, attribution and diluted shares into same-basis EPS.分項收入 × 分項營業利潤率，減公司費用；再處理利息、税、歸屬和稀釋股數，得到同口徑 EPS。 |
| 03 · Valuation | 目标日未来12个月 EPS × Exit P/E；估值日、目标日、盈利窗口均明确。24M仅作参考，未计算时不填值。Target-date next-12-month EPS × exit P/E; state valuation date, target date and earnings window. 24M is secondary and remains blank if uncalculated.目標日未來12個月 EPS × Exit P/E；估值日、目標日、盈利窗口均明確。24M僅作參考，未計算時不填值。 |
| 04 · Version | 模型规则版本、输入快照版本与公司适配版本分开；记录旧值、新值、证据、原因与结果变化。Version model rules, input snapshots and company adapters separately; retain old/new values, evidence, rationale and output changes.模型規則版本、輸入快照版本與公司適配版本分開；記錄舊值、新值、證據、原因與結果變化。 |
| 05 · Test | 到期终值、窗口触及、EPS误差与区间覆盖分别验证；没有样本外检验不发布概率或“准确率”。Test endpoint value, window touch, EPS error and interval coverage separately; publish no probabilities or accuracy claim without out-of-sample evaluation.到期終值、窗口觸及、EPS誤差與區間覆蓋分別驗證；沒有樣本外檢驗不發佈概率或“準確率”。 |
如何检验，而不是如何证明自己正确How to test, not how to prove ourselves right如何檢驗，而不是如何證明自己正確
回填只使用历史当时可得的资料；现在补做的历史模型标为回顾性研究，不能冒充当年公开预测。滚动12个月窗口高度重叠，不计作独立样本。财报重述、拆股、股息、价格源、缺失样本及选股偏差必须记录。Backfills use only information available at the historical cutoff. Reconstructions made today are retrospective research, not forecasts publicly made then. Overlapping rolling 12-month windows are not independent samples. Record restatements, splits, dividends, price sources, missing observations and selection bias.回填只使用歷史當時可得的資料；現在補做的歷史模型標為回顧性研究，不能冒充當年公開預測。滾動12個月窗口高度重疊，不計作獨立樣本。財報重述、拆股、股息、價格源、缺失樣本及選股偏差必須記錄。
到期误差 = (到期价格 − 原目标价) ÷ 原目标价；盈利误差 = (实际同口径 EPS − 预测 EPS) ÷ |预测 EPS|。EPS 为零时不计算该比例。窗口触及 ≠ 到期达标；Bear/Base/Bull 是经营情景，不天然具有概率。Endpoint error = (realized endpoint price − original target) / original target. Earnings error = (actual same-basis EPS − forecast EPS) / |forecast EPS|; undefined when forecast EPS is zero. A window touch is not endpoint success. Bear/Base/Bull are operating scenarios, not inherent probabilities.到期誤差 = (到期價格 − 原目標價) ÷ 原目標價；盈利誤差 = (實際同口徑 EPS − 預測 EPS) ÷ |預測 EPS|。EPS 為零時不計算該比例。窗口觸及 ≠ 到期達標；Bear/Base/Bull 是經營情景，不天然具有概率。
当前状态：本板块尚未完成新框架的校准与样本外检验，不公布模型胜率。原个股页面的分析师目标价历史触及记录仍是另一套统计，不移植为本模型的验证成绩。Current status: this framework has not completed calibration or out-of-sample tests; no model win rate is published. Analyst-target touch histories on equity pages remain separate statistics and are not relabeled as validation of these models.當前狀態：本板塊尚未完成新框架的校準與樣本外檢驗，不公佈模型勝率。原個股頁面的分析師目標價歷史觸及記錄仍是另一套統計，不移植為本模型的驗證成績。
版本记录Version record版本記錄
v0.1.0 · 2026-09-26：首次定义模型族、会计桥、阶段适配和验证规范；尚无新数值校准结果。未来结构性变更升级主版本，输入更新单独创建运行快照。v0.1.0 · 2026-09-26: first model-family, accounting-bridge, stage-adapter and validation specification. No newly calibrated numerical results. Structural changes require a major version; input updates create separate run snapshots.v0.1.0 · 2026-09-26：首次定義模型族、會計橋、階段適配和驗證規範；尚無新數值校準結果。未來結構性變更升級主版本，輸入更新單獨創建運行快照。