三笔跨期存量支取：
理论推导、测量层与策略引擎的完整量化体系

一、引言 / 1. Introduction

1.1 研究动机 / Research Motivation

1.1.1 2026年的"不可能三角" / The 2026 Impossible Trinity

2026年夏季，美国宏观经济数据显示出一个违反直觉的组合：联邦基金利率维持在3.50–3.75%的高位，但同时四大云厂商（微软、亚马逊、谷歌、Meta）的年度资本开支计划达到6,900–7,250亿美元（同比增长约75%），标普500前10只股票占总市值的38%——为百年六轮大泡沫中最高。传统经济学对此类"高利率+高资产价格+高财政赤字"的组合缺乏连贯的解释框架。本文论证这一矛盾状态是"三笔跨期存量支取同时到期"的结构性特征。

In the summer of 2026, U.S. macroeconomic data exhibits a counterintuitive configuration: the Federal Funds Rate is maintained at 3.50–3.75%, yet the four major cloud providers (Microsoft, Amazon, Google, Meta) plan annual capital expenditures of $690–725 billion (≈75% YoY increase), and the top 10 S&P 500 stocks account for 38% of total market capitalization—the highest across six major bubbles in a century. Traditional economics lacks a coherent framework for such a "high rates + high asset prices + high fiscal deficits" combination. This paper argues that this contradiction is a structural signature of the simultaneous maturity of three intertemporal stock drawdowns.

1.1.2 ABCT与MMT的百年争论 / A Century-Long Debate

奥地利学派商业周期理论(ABCT)的核心命题是：央行将利率压低至自然利率以下，导致企业家系统性高估社会真实储蓄水平，进而产生"不当投资"(malinvestment)。当利率恢复正常时，这些错误投资必须被清算。MMT则提出：主权货币发行国永远具有偿付能力，真正的约束是通胀而非财政赤字，因此危机中的"托底"是必要的政策选择。然而，ABCT低估了法币体系冻结清算的能力，MMT低估了物理约束（能源EROI衰减、人口TFR塌陷）的累积——两者各自捕捉了真相的一半。本文提出：将"不当投资"从单一的利率-信贷维度拆解为能源、人口、货币三类物理抵押品，从而将ABCT三维化；将MMT的"无限支付能力"翻译为受物理存量约束的"冻结系数"，从而将MMT物理化。

The ABCT's core proposition states that when central banks depress interest rates below the natural rate, entrepreneurs systematically overestimate the true level of societal savings, leading to malinvestment. When rates normalize, these errors must be liquidated. MMT counters that sovereign currency-issuing governments are never insolvent; the true constraint is inflation, not fiscal deficits, making crisis backstops a necessary policy choice. However, ABCT underestimates the fiat system's capacity to freeze liquidation, while MMT underestimates the accumulation of physical constraints (EROI decline, TFR collapse)—each capturing only half the truth. This paper proposes: decomposing malinvestment from a single interest-rate-credit dimension into three physical collateral types (energy, demographics, money), thus three-dimensionalizing ABCT; and translating MMT's "unlimited payment capacity" into a physically-constrained "freeze coefficient," thus physicalizing MMT.

1.2 核心贡献 / Core Contributions

1. 理论统一：建立三笔跨期存量支取的形式化数学框架，提供ABCT与MMT的统一量化表达。
2. 测量层构建（本文新增）：把宏观五票正式定义为框架的可观测化测量层，给出五票与三存量的一一映射、规则化公式（含污染度P的三通道线性映射）与票源铁律——四种票源语义（实盘真宏观 / 回测tech5 / 实验proxy / Shiller真宏观）在任何实证中禁混用。
3. 因子工程：构建四层数据加工管线（清洗→基础因子→高阶合成→正交提纯），从16个数据源生产25个标准化量化因子。
4. 三维修正：引入Gini三通道修正(ε/μ/π)、Ω底层空心化修正、γ-GSM黄金压力修正，建立完整的δ(S'')清算破坏力函数。
5. 概率预警：推导SBRI系统性破产风险指数的连续概率表达，替代传统二元触发机制。
6. 策略引擎与回测（本文并入）：v2.2三档状态机与三档多资产配置、五票退出进入决策矩阵、E10R5双档熔断、ESC3_025升坡门控的完整规格，以及8国30年含成本回测全集——把"预测回撤深度"的答案转化为"用什么仓位、按什么规则迎接"的可执行结构。

1. Theoretical unification: Establish a formal mathematical framework for three-stock drawdowns, providing a unified quantitative expression of ABCT and MMT.
2. Measurement layer (new): The five macro votes formally defined as the framework's observable layer, with the one-to-one mapping to the three stocks, rule-based formulas (including the three-channel linear mapping for pollution P), and the vote-source iron law—four non-interchangeable source semantics (live true-macro / backtest tech5 / experimental proxy / Shiller true-macro).
3. Factor engineering: Construct a four-layer data processing pipeline producing 25 standardized factors from 16 data sources.
4. Three-tier corrections: Introduce Gini three-channel correction (ε/μ/π), Ω bottom-hollowing correction, and γ-GSM gold stress correction to build a complete δ(S'') liquidation destructiveness function.
5. Probabilistic early warning: Derive a continuous probabilistic expression for the SBRI Systemic Bankruptcy Risk Index, replacing binary trigger mechanisms.
6. Strategy engine and backtests (merged in): Full specifications of the v2.2 three-tier state machine and allocations, the five-vote exit/re-entry decision matrix, the E10R5 dual-tier breaker, and the ESC3_025 ramp gate, together with the cost-inclusive 30-year, eight-country backtest suite—converting the "predicted drawdown depth" into an executable structure of "with what positions, by what rules."

二、理论框架：三笔跨期存量支取的数学结构 / 2. Theoretical Framework

2.1 ABCT的形式化表达 / Formal ABCT Specification

设经济中存在两类行为主体：家庭（储蓄者）和企业（投资者）。家庭的跨期效用函数为：

Let the economy contain two types of agents: households (savers) and firms (investors). The household intertemporal utility function is:

其中 $\rho$ 为时间偏好率（自然利率），$r$ 为市场利率。最优消费路径的一阶条件给出欧拉方程：

where $\rho$ is the time preference rate (natural rate), and $r$ is the market rate. The first-order condition yields the Euler equation:

当央行将市场利率压低至 $r < \rho$（即 $r < r^*$，其中 $r^*$ 为自然利率），家庭的最优反应是减少储蓄、增加当期消费。这一信号的"系统性误导"效应是ABCT的核心：企业观测到低利率和强消费，错误推断社会拥有充足储蓄来支撑长周期投资。

When the central bank depresses the market rate to $r < \rho$ (i.e., $r < r^*$), the household's optimal response is to reduce savings and increase current consumption. This "systematic misguidance" effect is the core of ABCT: firms observe low rates and strong consumption and mistakenly infer that society has sufficient savings to support long-cycle investment.

2.2 三笔存量支取的数学定义 / Mathematical Definition of Three-Stock Drawdowns

我们将ABCT中的"不当投资"从单一的利率-信贷维度推广到三个互相抵押的物理维度。设三笔存量为：

We generalize "malinvestment" from a single interest-rate-credit dimension to three mutually collateralized physical dimensions. Define three stocks as:

每一笔存量的"可持续度"由其剩余存量与支取速率的比值决定：

The "sustainability" of each stock is determined by the ratio of remaining stock to withdrawal rate:

式 (4) 实证口径注（2026-09-04 物理重锚）：e_E(t) 的可测量口径为发现账剩余储量（RDE 估计：累计发现 − 累计采出，池规模 D∞ = 2,782 Gb 由发现动力学内生外推，成熟期 Hubbert 线性 r = −0.997）。据此 S_E 的直接实例化 = 发现账剩余年限 R_disc/q：2026 = 29.1 年，对照账面 R/P 52.4 年（账面储量含重估上调——1980-2023 年账面增长的 ~85% 来自重估而非新发现，两账 1.80× 差距本身是披露项）。存量侧 E(t) 对应采出进度 Q/D∞：1929 = 1.1% 已采 → 2026 = 62.5% 已采（37.5% 剩余），详见 §6.8 相空间重锚表。

Eq. (4) empirical-caliber note (physically re-anchored 2026-09-04): the measurable caliber of e_E(t) is the discovery-account remaining stock (RDE estimate: cumulative discoveries minus cumulative production, with pool size D∞ = 2,782 Gb endogenously inferred from discovery dynamics; mature-period Hubbert linearization r = −0.997). The direct instantiation of S_E is therefore the discovery-account remaining years R_disc/q: 29.1 years in 2026, versus the book R/P of 52.4 years (book reserves include revaluation uplifts—~85% of 1980–2023 book growth came from revaluation rather than new discoveries; the 1.80× gap between accounts is itself a disclosure item). The stock side E(t) corresponds to depletion progress Q/D∞: 1.1% extracted in 1929 → 62.5% in 2026 (37.5% remaining); see the re-anchored phase table in §6.8.

核心命题（三线共振假说 · 2026-09-04 实证修正为"双线共振 + 清算/冻结条件"）：当三条可持续度曲线同时逼近零点——即 $S_E \to 0 \land S_P \to 0 \land S_M \to 0$——经济系统进入"三线共振"状态（理论极限态）。实证修正：百年样本中完整触发的仅有信用×人口双线（1929 / 2026 两次；能源线在两个时点均未临界——1929 池 98.9% 完整，2026 深耗竭但 S_E = 29.1 年远非零），完整论证见 §13 命题一。此时：

1. 单一部门的冲击无法被其他部门的支取空间吸收；
2. 交叉抵押机制失效——曾经互相掩盖的三笔借款同时要求还款；
3. 系统性危机的概率从独立事件的乘积跃升至近1（共振放大）。

Core Proposition (Triple Resonance Hypothesis · empirically revised 2026-09-04 to "dual resonance + clearing/freezing conditions"): When all three sustainability curves simultaneously approach zero—i.e., $S_E \to 0 \land S_P \to 0 \land S_M \to 0$—the economic system enters a "triple resonance" state (a theoretical limiting state). Empirical revision: only the credit × demographics pair has fully triggered in the century sample (1929 / 2026; the energy line was sub-critical at both points—98.9% of the pool intact in 1929, and deep-but-not-critical depletion in 2026 with S_E = 29.1 years, far from zero); see Proposition 1 in §13 for the full argument. In this state: (1) sector-specific shocks cannot be absorbed by drawdown space in other sectors; (2) cross-collateralization mechanisms fail as all three loans demand repayment simultaneously; (3) the probability of systemic crisis jumps from the product of independent events to near-unity (resonance amplification).

2.3 交叉抵押机制的数学表达 / Cross-Collateralization Mechanics

三笔支取之间的交叉抵押关系可以通过一个3×3的耦合矩阵 $\mathbf{A}$ 来描述：

Cross-collateralization among the three drawdowns can be described by a 3×3 coupling matrix $\mathbf{A}$:

该矩阵的核心性质：$a_{13} < 0$ 意味着货币宽松可以替代能源约束（压低利率→掩盖EROI下降），$a_{31} > 0$ 意味着能源危机迫使货币超发（1941年石油禁运→珍珠港→战争开支），$a_{23} < 0$ 意味着人口萎缩被货币宽松对冲（日本：30年QQE打一场已输的战争）。

Key properties: $a_{13} < 0$ means monetary easing can substitute for energy constraint (lower rates → mask EROI decline); $a_{31} > 0$ means energy crisis forces monetary expansion (1941 oil embargo → Pearl Harbor → war spending); $a_{23} < 0$ means demographic decline is hedged by monetary easing (Japan: 30 years of QQE fighting an already-lost war).

2.4 MMT的物理化：冻结系数的推导 / Physicalizing MMT: Derivation of the Freeze Coefficient

其中 $M_{max}$ 为央行资产负债表上限（受通胀约束），$\Delta_{clearing}(t)$ 为当前需要清算的"不当投资"总量。冻结系数 $\phi$ 是MMT"无限支付能力"的物理化表达——它揭示支付能力并非无限，而是取决于货币政策空间与清算需求之比。

where $M_{max}$ is the central bank balance sheet ceiling (constrained by inflation), and $\Delta_{clearing}(t)$ is the total malinvestment requiring liquidation. Freeze coefficient $\phi$ physicalizes MMT's "unlimited payment capacity"—it reveals that payment capacity is not unlimited but depends on the ratio of monetary policy space to clearing requirements.

2.5 数据底盘与三条慢变量观测轨迹 / 2.5 Data Base and Slow-Variable Trajectories

指标层使用 12 国加权样本（M2 加权）与五大类 25 项宏观因子，叠加四项理论因子（ψ 制度转化、Φ−ΔΦ 产业能力、R−P−I 三维迁移、R&D×GDP 交互项）；校准样本为美国 1929 / 1973 / 1987 / 2000 / 2007–08 / 2021 六次泡沫时点（其中回撤方程权重由 1929 / 1973 / 2000 / 2007 四次出清精确拟合）。所有 min-max 缩放结果截断至 [0, 10]。三条慢变量的观测轨迹与作用通道：

The indicator layer uses a 12-country M2-weighted sample with 25 macro factors in five classes, plus four theoretical factors (ψ institutional conversion, Φ−ΔΦ industrial capability, R−P−I migration triple, R&D×GDP interaction). Calibration uses six US bubble points—1929 / 1973 / 1987 / 2000 / 2007–08 / 2021—with drawdown weights exact-fitted on the four clearings of 1929 / 1973 / 2000 / 2007. All min-max scaled values are truncated to [0, 10]. Slow-variable trajectories and channels:

指标层不止于美国主样本：12 国的 BII 与 RCS 读数以 M2（万亿美元）× 美国溢价 2.5×为样本权重交叉校准（合计 123.6），用于 v2.2 国别修正项的触发判别：

The indicator layer extends beyond the US master sample: BII and RCS readings across 12 countries are cross-calibrated with sample weights = M2 (in $T) × a 2.5× US premium (total 123.6), driving the v2.2 country-correction triggers:

2.6 v3.0 → v3.3 体系演进对照 / 2.6 The v3.0 → v3.3 Lineage

体系的每一版升级都改变一个可分离的部件，全部部件的谱系如下（跨版本比较必须显式声明版本）：

Each version upgrade changes one separable component; the full lineage (cross-version comparison requires explicit declaration):

三、测量层：从框架到五票 / 3. Measurement Layer: From Framework to Five Macro Votes

第二章建立了三笔跨期存量支取的形式化数学结构（E/P/M 三条慢变量及其耦合矩阵），但理论框架本身不直接产生可交易的决策信号——它需要一层"可观测化"的翻译。本章正式定义这一翻译层：宏观五票。五票不是独立于框架的新发明，而是三存量在可测数据上的投影——每张票对应框架中一个明确的数学对象。本章给出五票与三存量的映射推导、规则化公式（含污染度 P 的三通道线性映射）、票源铁律（四种票源语义禁混用），以及回测口径与实盘口径之间已知断裂的前向双轨验证方案。

Chapter 2 established the formal mathematical structure of three intertemporal stock drawdowns (E/P/M slow variables and their coupling matrix), but the theoretical framework does not directly produce tradable decision signals—it requires an "observable" translation layer. This chapter formally defines that layer: the five macro votes. The five votes are not an invention independent of the framework, but projections of the three stocks onto measurable data—each vote corresponds to a well-defined mathematical object in the framework. We present the mapping from the three stocks to the five votes, the rule-based formulas (including the three-channel linear mapping for pollution P), the vote-source iron law (four non-interchangeable source semantics), and the forward dual-track validation scheme for the known gap between backtest and live calibers.

3.1 理论对象 → 五票映射表 / Theory-to-Vote Mapping Table

五票与三存量的对应关系如下表所示。每张票的"框架对象"列指向第二章或第四章中的具体数学定义，"观测数据"列给出票值的实际数据来源。这一映射是本章的核心理论贡献：它把抽象的存量耗竭定律连接到月度可测的宏观指标。

The correspondence between the five votes and the three stocks is shown below. The "Framework Object" column points to the specific mathematical definition in Chapter 2 or 4; the "Observation Data" column gives the actual data source for each vote. This mapping is the core theoretical contribution of this chapter: it connects abstract stock-depletion laws to monthly observable macro indicators.

关键定位：由于实盘真宏观五票的完整月度历史序列无法获取（2026-08 前多次尝试失败），上表映射在当前阶段定位为"理论动机 / 机制解释"而非"实盘测量"——回测使用 Shiller 真宏观五票作 proxy（§3.4 详述），实盘使用规则化真五票规格（§3.2 详述），两者口径差异通过前向双轨验证弥合（§3.4 详述）。

Key positioning: since complete monthly historical series for the live true-macro five votes are unavailable (multiple attempts failed before 2026-08), the mapping above is currently positioned as "theoretical motivation / mechanism explanation" rather than "live measurement"—backtests use Shiller true-macro five votes as a proxy (§3.4), live trading uses the rule-based true-five-vote specification (§3.2), and the caliber gap between them is bridged through forward dual-track validation (§3.4).

3.2 五票规格表 / The Five-Vote Specification Table

下表给出五票的规则化计算公式、数据源与票档阈值。所有阈值均为拟定值，在首次全流程回算时以当前读数与历史泡沫期（2000/2007/2021）对照校准后冻结。票值高 = 泡沫证据强 = 支持退出/防御。每票分三档：0 / 0.5 / 1，票和 0–5。

The table below gives the rule-based formulas, data sources, and vote-tier thresholds for the five votes. All thresholds are proposed values to be calibrated against historical bubble periods (2000/2007/2021) during the first full-process recalculation and then frozen. A high vote value = strong bubble evidence = supports exit/defensive posture. Each vote has three tiers: 0 / 0.5 / 1, with the sum ranging 0–5.

注：BII 的 L 维（杠杆存量，权重 0.22）与票⑤同源数据（居民债务/GDP），但公式独立计算、不重复计票。原版三票信号（SOFR-OIS > 7bp、保证金同比转负、IPO 崩塌）定位为"顶部/清算确认信号"，在退出票表之外——它们是清算已发生的证据，而非泡沫仍在上膨胀的证据。

Note: the L dimension (leverage stock, weight 0.22) of BII shares source data with vote ⑤ (household debt/GDP), but the formulas are computed independently and do not double-count. The original three signals (SOFR-OIS > 7bp, margin debt turning negative, IPO collapse) are positioned as "top/clearing-confirmation signals" outside the exit-vote table—they are evidence that clearing has begun, not that the bubble is still inflating.

3.3 污染度 P 规则化公式 / Rule-Based Pollution P Formula

污染度 P 是五票中唯一曾依赖 AI 主观研判（0–10）的票项。自 2026-08-27 定版起，P 改为纯规则化计算，AI 研判降级为"参考意见"，不再进入票值。规则化公式如下：

Pollution P is the only vote that previously relied on AI subjective judgment (0–10). From the 2026-08-27 specification onward, P is replaced by a purely rule-based calculation; AI judgment is downgraded to "reference opinion" and no longer enters the vote value. The rule-based formula:

三个分项的线性映射规则：

1. I_level 通胀水平：max(CPI 同比, 核心 PCE 同比)，线性映射 [0%, 8%] -> [0, 10]
2. I_trend 通胀趋势：核心 PCE 3 个月年化环比 - 12 个月年化环比（加速度项），线性映射 [-2pp, +2pp] -> [0, 10]，零加速度 = 5 分
3. I_money 货币信用：0.5*M2 同比 + 0.5*银行信贷同比，线性映射 [0%, 12%] -> [0, 10]

Three sub-component linear mapping rules:

1. I_level (inflation level): max(CPI YoY, core PCE YoY), linear map [0%, 8%] -> [0, 10]
2. I_trend (inflation trend): core PCE 3-month annualized QoQ - 12-month annualized QoQ (acceleration term), linear map [-2pp, +2pp] -> [0, 10], zero acceleration = 5
3. I_money (monetary credit): 0.5*M2 YoY + 0.5*bank credit YoY, linear map [0%, 12%] -> [0, 10]

诚实披露：规则化 P 首次回算值预计显著低于 AI 研判末值（约 7）——AI 研判含定性前瞻（财政、关税、预期传导），规则化后仅保留可测输入。这是消除主观性的预期代价。首次回算于下次全流程校正执行，回算值与 AI 末值的差异记入变更记录。

Honest disclosure: the rule-based P's first recalculation value is expected to be significantly lower than the AI-judgment terminal value (~7)—the AI judgment included qualitative forward-looking factors (fiscal policy, tariffs, expectation transmission) that the rule-based version retains only measurable inputs. This is the expected cost of eliminating subjectivity. The first recalculation will be executed during the next full-process correction; the difference between the recalculation and the AI terminal value will be recorded in the change log.

3.4 票源铁律：四种票源语义禁混用 / The Vote-Source Iron Law

"宏观五票"在本体系存在四种语义，任何实证必须先指明票源，不同票源的结论不可互换。这是本章最重要的方法论纪律。四种票源的完整对照如下：

"Five macro votes" carries four distinct semantics in this system; any empirical claim must declare its source first, and conclusions across sources are non-interchangeable. This is the most important methodological discipline of this chapter. The full comparison of the four sources follows:

3.5 前向双轨验证披露 / Forward Dual-Track Validation Disclosure

回测票源 = Shiller 真宏观五票（CAPE 分位 / CPI / 实际利率 / 加息 / 价格过热）——proxy 性质。实盘真五票 = §3.2 规格。对照回测（proxy 保真度验证）已尝试、不可行：真五票完整历史序列无法获取（2026-08 前多次尝试失败，用户确认）。

The backtest vote source = Shiller true-macro five votes (CAPE percentile / CPI / real rate / rate hike / price overheating)—of a proxy nature. The live true five votes = the §3.2 specification. Side-by-side backtest validation (proxy fidelity check) has been attempted and is infeasible: complete historical series for the true five votes cannot be obtained (multiple attempts failed before 2026-08, confirmed by the user).

可行的替代验证方案为前向双轨验证：自实盘切换日起，sim-us-macro 同时记录两套票值（Shiller 真宏观 + 规则化真五票），前向积累信号一致率统计——12 个月后可出首个前向对照报告。在首次回算与并行验证完成前，实盘状态机仍由 Shiller 真宏观票驱动，规则化真五票作为参考并行记录。

The feasible alternative is forward dual-track validation: from the live-switch date onward, sim-us-macro simultaneously records two sets of vote values (Shiller true-macro + rule-based true five votes), forward-accumulating signal-consistency statistics—12 months later, the first forward comparison report can be produced. Until the first recalculation and parallel validation are complete, the live state machine remains driven by Shiller true-macro votes, with the rule-based true five votes recorded in parallel as reference.

四、因子工程与指标构建 / 4. Factor Engineering and Index Construction

本章把数据加工管线与核心指标的解析推导合并为连续的八节，覆盖从原始数据到 BII / RCS / S''' 的完整构建链。四层因子工程管线处理 16 个宏观数据源的 25 个标准化因子，经清洗->基础因子->高阶合成->正交提纯四步，输出直接喂入回撤预测方程（§5）和策略引擎（§8）。

This chapter merges the data-processing pipeline with the analytical derivation of core indices into eight continuous subsections, covering the complete construction chain from raw data to BII / RCS / S'''. The four-layer factor engineering pipeline processes 25 standardized factors from 16 macro data sources through cleaning -> basic factors -> composite synthesis -> orthogonal purification, with outputs feeding directly into the drawdown prediction equation (§5) and the strategy engine (§8).

4.1 数据架构 / Data Architecture

4.2 L1: 数据清洗 / Data Cleaning

4.2.1 日历对齐 / Calendar Alignment

4.2.2 频率统一：Chow-Lin插值 / Frequency Unification

4.2.3 MAD稳健去极值 / MAD Robust Outlier Detection

4.3 L2: 基础因子构造 / Basic Factor Construction

4.4 L3: 高阶合成因子 / Composite Factor Synthesis

4.5 L4: Schmidt正交化提纯 / Schmidt Orthogonalization

4.6 BII泡沫强度指数 / Bubble Intensity Index

披露补充 · 五维构造完整规格：每个子维度 $s_i$ 由其原始输入经 min-max 缩放并截断至 [0,10] 得到。权重设计体现"存量重于流量"：杠杆存量（0.22）与非传统信贷（0.20）合计 0.42，超过估值偏离（0.28）——这与杠杆泡沫文献的经验发现一致：负债推动的泡沫破裂后的产出损失显著更大 [26]。2026 年 BII = 6.74，低于 2021 年的 7.25：它不是利率泡沫——在 3.75% 利率下由企业现金流、私人信贷与监管额度吹起。

Disclosure supplement · full five-dimension specification: each sub-dimension $s_i$ is min-max scaled from raw inputs and truncated to [0,10]. Weights embody "stocks over flows": leverage stock (0.22) plus non-traditional credit (0.20) total 0.42, exceeding valuation deviation (0.28)—consistent with the empirical finding that debt-financed bubbles are followed by larger output losses [26]. The 2026 BII of 6.74 sits below 2021's 7.25: this is not a rate-driven bubble—it was inflated at 3.75% policy rates by corporate cash flow, private credit, and regulatory quotas.

v3.0 五维升级中信贷扩张权重自 0.25 降至 0.12，让位于新增的非传统信贷维度——捕捉 AI 轻资本范式下传统银行信贷渠道的失效。v2.2 在 12 国校准上追加三项国别修正：

In the v3.0 upgrade the credit-expansion weight fell from 0.25 to 0.12, ceding room to the new non-traditional-credit dimension—capturing the failure of the traditional bank-credit channel under the AI light-capex paradigm. v2.2 adds three country-level corrections on the 12-country calibration:

4.6.1 BII v3 · M 维施政空间口径升级（2026-09-02 增补）/ 4.6.1 BII v3: The M-Dimension Policy-Space Upgrade

披露增补：式 (16) 的 M 维（货币宽松，权重 0.18）自 2026-09 起在连续研究序列 BII v3（366 月，1996-01–2026-06，outputs/bii_rcs_frag_continuous_v3.json）中升级为三分量「有效宽松/施政空间」口径——五维框架与权重向量 α = [0.28, 0.12, 0.18, 0.22, 0.20] 不动，M 维内部新增第三分量 m3 = 12MΔ(债务/GDP) 负向挤占映射。这是 BII 首次携带财政维度信息：原五维的宏观变量中财政完全缺席，而预研证明财政消耗速度 x_d 与 M 维两既有分量正交（r = −0.002）、对四态（快崩/慢衰退/繁荣/浅回调）的判别力 d′ = 2.629（快崩段 +10.3pp/yr vs 慢衰退段 −0.6pp/yr）——纯增量信息，非噪声替换。语义上，M 维从「已多宽松」（水平口径）升格为「有效宽松/施政空间」：计入政府发债对流动性的吸收端与对宽松效果的挤占端——QE 扩表若同步伴随赤字海啸，净宽松远小于其名义规模（2020 段内语义反转由此修正，见下文 E4）。r−g（实际利率减增长）作为披露性背景变量保留（d′ = 1.182 但与实际利率分量部分冗余，不作独立分量）。口径关系声明：本节正文 BII = 6.74 系人工校准口径（式 (25) 标定用）；BII v3 连续序列为独立的研究监测口径（canonical，用于状态判别与监测锚点），两者分属不同序列，禁止直接对比。

Disclosure supplement: as of 2026-09 the M dimension (monetary easing, weight 0.18) of Eq. (16) is upgraded within the continuous research series BII v3 (366 months, 1996-01–2026-06) to a three-component "effective easing / policy space" caliber—the five-dimension framework and weight vector α = [0.28, 0.12, 0.18, 0.22, 0.20] unchanged, with a third component m3 = a negative-crowding-out mapping of 12MΔ(debt/GDP) added inside M. This is the first time BII carries fiscal information: the original five dimensions contain no fiscal variable, whereas pre-research shows the fiscal drawdown speed x_d is orthogonal to both existing M components (r = −0.002) with four-state discriminability d′ = 2.629 (crash episodes +10.3pp/yr vs slow recessions −0.6pp/yr)—pure incremental information, not a noise substitution. Semantically, M moves from "how much easing" (a level notion) to "effective easing / policy space": QE expansion accompanied by deficit tsunamis delivers far less net easing than its nominal size (the 2020 in-episode semantic reversal is corrected by this, see E4 below). The r−g spread is retained as a disclosed background variable (d′ = 1.182 but partially redundant with the real-rate component). Caliber statement: the BII = 6.74 of this section is the manually calibrated caliber (used for Eq. (25) calibration); the BII v3 continuous series is an independent research-monitoring caliber (canonical, for state discrimination and monitoring anchors)—the two series must not be directly compared.

f_db 的语义：x_d → 10pp/yr（财政消耗极端，如 2008-12 的 +10.53、2020-06 的 +29.74）时 f_db → 0，施政空间耗尽；x_d ≤ −3（债务去化）时 f_db = 1，空间充裕。阈值（x=10→0、−3→1）为预研固化的先验（取自事件段分布的天然空隙），非本样本拟合；权重 0.30 经 {0.15, 0.30, 0.45} 三档敏感性扫描，方向性结论全部成立。

Semantics of f_db: as x_d → 10pp/yr (extreme fiscal drawdown, e.g., 2008-12 at +10.53 and 2020-06 at +29.74), f_db → 0—policy space exhausted; as x_d ≤ −3 (debt deleveraging), f_db = 1. Thresholds (x = 10 → 0, −3 → 1) are priors fixed in pre-research (natural gaps in the event-segment distribution), not in-sample fits; the weight 0.30 passes a three-tier sensitivity sweep over {0.15, 0.30, 0.45} with all directional conclusions intact.

五臂裁定（预注册 E0–E5，2026-09-02 全跑）。设计轴 = 财政变量以何种映射进入 M 维：a 正向燃料 / b 负向挤占 / c 不对称警戒 / d 机械压力 / e 速度替换。数据裁定 b 臂胜出：

Five-arm adjudication (pre-registered E0–E5, all run 2026-09-02). The design axis = the mapping through which the fiscal variable enters M: (a) positive fuel, (b) negative crowding-out, (c) asymmetric alert, (d) mechanical pressure, (e) speed replacement. Arm b wins on data:

裁定要点：b 臂是唯一在全部三对关键区分度上超越 v1 原版的臂；E4 的 2020 段内语义反转消除 88%（段内 BII 斜率 +0.188/月 → +0.023/月——财政大消耗 x_d 冲至 +18~30pp/yr 压低 M′，抵消央行扩表托底）；E1 扰动温和（corr 0.967 > 0.95 局部性线）。值得注意的是语义直觉再次输给数据：设计稿推荐的起点是 c 臂（不对称警戒），被实测彻底推翻——2020 年 x_d 冲满格使 c 臂的映射反而加剧 BII 托升（+0.291/月，比原版还差 55%）。「探针必须跑，换理由死」由此成为本体系的制度化纪律。

Adjudication notes: arm b is the only arm that beats the original v1 on all three critical discrimination pairs; E4 shows the 2020 in-episode semantic reversal reduced by 88% (in-episode BII slope +0.188/mo → +0.023/mo—fiscal drawdown x_d surging to +18–30pp/yr depresses M′, offsetting the central-bank balance-sheet backstop); E1 disturbance is mild (corr 0.967 > the 0.95 locality line). Notably, semantic intuition again loses to data: the design memo's recommended starting point was arm c (asymmetric alert), flatly overturned by measurement—x_d maxing out in 2020 makes arm c's mapping amplify the BII backstop instead (+0.291/mo, 55% worse than the original). "Probes must be run; changing the rationale is not allowed" is thereby institutionalized.

E3 诚实归档（预注册条款触发）：式 (25) 的慢衰退外推无任何臂改善（b 臂 19.6pp vs v1fx 19.4pp，噪声级）——根源在式 (25) 泡沫清算口径本身对慢衰退系统性高估，M 升级的真实价值 = 改状态分辨率（E2/E4），不改回撤精度（E3）；慢衰退外推由独立的慢衰退判别器口径另立解决。no-look-ahead 声明：债务取月内最后公布值（T+1 公布，泄漏 ≤2 工作日）；GDP 季度 forward-fill（只用 ≤ 当月最近季度）；12MΔ 仅用 t 与 t−12 两点；五臂选择标准在实验前锁定于设计稿。

E3 honest archival (pre-registered clause triggered): no arm improves the Eq. (25) slow-recession extrapolation (arm b 19.6pp vs v1fx 19.4pp, noise-level)—the root cause lies in the bubble-clearing caliber of Eq. (25) itself, which systematically overestimates slow recessions; the real value of the M upgrade = state resolution (E2/E4), not drawdown precision (E3). No-look-ahead: debt taken at the last published value within the month (T+1 publication, leakage ≤ 2 working days); GDP quarterly forward-fill (nearest quarter ≤ current month only); 12MΔ uses points t and t−12 only; the five-arm selection criteria were locked in the design memo before experiments.

监测锚点（分位数保持法新建，无拟合——v1 从无正式监测线，此为新建而非「阈值刷新」）：快崩态 P95 读数 2.773 保持其全样本分位（0.5929）映射到 v3 序列得警戒线 3.304；慢衰退观察带 P10–P90 = [2.043, 4.507]。锚点语义为「确认型」：BII 是形成期指标，跌破警戒线即进入快崩清算读数区，非预警。当前读数 2026-06 v3 = 4.636（prosperity · 形成期高位，高于警戒线与观察带上沿）。

Monitoring anchors (newly built by quantile preservation, no fitting—v1 never had formal monitoring lines, so these are new constructions rather than "threshold refreshes"): the crash-state P95 reading 2.773, mapped at its full-sample quantile (0.5929) onto the v3 series, yields the alert line 3.304; the slow-recession observation band P10–P90 = [2.043, 4.507]. The anchor semantics are "confirmatory": BII is a formation-phase index—crossing below the alert line marks entry into crash-clearing territory, not an early warning. Current reading: 2026-06 v3 = 4.636 (prosperity, formation-phase high, above both the alert line and the band ceiling).

4.7 RCS真实约束指数 / Real Constraint Score

式 (18b) 替代理由（数据质量名义，2026-09-03 部署）：v2.1 的 EROI 分量并非实测序列，而是按已知拐点（1930/1970/2000/2020）分段线性插值的文献锚点构造——它与日历年份的秩相关 ρ=+1.000（完美单调），意味着其全部「信息」都是时间趋势的人工产物；控制年份后，EROI 与同期金价的相关由 raw −0.905 反号为 +0.095（G3 对照检验），证明其表面预测力来自与金价共享的时间趋势耦合，而非能源基本面。替代口径 EI_REAL 完全由 FRED 实测序列构造（分子 TETCBUS 总能耗 12 月滚动和、分母 GDPC1 实际 GDP 季率、分子分母同源 1949 起连续），零价格变量、零文献插值。金价伪影检验 G2：EI_REAL vs 金价 raw −0.897 → 控年份 +0.071≈零（年代内 −0.189 弱负）——「实际无相关、弱相关」成立，无金价伪影载体。月度部署闸门（预注册）：M1 对账通过（月末 vs 年度口径 median|Δ|≤2%）；M2 主判据「月度年末非重叠偏ρ v2.2≥v2.1」实测 Δ=−0.001（打平非恶化），三辅助证据全部支持不劣（M3 全月度重叠偏ρ +0.189>+0.160；年度同窗 1977–2024 +0.304>+0.230；年度全段 +0.240>+0.176）→ 经裁决以数据质量名义部署，预测力名义留痕；M3 全月度口径仅披露。等价触发线：v2.1 θ=5.0 同分位（95.5%）映射至 v2.2 得 θ₂=4.85，Phase-2 触发条款随之改 v2.2>4.85（旧条款 v2.1>5.0 留痕弃用）。语义变化如实披露：v2.2 口径下 1996+ 窗口 0 个月超线（历史峰值 4.81 = 2005-09），v2.1 中 2005 触发段的 EROI 分量读数 7.23 vs s_ei 2.89——该段超线系插值伪影虚增。诚实边界：EI_REAL 与年份 ρ=−0.991（技术进步使能耗强度单调下降），同样携带时间趋势，但该趋势是实测经济事实而非插值产物；GDPC1 发布错位（advance 约 30 天 + 保守 1 月）已由 avail 滞后规则（季度结束月 ≤ m−2）显式处理。

Eq. (18b) replacement rationale (data-quality basis, deployed 2026-09-03): the v2.1 EROI component is not a measured series but a literature-anchor piecewise-linear interpolation over known turning points (1930/1970/2000/2020)—its rank correlation with calendar year is ρ=+1.000 (perfectly monotone), meaning all of its "information" is an artificial time-trend artifact; controlling for year, EROI's correlation with contemporaneous gold flips from raw −0.905 to +0.095 (G3 control test), proving its apparent predictive power stems from a shared time-trend coupling with gold, not energy fundamentals. The replacement EI_REAL is constructed entirely from measured FRED series (numerator: TETCBUS total energy consumption 12-month rolling sum; denominator: GDPC1 real GDP quarterly annual rate; both continuous from 1949)—zero price variables, zero literature interpolation. Gold-artifact test G2: EI_REAL vs gold raw −0.897 → year-controlled +0.071 ≈ zero (within-decade −0.189 weakly negative)—no gold-artifact carrier. Monthly deployment gates (pre-registered): M1 reconciliation passed (month-end vs annual caliber median|Δ|≤2%); M2 primary criterion "monthly year-end non-overlapping partial-ρ v2.2≥v2.1" measured Δ=−0.001 (tie, not deterioration), with all three auxiliary lines supporting non-inferiority (M3 full-monthly +0.189>+0.160; annual same-window 1977–2024 +0.304>+0.230; annual full-span +0.240>+0.176) → deployed on a data-quality basis by adjudication, predictive-power basis archived; M3 full-monthly disclosed. Equivalent trigger line: v2.1 θ=5.0 at the same quantile (95.5%) maps to v2.2 θ₂=4.85, and the Phase-2 trigger clause is revised to v2.2>4.85 (old clause v2.1>5.0 archived and deprecated). Semantic change disclosed honestly: under the v2.2 caliber the 1996+ window has zero months above the line (historical max 4.81 = 2005-09); the v2.1 2005 trigger segment's EROI component read 7.23 vs s_ei 2.89—that crossing was interpolation-artifact inflation. Honest boundary: EI_REAL correlates with year at ρ=−0.991 (technical progress monotonically lowering energy intensity)—it too carries a time trend, but one that is a measured economic fact rather than an interpolation artifact; GDPC1 publication lag (advance ~30 days + 1 conservative month) is explicitly handled by the avail rule (quarter-end month ≤ m−2).

式 (18c) 上代理由（趋势形态伪影证伪，2026-09-05 五环落地）：v2.2 的 s_ei 水平值携带 EI_REAL 与年份 ρ=−0.991 的长期改善趋势——77 年效率红利主导水平读数，使 2026 年读数 0.88 呈「约束松」的趋势形态伪影（水平值影子病卡线：金价 raw −0.79 → 控年份仍 +0.30）；且 v2.1 EROI 插值与 v2.2 实测水平值数据源不可互比（EROI 控年份残存联动 +0.73 为三代最脏）。v2.3 把趋势形态构造性杀死：病检全过——T no（去趋势对时间构造正交）/ G no（对金价 raw +0.17，三代唯一过线者）/ 信号方差提升 8×（0.189 vs v2.2 的 0.024）/ 非日历映射 ρ(S,t) = −0.44。核心发现——2026 读数方向反转：去趋势口径 7.71（月度池）「约束紧」——2015-2026 效率红利停摆 11 年、EI_REAL 偏离长期趋势 +10.4% 为 1949 年来最高；RCS-E 由 1.67 → 2.83，实测总分 4.83 → 5.42。八时点场景回检：2000 年起 Δ 转正单调放大（2008 GFC +0.68 → 2026 +1.44）；危机前拐头仅 2008 成立（诚实披露：状态度量非预警器）。诚实边界：与金价 2000-25 周期共动 +0.87 属真实宏观联动（非伪影，禁读因果）；log 去趋势版符号不稳（2000 前正后负）已弃，主口径取线性；式(25) DD 联动 80.3% 属叙事层系数外推失真仅供参考。部署状态：观测层上线（不动生产仓位）；0.75 总分修复裁决挂起——按 2026-09-05 指令部署后再议原值/去趋势值修改，v2.3 实测总分 5.42 届时作分量复算层新基准。产物：_calc_ei_dt_monthly.py + ei_dt_monthly.json（583 月管道）。

Eq. (18c) upgrade rationale (trend-shape artifact falsified, five-ring landed 2026-09-05): the v2.2 s_ei level reading carries EI_REAL's ρ=−0.991 long-run improvement trend—77 years of efficiency gains dominate the level, making the 2026 reading of 0.88 a trend-shape artifact of "loose constraint" (level shadow pathology: gold raw −0.79 → year-controlled still +0.30). v2.3 kills the trend shape by construction: all pathology checks pass—T no (detrending is orthogonal to time by construction) / G no (gold raw +0.17) / signal variance ×8 (0.189 vs 0.024) / non-calendar mapping ρ(S,t) = −0.44. Core finding—the 2026 reading flips direction: the detrended slot reads 7.71 "tight"—the efficiency dividend has stalled for 11 years (2015-2026) with EI_REAL +10.4% above its long-run trend, the highest since 1949; RCS-E moves 1.67 → 2.83, measured total 4.83 → 5.42. Eight-checkpoint replay: Δ turns positive from 2000 and widens monotonically (2008 +0.68 → 2026 +1.44); pre-crisis turning occurs only in 2008 (honest disclosure: a state measure, not an early-warning device). Boundaries: the +0.87 co-movement with gold 2000-25 is genuine macro linkage (not an artifact; causal reading prohibited); the log-detrended variant was dropped for sign instability; the Eq. (25) DD linkage of 80.3% is narrative-layer coefficient extrapolation, for sensitivity reference only. Deployment: observation layer live (no production positions touched); the 0.75 total-score repair adjudication is pending per the 2026-09-05 directive, with the v2.3 measured total of 5.42 as the new component-recomputation baseline at that time.

当前读数：RCS=7.95（六轮最高）——分量显示值：能源 6.4 + 人口 8.0（等权合成 7.2；与总读数 7.95 的 +0.75 差额为 2026-09-04 审计发现的已知不一致，记入修订清单待溯源，本版不擅自取舍）。关键发现：能源约束的形态从"石油不够"切换为"电力不够"。（能源 6.4 为前瞻口径；实测月度口径 RCS-E 三代读数见 sim-us-macro 面板双轨列与式 (18b)/(18c)：v2.2 水平位 2026-05 = 1.67 / v2.3 去趋势位 = 2.83；人口腿 P ≈ 8.0，v2.3 实测总分 5.42，全史分布 p50=2.76。）

Current reading: RCS=7.95 (highest of six cycles) — component display values: Energy 6.4 + Demographics 8.0 (equal-weight composite 7.2; the +0.75 gap vs. the headline 7.95 is a known inconsistency found in the 2026-09-04 audit, logged in the revision list pending traceability—this version does not adjudicate). Key finding: the form of energy constraint has shifted from "not enough oil" to "not enough electricity." (Energy 6.4 is the forward-looking basis; the measured monthly RCS-E has three generations on the sim-us-macro dual-track column per Eqs. (18b)/(18c): v2.2 level 2026-05 = 1.67 / v2.3 detrended = 2.83; demographics leg P ≈ 8.0, v2.3 measured total 5.42, full-history p50 = 2.76.)

能源约束 v2.1 六分量完整规格 / Energy v2.1 Six-Component Specification

A_idr 的设计原理：进口依赖率的绝对水平有口径误导（美国原油净进口 40% 但总能量净出口），但其二阶差分可捕捉趋势反转——当 IDR 从下降转为上升，加速度提前转正，领先于 G 与 D。2026 年 A_idr = 8.67（IDR 从 −5% 反弹至 −3%，AI 电力需求驱动），正确提升了约束得分。六分量对五轮历史的回检：

A_idr design rationale: the absolute level of import dependence is definitionally misleading (US crude net imports 40% yet a net total-energy exporter), but its second difference captures trend reversal—when IDR turns from falling to rising, the acceleration turns positive ahead of G and D. The 2026 A_idr reads 8.67 (IDR rebounding from −5% to −3%, AI power demand), correctly raising the stress score. Backcheck across five historical points:

口径标注（2026-08-27 勘误）：上表为文献估计版——G 列（1973 +0.35 / 2007 +0.60）为石油净进口口径，而式 (18) 规格原文为「一次能源物理平衡」（EIA 实测：1973 +0.165 / 2007 +0.299）；2026 行的 D_e=+3.0% 与 A_idr=+2.3 为前瞻口径（AI 数据中心电力需求规划增速 vs 电网并网速度、原油进口依赖预期反弹），EIA 实测口径下两者分别为 −0.56pp 与 −2.59pp。真实数据回放见下节。

Definition note (2026-08-27 erratum): the table above is the literature-estimate version—the G column (1973 +0.35 / 2007 +0.60) uses the petroleum net-import basis, whereas the spec text of Eq. (18) defines G on total primary energy (EIA measured: 1973 +0.165 / 2007 +0.299); the 2026 row's D_e=+3.0% and A_idr=+2.3 are forward-looking (planned AI datacenter power demand vs grid interconnection, expected rebound in crude import dependence), while EIA measured values are −0.56pp and −2.59pp respectively. See the real-data replay below.

六分量真实数据回放（EIA/FRED，2026-08-27 补全）/ Six-Component Real-Data Replay (EIA/FRED, added 2026-08-27)

表二 o7 项落地：用 EIA 总能源月报（T01.02/T01.03/T07.01/T03.01，1949–2025）、EIA 原油首购价（1900–2025）与 FRED WTI（2026 YTD）回放全部六分量（量化规则定版见 academic/11-rcs-energy-backcheck.md）。九时点回放：

Table-2 item o7 implemented: all six components replayed with EIA Monthly Energy Review (T01.02/T01.03/T07.01/T03.01, 1949–2025), EIA crude first-purchase price (1900–2025), and FRED WTI (2026 YTD); quantified rules finalized in academic/11-rcs-energy-backcheck.md. Nine-point replay:

判别力结论（verdict=PARTIAL，J1–J6 五过一挂）：① 1973-74 实体供给冲击由六分量准确捕捉（v1=1.62/4.22 vs v2.1=4.98/5.01，全历史 top5）——油价管制下名义年均价仅 1.24×10 年均值，价格分量系统性漏报，G/D/A_idr（进口依赖 21.5%→34.8%，二阶差分 +3.19pp）完成捕捉；② 2005-08 页岩前依赖峰值（IDR≈60%）进 top10（2005 为全历史最高 5.54）；③ 1929/2021 无约束期正确低读。代价与边界：物理分量粘性导致危机后 3–5 年残留信号（1985-88 v2.1≈4.5–4.7 而油价已崩至 0.6–1.2×）；EROI 为文献插值基线（现代读数常年偏高）——该缺陷已于 2026-09-03 通过 v2.2 双轨部署修复：能源质量位替换为实测能耗强度 EI_REAL（式 18b），EROI 与年份 ρ=+1.000 的完美时间趋势被证伪为插值伪影（控年份后与金价相关反号：raw −0.905 → +0.095），EI_REAL 控年份 +0.071≈零（G2 检验，无金价伪影载体）；月度部署闸门 M1 通过 / M2 打平非恶化（Δ−0.001，三辅助证据支持不劣）/ M3 全月度披露；θ₂ 等价线 4.85 对齐 v2.1 的 5.0 同分位；判别主窗口按油价冲击定义与 v1 同源，AUC 对价格分量天然有利。两轮危机形态互补——1973 实体供给型看六分量、1979 价格型看价格——这正是价格权重自 v1 的 0.50 降至 v2.1 的 0.20 并保留 0.20 的实证依据。2026 读数双口径：EIA 实测 2.40（油价 1.31×、净出口加深、发电快于售电）vs 前瞻口径 6.4（AI 电力规划剪刀差）；AI 效应在 EIA 实测口径尚不可观测，待数据中心用电分项数据发布后收敛。验证细节与诚实披露 D1–D5 见 academic/11-rcs-energy-backcheck.md。

Discrimination verdict (PARTIAL, J1–J6: five pass, one boundary fail): (i) the 1973–74 physical supply shock is caught by the six components (v1=1.62/4.22 vs v2.1=4.98/5.01, all-time top-5)—under price controls the nominal annual average was only 1.24× the 10-year mean, so the price component systematically missed it while G/D/A_idr (import dependence 21.5%→34.8%, second difference +3.19pp) captured the shock; (ii) the pre-shale dependence peak of 2005–08 (IDR≈60%) enters the top-10 (2005 is the all-time maximum, 5.54); (iii) 1929/2021 correctly read low. Costs and boundaries: physical-component stickiness leaves 3–5 years of residual signal (1985–88 v2.1≈4.5–4.7 with oil already collapsed to 0.6–1.2×); EROI is a literature-interpolated baseline (permanently elevated in the modern era)—this defect was repaired on 2026-09-03 via the v2.2 dual-track deployment: the energy-quality slot is replaced by the measured energy intensity EI_REAL (Eq. 18b); EROI's perfect time trend vs year (ρ=+1.000) was falsified as an interpolation artifact (year-controlled gold correlation flips sign: raw −0.905 → +0.095), while EI_REAL's year-controlled +0.071 ≈ zero (G2 test, no gold-artifact carrier); monthly deployment gates M1 passed / M2 tie-not-deterioration (Δ−0.001, three auxiliary lines support non-inferiority) / M3 full-monthly disclosed; the θ₂ equivalent line 4.85 aligns with v2.1's 5.0 at the same quantile; the primary discrimination windows are oil-shock-defined and thus price-sympathetic, favoring v1's AUC by construction. The two shocks are complementary—1973 physical-supply type favors six components, 1979 price type favors price—empirically justifying the weight reduction of price from 0.50 (v1) to a retained 0.20 (v2.1). 2026 dual-basis reading: EIA measured 2.40 vs forward-looking 6.4; the AI effect is not yet observable in EIA measured data, pending datacenter-electricity breakdowns.

BII / RCS 口径演化与多轨读数审计（2026-09-05 增补） / Index Caliber-Evolution and Multi-Track Audit

口径演化披露：BII 与 RCS 自本文冻结（2026-08-27）以来各自完成「连续化 → 分量替换/校正」演化，全部节点如下表。本文正文叙事值（BII 6.74 / RCS 7.95，60 事件人工校准 + ex-ante 情景）与实测管道读数（BII v3 / RCS v2.3）分属不同序列、禁止直接对比——前者为式 (25) 标定层，后者为研究监测层。

泡沫时点双轨对照（人工校准/叙事 vs 实测连续序列，v3 + 能源腿对应版本）：

2000 年两点吻合，2000 年后人工校准值系统性高于实测序列（2021 年 BII 差 3.35）——与 RCS 侧「叙事数 vs 实测数」的结构性差异独立复现。已知缺口如实披露：BII 的 N 维（0.20 权重）无逐月源悬空、V 维仅 PE10 单因子近似、C 维以银行信贷近似私人信贷——6.74 → 4.66 的差异由此三层叠加（人工校准偏高 + 分量缺省 + m1 bug 历史已修正）。换腿探针（2026-09-05）：连续序列 v3 的 RCS 能源腿原为 v2.1（口径滞后）；换腿探针（outputs/bii_rcs_frag_continuous_v3_e23_probe.json，A1 基线与 canonical 366/366 逐位一致）测得 full_v2.3 与 legacy 相关 0.946、当前读数 5.44 vs 5.49（读数稳健）；四态分离出现分化——crash|slow_rec 的 d′：v2.1 −1.01 / v2.2 −1.71（最强）/ v2.3 −0.89，v2.3 的慢衰退分离弱化（slow_rec|prosperity 0.06 vs v2.1 0.235）——与 D_e/A_idr 重校准探针（区间失配修复，var ×14.6/×2145）合并观察后已于同日走完规格五环：v2.3R 转正（48 号起诉书 · 观测层正读数，v2.1/v2.2/v2.3 降历史对照）；四态分离 crash|slow_rec d′：v2.3R −1.44（vs v2.3 −0.89 恢复分离，仍弱于 v2.2 −1.71）、slow_rec|prosperity 0.282（vs v2.3 0.06 恢复）；第四对 slow_rec|shallow v2.3R −0.101 弱于 v2.3 −0.455，但该对四轨符号不稳属噪声主导（起诉书 §3 勘误：「四对全优」不成立）。θ 四轨 5.0/4.85/4.59/5.71——v2.3R 的 5.71 线 1996+ 触发月恰为 2008-05/06/07（GFC 油价顶峰），同时修正 v2.1 假阳（2005/2007 EROI 伪影段）与 v2.2 假阴（现代窗口零触发）。式 (25) DD 参考链：论文 65.8% → v2.1 80.0 / v2.2 83.6 / v2.3 80.3 / v2.3R 80.3（系数在叙事校准层值域 0.36~7.95 拟合，实测 5.4 附近代入属域外推已知失真，仅供敏感性参考）。

Caliber-evolution disclosure: since the paper freeze (2026-08-27), BII and RCS have each undergone "continuization → component replacement/correction" (full timeline table above). The narrative values (BII 6.74 / RCS 7.95, 60-event manual calibration + ex-ante scenarios) and the measured pipeline readings (BII v3 / RCS v2.3) belong to different series and must not be compared directly—the former is the Eq. (25) calibration layer, the latter the research monitoring layer. Bubble-checkpoint comparison: the two calibers agree in 2000 but the manual values run systematically above the measured series afterward (2021 BII gap 3.35), independently replicating the RCS narrative-vs-measured structural gap. Disclosed gaps: BII's N dimension (0.20) has no monthly source (four-dimension /0.80 renormalization), V is single-factor PE10, C approximates private credit with bank credit. Leg-swap probe (2026-09-05): full_v2.3 correlates 0.946 with legacy, current reading 5.44 vs 5.49 (robust); four-state separation diverges—crash|slow_rec d′: v2.1 −1.01 / v2.2 −1.71 (best) / v2.3 −0.89, with v2.3's slow-recession separation weakened—merged with the D_e/A_idr recalibration probe, the five-ring process concluded the same day: v2.3R promoted (indictment No. 48; observation-layer primary reading, v2.1/v2.2/v2.3 demoted to historical reference). Four-state separation, crash|slow_rec d′: v2.3R −1.44 (restoring separation vs v2.3's −0.89, still below v2.2's −1.71), slow_rec|prosperity 0.282 (vs v2.3's 0.06); the fourth pair slow_rec|shallow is sign-unstable across all four tracks (noise-dominated, per indictment §3 erratum: "all-four-pairs dominance" does not hold). θ across tracks 5.0/4.85/4.59/5.71—v2.3R's 5.71 line triggers exactly at 2008-05/06/07 (GFC oil peak), simultaneously correcting v2.1's false positives (2005/2007 EROI-artifact segments) and v2.2's false negatives (zero triggers in the modern window). Eq. (25) DD reference chain: paper 65.8% → v2.1 80.0 / v2.2 83.6 / v2.3 80.3 / v2.3R 80.3 (coefficients fitted on the narrative calibration range 0.36~7.95; substitution at measured ~5.4 is a known out-of-domain extrapolation, sensitivity reference only).

人口约束 v3 双层结构 / Demographic v3 Two-Tier Structure

核心判断：土地不是周期变量而是上限变量。人均土地在单轮泡沫周期内几乎不动——若作为第六个等权周期分量，将产生"常年饱和或无信号"与"角色错位"两个坏结果。正确做法是双层纳入：上限修正层（决定 RCS 在哪饱和）+ 泄压流量层（可周期化的新指标 D_land 与 V_land，分别对标能源 D 与移民分量）。国别饱和上限与 v2 → v3 回检：

The core judgment: land is an upper-bound variable, not a cyclical one. Per-capita land barely moves within a single bubble cycle—as a sixth equal-weight component it would yield either permanent saturation or no signal, plus role misallocation. The correct treatment is two-tier: an upper-bound correction layer (where RCS saturates) and a pressure-relief flow layer (the cycle-able indicators D_land and V_land, mirroring energy-D and the migration component). Country saturation caps and the v2 → v3 backcheck:

三大结构性发现：① 中国读数修正（8.0 → 8.6）是本次修订的最大单一发现——大豆进口同时是能源负债、土地负债与水负债；② 虚拟土地贸易与隐含能源贸易同构同源（粮食 = 土地 + 水 + 化肥能源的三重封装），三份备忘录形成完整计量链条；③ 人口与土地独立成账——俄罗斯与中国的 RCS 读数相近时，出清路径与政策空间完全不同。数据源：FAO FAOSTAT / AQUASTAT + USDA 贸易数据 + UN WPP 2024。

Three structural findings: (1) the China revision (8.0 → 8.6) is the revision's largest single discovery—soybean imports are simultaneously an energy, land, and water liability; (2) virtual-land trade is isomorphic to embodied-energy trade (food = land + water + fertilizer-energy, triple-packaged), completing the measurement chain across the three memos; (3) demographics and land are independent ledgers—Russia and China at similar RCS readings face entirely different clearing paths and policy space. Sources: FAO FAOSTAT / AQUASTAT, USDA trade data, UN WPP 2024.

4.8 S'''三笔可持续度：三阶修正的完整推导 / S''' with Three-Tier Corrections

4.8.1 基础定义 / Base Definition

4.8.2 第一阶：Gini三通道分配不均修正 / Tier 1: Gini Three-Channel Correction

4.8.3 第二阶：Omega底层空心化修正 / Tier 2: Bottom-Hollowing Correction

口径脚注 / Calibration Footnote：Ω 存在两种分解粒度，本文主线采用分量分解法（式 22）：ω_F·ΔF_bottom + ω_L·ΔL_bottom + Ω_存量 = 0.38，经中间态 S'' = −2.97 收敛到 S''' = −3.35；维度配分法口径取 Ω = 0.15，从三笔原始配分（E = +1、P = −2、M = +3）直接合成：S''' = (1−2)/2 − 3 + 0.15 = −0.5 − 3 + 0.15 = −3.35。两条路径的终值与最终 δ(S''') 完全一致，差异仅在中间态的分解粒度——全体系披露版将其并列存档以保持与站点历史口径的可追溯性。

4.8.4 第三阶：gamma-GSM黄金压力修正 / Tier 3: Gold Stress Modification

gamma-GSM的经济逻辑：黄金3年CAGR突破20%阈值，表明市场对法币信心的折价进入非线性加速区。1973年（gold 3yr CAGR=39.6%, GSM=1.96）和2026年（33.1%, GSM=1.31）是历史上仅有的两次GSM触发。GSM通过修饰kappa(G)来影响清算破坏力——kappa'(G,GSM)=3.92意味着在当前分配不均基础上，法币信心折价进一步放大了13%。

Economic logic of gamma-GSM: when the 3-year gold CAGR breaks the 20% threshold, it signals that the market's discount on fiat confidence has entered a nonlinear acceleration zone. 1973 (gold 3yr CAGR=39.6%, GSM=1.96) and 2026 (33.1%, GSM=1.31) are the only two historical GSM triggers. GSM affects liquidation destructiveness by modifying kappa(G)—kappa'=3.92 means fiat confidence discount further amplifies by 13% on top of existing inequality amplification.

4.8.5 delta(S'')清算破坏力函数的完整形式 / Complete delta(S'') Function

4.9 支付体系脆弱性 FRAG / 4.9 Payments Fragility FRAG

FRAG（0–10）度量清算传染的制度缓冲缺失：金本位下的 1929 年 FRAG = 10（银行系统裸奔、无存款保险），随存款保险、最后贷款人、货币互换网络的建立单调下降，2021 年后稳定于中低区间。其时间形式由式 (D1) 给出。FRAG 是回撤方程中敏感度第一的参数（§5.4 蒙特卡洛 Tornado 排序第一，约 ±9pp）。

FRAG (0–10) measures the institutional absence of contagion buffers: 1929 under the gold standard scores FRAG = 10 (an unbacked banking system, no deposit insurance), declining monotonically as deposit insurance, lender-of-last-resort, and swap networks were built, and stabilizing at mid-low levels after 2021. Its time form is given by Eq. (D1). FRAG is the most sensitive parameter in the drawdown equation (ranked first in the §5.4 Monte Carlo Tornado, ~+/-9pp).

该指数衰减形式编码了制度建设的累积记忆：每约 35 年，金本位时代的裸奔脆弱度衰减 63%。2026 年外推读数 FRAG ≈ 2.7——存款保险、贴现窗口与央行互换网络把 1929 式的传染链条制度性切断，这正是"冻结系数"理论的微观制度基础，也是 §5 回撤方程中 FRAG 与 BII×RCS 交互项系数为负的来源。

The exponential decay encodes the cumulative memory of institutional construction: every ~35 years, the gold-standard era's raw fragility decays by 63%. The 2026 extrapolated reading is FRAG ≈ 2.7—deposit insurance, discount windows, and central-bank swap networks institutionally sever the 1929-style contagion chain. This is the micro-institutional foundation of the "freeze coefficient" theory, and the source of the negative interaction coefficient on FRAG with BII×RCS in the §5 drawdown equation.

4.9.1 双谱系澄清与 inst_mem_10 改名（2026-09-02 增补）/ 4.9.1 Dual-Lineage Clarification and the inst_mem_10 Renaming

披露增补：全体系历史上同时存在两个 "frag"，语义相反，2026-09-02 起以改名彻底切割——谱系 A（大写 FRAG，即本节正文）：方法论脆弱度，1929 = 10、2021 = 2.5（解冻底）、2026 = 7.2，语义「越高越危险」，为式 (25) 中敏感度第一的参数；谱系 B（原小写 frag_10）：连续序列 canonical 中的制度记忆时间占位——式 (D1) 的平线形式（1996-01 = 3.61 → 2026-06 = 2.96，全窗口跨度仅 0.64），测度「制度建设记忆」而非脆弱度，与谱系 A 的「越高越危险」语义相反。处置：谱系 B 字段更名为 inst_mem_10（institution-memory），数值零改动，向后兼容别名保留——把这一「假 FRAG」从命名上退役，杜绝谱系混用。

Disclosure supplement: the system historically carried two "frags" with opposite semantics, severed by renaming as of 2026-09-02—Lineage A (upper-case FRAG, i.e., this section): the methodological fragility, 1929 = 10, 2021 = 2.5 (thaw floor), 2026 = 7.2, semantics "higher = more dangerous," the most sensitive parameter of Eq. (25); Lineage B (formerly lower-case frag_10): the institutional-memory time placeholder in the continuous canonical series—the flat form of Eq. (D1) (1996-01 = 3.61 → 2026-06 = 2.96, full-window span only 0.64), measuring "institutional memory of construction," not fragility, with semantics opposite to Lineage A. Disposition: Lineage B is renamed inst_mem_10 (institution-memory), values unchanged, backward-compatible alias retained—retiring this "false FRAG" nominally to preclude lineage mixing.

4.9.2 FRAG_C_s2 奥派累积脆弱度与 L12 流动性脉搏观测哨（2026-09-02 增补）/ 4.9.2 FRAG_C_s2 and the L12 Liquidity-Pulse Sentinel

在谱系澄清基础上，以大写 FRAG 脆弱度为主基底叠加奥派「施政空间耗竭/债务不可逆」上升项，重建 366 月脆弱度研究序列 FRAG_C（两臂：流存量均衡臂 = 诚实基线；FRAG_C_s2 存量加权臂 = 奥派/方法论对齐臂，经五环流程转正为只读观测哨读数）。核心构造发现：奥派「施政空间耗竭」本质是存量（债务复利累积 + QE 深度），而非 12 个月流量——2026-05 流重臂塌回 4.31（12M 信贷增速放缓）而存量加权臂仍持 7.44，精准贴近方法论脆弱度 7.2；若只计流量，会在放水暂停后误报安全。这一「脆弱度被喂养上行的复利结构」正是旧式式 (D1) 时间占位（贴地 ~3.0）永远捕捉不到的维度。

Upon the lineage clarification, the upper-case FRAG serves as the base with Austrian "policy-space exhaustion / debt irreversibility" add-on terms, rebuilding the 366-month fragility research series FRAG_C (two arms: a flow-stock-balanced arm as the honest baseline; the FRAG_C_s2 stock-weighted arm, aligned with the Austrian/methodological reading and promoted—via the five-ring process—to a read-only sentinel reading). The core constructional finding: Austrian "policy-space exhaustion" is fundamentally a stock (compounding debt accumulation plus QE depth), not a 12-month flow—in 2026-05 the flow-weighted arm collapses to 4.31 (slowing 12M credit growth) while the stock-weighted arm holds 7.44, precisely proximate to the methodological fragility 7.2; counting flows alone would falsely report safety after liquidity pauses. This "compounding structure of fragility being fed upward" is exactly the dimension the legacy Eq. (D1) time placeholder (hugging ~3.0) can never capture.

三分量的奥派语义：$\overline{\text{drain}}_{36m}$ 是施政空间耗竭的时间累积形式（半衰期 3 年编码「消耗事件的影响随时间衰减但不即逝」）；$\Delta_{credit}$ 是债务不可逆/大而不倒深度（累计增量、只上不下）；$m_{cb}$ 是 QE 冻结能力深度。与 L12（12 个月准备金供给变化）构成双读数：s2 度量累积危险（年尺度），L12 度量当期流动性脉搏（月季尺度）。严格 no-look-ahead：每分量只用截至当月数据。

Austrian semantics of the three components: $\overline{\text{drain}}_{36m}$ is the time-cumulated form of policy-space exhaustion (the 3-year half-life encodes "drawdown events fade with time but do not vanish"); $\Delta_{credit}$ is debt irreversibility / too-big-to-fail depth (cumulative increment, upward-only); $m_{cb}$ is QE freeze-capacity depth. Together with L12 (12-month reserve-supply change) they form dual readings: s2 gauges accumulated danger (annual scale) while L12 gauges the current liquidity pulse (monthly-quarterly scale). Strict no-look-ahead: every component uses data up to the current month only.

关键月读数与奥派时滞的诚实呈现：

Key-month readings, with the Austrian time-lag honestly presented:

注意「2021-01 峰值」的奥派读法：当期显性回撤小（繁荣态），但 FRAG_C_s2 已把递延危险标到最高——放水越多、越大而不倒、越危险，冻结的清算在复利地积累。状态排序的奥派时滞同样诚实呈现：慢衰退期均值（5.16）低于繁荣期（6.25），因慢衰退发生在债务开始去化之后，而繁荣样本含 QE 峰值期——脆弱度在放水期积累、在真实慢校正时反而读数回落，这是机制本性而非缺陷。

Read the "2021-01 peak" the Austrian way: contemporaneous explicit drawdown was small (prosperity state), yet FRAG_C_s2 already marked the deferred danger at its maximum—more easing, more too-big-to-fail, more danger: the frozen clearing compounds. The Austrian lag in state ordering is equally honest: the slow-recession mean (5.16) sits below prosperity (6.25) because slow recessions occur after deleveraging begins while prosperity samples include the QE peak—fragility accumulates during easing and reads lower during genuine slow corrections; this is mechanism, not defect.

回测证据（研究层，s2 分位 ≥ 阈值 → wait/exit，hold 段优先级 E10 > s2 > any1）：

Backtest evidence (research layer; expanding-percentile s2 ≥ threshold → wait/exit; hold-segment priority E10 > s2 > any1):

参数稳健性：β ∈ [0.15, 0.35] 五档全正增益（24.17%–24.57%，Sh 1.835–1.866）；触发分布分散（p80 触发 51 月，分布于 2003–2023 每年 2–4 月），改善来自高脆弱期的系统性防御配置而非精准逃顶。象限条件收益（s2 分位 80% × L12 正负）呈完美单调性：

Parameter robustness: β ∈ [0.15, 0.35] yields positive gains across all five tiers (24.17%–24.57%, Sh 1.835–1.866); trigger months are dispersed (p80 fires 51 months, 2–4 per year over 2003–2023)—the improvement comes from systematic defensive allocation during high-fragility phases, not precise top-calling. Quadrant-conditional returns (s2 percentile 80% × L12 sign) show perfect monotonicity:

三重检验的综合裁定（verdict: PARTIAL）：检验 A（事件前预警）4/4 命中——US 危机顶前 12 月内 s2 分位极大值：2007 次贷 100%、2021 QE 泡沫 98.3%、2018 Q4 95.8%、2013 Taper 88.6%（2000 互联网泡沫分位不可算，诚实排除）；检验 B（信息量）FAIL——高分位月 P(后续 18 月 NDX MDD ≤ −12%) = 49.8% vs 基率 43.8%（lift 1.14 < 1.2），精确逃顶判别力温和；检验 C（L12 相位）2/3——2018 缩表前 −2.32 ✓、2019 回购危机前 −2.29 ✓、2022 熊市顶前 +3.60 ✗（QE 海啸释放型——「快崩两相位：收缩窒息型 vs 释放冻结型」的第三例确认）。PARTIAL 恰好支持其定位：若 s2 是强预测器就有理由升级为交易信号；PARTIAL 说明它就该是只读观测哨。式 (25) 对齐性旁证：5.62 × s2 于 2026-05 = 41.8pp，精准对齐方法论基准 5.62 × 7.2 = 40.5pp，消除旧 inst_mem 占位（16.7pp）+17pp 的量级错位——但式 (25) 保留原 frag 口径不动：其预测目标是当期回撤深度，而 s2 度量累积危险，两者测度对象不同，混入将引入「当期化 vs 累积」的语义分歧。

Composite verdict across three tests (PARTIAL): Test A (pre-event warning) 4/4 hits—maximum s2 percentiles within 12 months before US crisis tops: 2007 subprime 100%, 2021 QE bubble 98.3%, 2018 Q4 95.8%, 2013 taper 88.6% (2000 dot-com percentile not computable, honestly excluded); Test B (information content) FAILS—high-percentile months' P(NDX MDD ≤ −12% within 18 months) = 49.8% vs base rate 43.8% (lift 1.14 < 1.2), mild top-calling discriminability; Test C (L12 phase) 2/3—pre-2018-QT −2.32 ✓, pre-2019-repo −2.29 ✓, pre-2022-bear +3.60 ✗ (QE-tsunami release type—the third confirmation of "two crash phases: contraction-asphyxiation vs release-freeze"). PARTIAL precisely supports the positioning: were s2 a strong predictor there would be grounds to promote it to a trading signal; PARTIAL means it belongs as a read-only sentinel. An Eq.-(25) alignment aside: 5.62 × s2 at 2026-05 = 41.8pp, precisely aligned with the methodological benchmark 5.62 × 7.2 = 40.5pp, eliminating the +17pp magnitude misalignment of the old inst_mem placeholder (16.7pp)—yet Eq. (25) retains the original frag caliber: its prediction target is contemporaneous drawdown depth, whereas s2 measures accumulated danger; the two differ in measurement object, and mixing them would import the "contemporaneous vs cumulative" semantic divergence.

定位声明（路径 A，五环全流程落地）：FRAG_C_s2 与 L12 为双读数只读观测哨——不进式 (25)、不进状态机、不进五票、不触发任何调仓。五环对账：canonical s2 vs 冻结原型 366/366 逐位一致；v3 原字段零改动（JSON 级独立复验 4×366/366）；手工六观察点锚值复现 6/6；幂等复跑 md5 一致。A1–A5 断言成为月度管道常驻门禁。月度刷新链序：FRED 七系列 → v1 → m_v2 实验 → v2 → v3（A1–A5 门禁）→ pulse → 读数摘要。

Positioning statement (Path A, full five-ring process): FRAG_C_s2 and L12 are dual-readout read-only sentinels—not entering Eq. (25), the state machines, the five votes, or any rebalancing. Five-ring reconciliation: canonical s2 vs the frozen prototype 366/366 identical; v3 original fields untouched (JSON-level independent re-verification 4×366/366); six manual observation-point anchors reproduced 6/6; idempotent reruns md5-identical. The A1–A5 assertions serve as standing monthly-pipeline gates. Monthly refresh chain: FRED seven series → v1 → m_v2 experiment → v2 → v3 (A1–A5 gates) → pulse → reading digest.

4.10 繁荣质量指数 PQI / 4.10 The Prosperity Quality Index (PQI)（2026-09-02 增补）

动机与互补定位：BII 度量「泡沫有多大」（水平），PQI 度量「繁荣有多健康」（质量）。构造动因有三：其一，本体系的 prosperity 状态系残余类别（非衰退且回撤 > −15% 即繁荣），占 366 月样本的 66%（302 月）而内部零分辨率——2009-04 复苏初（BII v3 = 1.30）与 2021-11 泡沫顶（5.48）同标签，读数差 4.2 倍；其二，繁荣带读数 P10–P90 [2.34, 4.45] 与慢衰退带 [2.19, 4.57] 完全重叠，BII 单向性在低值端无锚定；其三，式 (25) 只预测回撤深度，繁荣期无预测目标（持续时长/结束方式/软硬着陆）。PQI 以三分量正交侧面合成 [0,1] 月度读数：实体增长 × 盈利验证 × 融资成本，低读数 = 至少两个侧面同时恶化。三顶验证显示其与 BII 的互补结构：2007-10 低 BII 信贷质量型顶（BII 周期峰仅 3.37，远低于繁荣带 P90 门槛）由 PQI 唯一捕捉——质量温度计补上了水平温度计的结构性盲区。

Motivation and complementary positioning: BII measures "how large the bubble is" (level); PQI measures "how healthy the prosperity is" (quality). Three motivations: first, the prosperity state of this system is a residual category (non-recession with drawdown > −15%), covering 66% of the 366-month sample (302 months) with zero internal resolution—2009-04 early recovery (BII v3 = 1.30) and 2021-11 bubble top (5.48) share one label with a 4.2× reading gap; second, the prosperity reading band P10–P90 [2.34, 4.45] fully overlaps the slow-recession band [2.19, 4.57], leaving BII unanchored at the low end; third, Eq. (25) predicts only drawdown depth—prosperity has no prediction target (duration, ending style, landing softness). PQI synthesizes a [0,1] monthly reading from three orthogonal facets—real growth × earnings verification × financing stress; a low reading means at least two facets deteriorate simultaneously. Three-top validation shows the complementarity with BII: the 2007-10 low-BII credit-quality top (BII cycle peak merely 3.37, far below the prosperity-band P90 threshold) is captured by PQI alone—the quality thermometer covers the level thermometer's structural blind zone.

样本纪律（先于构造声明）：prosperity 名义 302 月，但独立繁荣段仅约 4–5 段（1996–2000 / 2003–2007 / 2009–2019 / 2020–2021 / 2023–今）——PQI 全系列禁回归类构造，只走规则式合成 + 描述性基线；全部统计量（z-score、分位）取 expanding 口径（no-look-ahead，只用 ≤ 当月数据）。用途边界：研究读数 + 报告层，零仓位改动。

Sample discipline (declared before construction): prosperity spans nominally 302 months but only ~4–5 independent prosperity segments—PQI in its entirety forbids regression-type construction, permitting only rule-based synthesis and descriptive baselines; all statistics (z-scores, percentiles) use expanding calibers (no-look-ahead, data ≤ current month only). Usage boundary: research reading and reporting layer, zero position impact.

三分量的理论锚定：Q1 编码「增长跑赢信贷 = 可持续繁荣，信贷超前 = 庞氏段临近」——Minsky 金融不稳定假说的位移刻画 [5] 与 BIS 信贷缺口传统（credit-to-GDP gap 为官方早期预警指标）[29] 的正交合成，三顶中它是 2006–08 唯一持续早警的分量；Q2 直接度量回报分解中盈利增长对估值重估的覆盖度——「指数涨幅中有多少是盈利背书的」[30]；Q3 是繁荣的「债务端体检」：信用利差为违约风险溢价的实时定价 [31]，保证金债务增量为躁动期杠杆拥挤的计量 [27]，两者同步出现时繁荣的燃料端先于价格见顶。数据：FRED GDP（季度月度化）、A091 名义企业利润、NASDAQ、BAA−AAA（z 预喂自 1919）、FINRA 保证金债务；无效分量剔除（至少 2 个有效才输出读数，数据缺腿月份不硬算）。

Theoretical anchors of the three components: Q1 encodes "growth outpacing credit = sustainable prosperity; credit outrunning growth = Ponzi phase approaching"—an orthogonal synthesis of the Minskian displacement [5] and the BIS credit-gap tradition (credit-to-GDP gap as the official early-warning indicator) [29]; among the three tops it is the only component persistently pre-warning through 2006–08. Q2 directly measures the coverage of valuation re-rating by earnings growth in the return decomposition—"how much of the index rally is earnings-backed" [30]. Q3 is the prosperity's "debt-side physical": the credit spread prices default risk in real time [31], and margin-debt increments meter euphoria-phase leverage crowding [27]; when both appear together, the fuel side of the prosperity tops before prices do. Data: FRED GDP (quarterly, monthly-interpolated), A091 nominal corporate profits, NASDAQ, BAA−AAA (z pre-fed from 1919), FINRA margin debt; invalid components are dropped (a reading requires ≥2 valid components; months with missing legs are not force-computed).

Q2 反向毒药与条件化修复——本系列最重要的方法论发现：baseline 的 Q2 分量 d′ 符号为正（+0.42：低 PQI 组的前瞻回撤反而更浅），系 clamp 下限把两种截然不同的低 Q2 混读为 0：停滞型（1996–2000/2015：盈利平稳但跟不上估值，未来 12 月回撤中位 −7.1%）与塌方修复型（2009–13/2020–21：利润刚从底部 V 型弹起，未来 12 月中位 +1.1%）——修复型是买点，被 baseline 拖进「质量极差」阵营。修复（D4 核心改动）：低支撑 × 高波动联合触发时抬至 0.5 = 中性解除——不宣布高质量，只撤销「质量极差」的误判罚，把裁决权交还 Q1/Q3：

The Q2 reverse poison and its conditional repair—the series' most consequential methodological finding: the baseline Q2 component's d′ sign is positive (+0.42: the low-PQI group's forward drawdowns are shallower), because the clamp floor conflates two utterly distinct low-Q2 regimes: stagnation-type (1996–2000/2015: earnings flat but lagging valuation, forward-12M drawdown median −7.1%) and collapse-recovery-type (2009–13/2020–21: profits V-bouncing off the bottom, forward median +1.1%)—the recovery type is a buying opportunity dragged by the baseline into the "worst quality" camp. The repair (the D4 core change): when low support × high volatility jointly trigger, lift to 0.5 = neutral release—not declaring high quality, only rescinding the "worst quality" misjudgment and returning the verdict to Q1/Q3:

其理论依据是过度反应假说的宏观对应 [32]：剧烈盈利塌方后的高波动 = 均值回归窗口（修复期买点），而平静的低支撑 = 估值被缓慢透支的停滞危险。阈值 0.25/0.75 取自经验先验（低 Q2 组内波动分位条件中位 ≈ 0.733，无网格搜索）；敏感性扫描（0.20–0.30 × 0.70–0.80 五配置）证实方向性结论零依赖。触发集 35 个繁荣月全部落在修复期（2009×3 / 2010×3 / 2012×8 / 2013×12 / 2020×6 / 2021×3），互联网泡沫前夜 43 个低支撑月零混入。

The theoretical basis is the macroscopic counterpart of the overreaction hypothesis [32]: high volatility after a violent earnings collapse = a mean-reversion window (recovery-phase buying point), whereas calm low support = stagnation danger of slowly overdrawn valuations. Thresholds 0.25/0.75 are experiential priors (conditional median volatility percentile within the low-Q2 group ≈ 0.733, no grid search); sensitivity scans (0.20–0.30 × 0.70–0.80, five configurations) confirm zero dependence of directional conclusions. The trigger set's 35 prosperity months all fall in recovery phases (2009×3 / 2010×3 / 2012×8 / 2013×12 / 2020×6 / 2021×3), with zero contamination from the 43 low-support months preceding the dot-com top.

终版权重经三阶段预注册裁定：D6 的 DOM 判据（3/3 wins 零劣化）采纳 Q3×2 加权——警报器与判别器同步改善（2007-10 顶前 12M 警报命中 4 → 6/12，繁荣月报警率 0.125 → 0.133）；P4 组合验证（W1×R1 名义/实际口径组合）按预注册「组合不伤单项」判据边缘否决（excl_2020 口径劣化 0.025 > 0.02 容忍线），单项择优 = W1。每步单变量改动保证归因干净：D4 先换 Q2_cond，D6 再倍 Q3 权重。

The final weights were adjudicated through three pre-registered stages: D6's DOM criterion (3/3 wins, zero degradation) adopted the Q3×2 weighting—alarm and discriminator improving in sync (2007-10 pre-top 12M alert hits 4 → 6/12; prosperity-month alarm rate 0.125 → 0.133); P4 combination verification (W1×R1 nominal/real blend) was marginally rejected under the pre-registered "combinations must not hurt single arms" criterion (excl_2020 degradation 0.025 > the 0.02 tolerance line), single-arm optimum = W1. Each step is a single-variable change keeping attribution clean: D4 first swaps in Q2_cond; D6 then doubles the Q3 weight.

判别证据（全部可复现于落盘 JSON；三标准口径 = all / 排除政策急救月 / 排除 2020-05~08）：

Discrimination evidence (all reproducible from archived JSON; three standard calibers = all / excluding policy-rescue months / excluding 2020-05–08):

诚实边界与定位裁定：(i) 探针级信号——月度自相关使名义显著性高估，有效独立繁荣段仅 4–5 段（本节表内显著性按此折算解读）；(ii) 政策急救假阳性及其修复边界——2020-05~08（ZIRP+QE 使信贷脉冲 +11.7pp 被 Q1/Q3 惩罚、利润塌方压 Q2 至 0）PQI 读 0.03–0.15 而后续 12 月为大牛，「质量」概念在政策急救环境失真；Q2_cond 修复假警报 3.1 倍但全额豁免臂（A2）判别力不变——修复叙事、不修复判别；(iii) 「确认器不是警报器」——质量低点粘性口径（min 6M）判别力全场最强（d′ −1.046）但分位重校后警报命中率坍塌（2007-10 仅 1/12 vs W1 的 6/12）：低点粘性确认「质量已塌」而非先行警报，警戒线语义由此固定为确认型；(iv) 名义 GDP 通胀伪影——2022 滞胀年 Q1 虚高约 30 分位点，实际口径（CPI 平减）判别力一致改善，终版保留名义口径并以实际口径（R1）辅助并读（2026-06：R1 0.299 vs W1 0.256，差异即口径效应）；(v) 定位 = 质量温度计，非顶部择时器——与 BII（水平）、C4 状态机（阶段）构成三层互补。当前读数：2026-06 PQI′（W1）= 0.256，低于 0.30 警戒线，主导分量为 Q3 = 0.041（历史极低位：融资压力侧的警惕信号）。更新节奏：月频。

Honest boundaries and positioning verdict: (i) probe-grade signal—monthly autocorrelation inflates nominal significance; effectively independent prosperity segments number only 4–5 (read the table's significance accordingly); (ii) the policy-rescue false positive and its repair boundary—during 2020-05–08 (ZIRP+QE driving the credit pulse to +11.7pp, penalized by Q1/Q3; profit collapse pinning Q2 to 0) PQI read 0.03–0.15 while the subsequent 12 months were a raging bull—the "quality" notion fails in policy-rescue environments; Q2_cond repairs false alarms 3.1× but the full-exemption arm (A2) leaves discriminability unchanged—repairing the narrative, not the discrimination; (iii) "a confirmer is not an alarm"—the quality-trough-stickiness caliber (min 6M) has the strongest discrimination (d′ −1.046) but its alarm hit rate collapses after percentile recalibration (2007-10 only 1/12 vs W1's 6/12): trough stickiness confirms "quality has collapsed" rather than warning ahead—the alert-line semantics are hereby fixed as confirmatory; (iv) nominal-GDP inflation artifact—Q1 ran ~30 percentile points hot in the 2022 stagflation year; the real (CPI-deflated) caliber improves discrimination consistently; the final version retains the nominal caliber with the real caliber (R1) as an auxiliary co-reading (2026-06: R1 0.299 vs W1 0.256, the gap being purely the caliber effect); (v) positioning = a quality thermometer, not a top-timer—complementing BII (level) and the C4 state machine (phase) as a three-layer stack. Current reading: 2026-06 PQI′ (W1) = 0.256, below the 0.30 alert line, dominated by Q3 = 0.041 (a historical extreme low—a vigilance signal from the financing-stress side). Update cadence: monthly.

五、出清深度预测模型 v3.3 / 5. Drawdown Prediction Model v3.3

5.1 预测方程 / Prediction Equation

本节给出 v3.3 回撤预测方程的完整形式。该方程把第四章构建的 BII、RCS、FRAG 与 delta(S''') 四个指标合成为一个单一回撤深度预测。交互项 -17.59*(BII*RCS/10) 是对三笔耦合的数学承认：当 BII 与 RCS 同时很高时，清算破坏力不是线性相加——真实约束掐住名义修复的通道，形成非线性抵消。

This section presents the complete v3.3 drawdown prediction equation. It synthesizes the four indicators built in Chapter 4—BII, RCS, FRAG, and delta(S''')—into a single drawdown-depth prediction. The interaction term -17.59*(BII*RCS/10) is the mathematical acknowledgment of three-stock coupling: when BII and RCS are simultaneously high, clearing damage is not linearly additive—real constraints cap the nominal-repair channel, producing nonlinear offset.

2026 时点读数：BII=6.74，RCS=7.95，FRAG=5.0，delta(S''')=+13.13pp。代入方程：8.54(6.74) + 6.15(7.95) + 5.62(5.0) - 17.59*(6.74*7.95/10) + 13.13 = 65.8%。中位数 65.8%（v3.0 原为 59.6%，v3.3 修正 +6.2pp），90% 置信区间 58–73%。三情景分解：A 估值熊 -65%、B 信用熊 -76%、C 滞胀熊 -69%。关键结构判断：利率紧缩刺不破它——BII 的构成中货币宽松仅占 0.18 权重且读数不高；能源电力与人口掐着物理上限。^①

2026 readings: BII=6.74, RCS=7.95, FRAG=5.0, delta(S''')=+13.13pp. Substituting: 8.54(6.74) + 6.15(7.95) + 5.62(5.0) - 17.59*(6.74*7.95/10) + 13.13 ~ 65.8%. Median 65.8% (v3.0 read 59.6%; v3.3 corrections add +6.2pp), 90% CI 58-73%. Three scenarios: A valuation bear -65%, B credit bear -76%, C stagflation bear -69%. Key structural judgment: rate tightening cannot prick it—monetary easing carries only 0.18 weight in BII with a low reading; energy-electricity and demographics grip the physical ceiling.^①

5.2 校准方法 / Calibration Methodology

口径并注（2026-09-02，LOEO 替代验证）：式 (26) 为 n=4 精确拟合（4 方程 4 未知数 → R²=1.0，符号约束在 n=4 下空转）。全 57 事件无截距 OLS 的系数为 DD ≈ 2.46·BII − 1.19·RCS + 9.60·FRAG − 3.93·(BII×RCS/10)（pp 口径）——β₂(RCS) 符号翻转，违反式 (26) 约束；约束版（移除 RCS）为 DD ≈ 2.99·BII + 9.03·FRAG − 6.24·(BII×RCS/10)，符号满足且 LOEO ρ=0.610 略优于无约束版 0.582。三套权重的 2026 基础预测（不含 δ）收敛于 51.7~55.2pp——点预测对校准方式稳健，但不确定性陈述须按 LOEO 量级（MAE ≈ 15pp）而非参数扰动区间（±7.5pp）理解。详见 §12 局限 1 与 academic/35-loeo-drawdown-57.md。

Calibration annotation (2026-09-02, LOEO replacement): Eq. (26) is the n=4 exact fit (4 equations, 4 unknowns → R²=1.0; the sign constraints are vacuous at n=4). The full-57 no-intercept OLS coefficients are DD ≈ 2.46·BII − 1.19·RCS + 9.60·FRAG − 3.93·(BII×RCS/10) (pp scale) — β₂(RCS) flips sign, violating Eq. (26); the constrained variant (dropping RCS) is DD ≈ 2.99·BII + 9.03·FRAG − 6.24·(BII×RCS/10), satisfying signs with LOEO ρ=0.610 (marginally better than the unconstrained 0.582). All three weight sets converge on a 2026 base prediction (ex-δ) of 51.7~55.2pp — the point prediction is robust to calibration, but uncertainty must be read at the LOEO scale (MAE ≈ 15pp), not the parameter-perturbation interval (±7.5pp). See §12 Limitation 1 and academic/35-loeo-drawdown-57.md.

口径并注（2026-09-02）：式 (27) 的 RMSE_cv ≈ 9.0% 为 12 国点估计校准口径（用校准后的国别读数做点预测误差），非事件级全样本外推口径。57 事件 LOEO 的事件级误差量级为 MAE ≈ 15pp（v3.3 三因子）/ 8.4pp（7F 全特征），故「±9%」不可与「±15pp」互换引用——前者描述校准器对已知读数的拟合精度，后者才是对完全留出事件（含未见危机类型与国别）的真实泛化误差。对 2026 点预测的不确定性陈述，应采用 LOEO 量级（§12 局限 1），而非式 (27) 的 12 国拟合口径。

Calibration annotation (2026-09-02): The RMSE_cv ≈ 9.0% in Eq. (27) is the 12-country point-estimate calibration caliber (point-error over calibrated country readings), not the event-level full-sample extrapolation caliber. The event-level LOEO error across 57 events is MAE ≈ 15pp (v3.3 3-factor) / 8.4pp (7F full-feature), so "±9%" and "±15pp" are not interchangeable — the former measures the calibrator's fit on known readings, the latter is the true generalization error on fully held-out events (including unseen crisis types and countries). Uncertainty statements on the 2026 point prediction should use the LOEO scale (§12 Limitation 1), not the 12-country fit caliber of Eq. (27).

5.3 Forward Stagewise 特征选择 / Forward Stagewise Feature Selection

上述方程的形式来自理论推导，但系数校准和模型选择基于57个历史危机事件、29个国家的实证数据。我们构建了覆盖1929—2022年的危机回撤数据库，包含57个事件（发达经济体39个、新兴经济体18个），涉及股市泡沫、银行危机、货币危机、主权债务危机和全球系统性冲击五种类型。以BII、RCS、FRAG三个核心指标为基准3F模型（样本内rho=0.607），采用Forward Stagewise逐步回归进行特征选择，L0级事件（技术回调）赋以2倍权重以补偿小样本偏差。

The equation form is theoretically derived, but coefficient calibration and model selection are based on empirical data from 57 historical crisis events across 29 countries. We constructed a crisis drawdown database covering 1929–2022, comprising 57 events (39 in advanced economies, 18 in emerging markets), spanning stock bubbles, banking crises, currency crises, sovereign debt crises, and global systemic shocks. Using the 3F baseline model (BII, RCS, FRAG; in-sample rho=0.607), we employ Forward Stagewise stepwise regression for feature selection, with L0 events (technical corrections) weighted 2x to compensate for small-sample bias.

最终模型为7因子线性回归，样本内rho=0.9177，MAE=8.63pp，R2=0.8564。CRISIS_LEVEL（危机严重程度分级0/1/2）是最大增量贡献者（delta_rho=+0.247），表明危机的内生严重度比泡沫指标本身更能预测回撤深度。QE_DUMMY和QE_TAPER分别捕捉量宽政策的存在和退出效应，与第二章"冻结系数"的理论预期一致。CONTAGION（跨国传染评分）在特征搜索中未被选中（最佳rho=0.891，低于阈值），表明传染效应的信息已被CRISIS_LEVEL吸收。

The final model is a 7-factor linear regression with in-sample rho=0.9177, MAE=8.63pp, R2=0.8564. CRISIS_LEVEL (severity grading 0/1/2) is the largest incremental contributor (delta_rho=+0.247), indicating that endogenous crisis severity predicts drawdown depth better than bubble indicators themselves. QE_DUMMY and QE_TAPER capture the presence and exit effects of quantitative easing, consistent with the "freeze coefficient" theoretical expectation. CONTAGION (cross-country contagion score) was not selected (best rho=0.891, below threshold), indicating contagion information is already absorbed by CRISIS_LEVEL.

5.4 蒙特卡洛稳健性检验 / Monte Carlo Robustness

对2026时点的全部输入参数施加 +/-20% 独立扰动，N=10,000次模拟：P5/P50/P95 = 58%/65.8%/73%，90%区间宽度约15pp。Tornado敏感度排序：FRAG第一（约 +/-9pp），其后为私人信贷/GDP、居民债务/GDP、老年抚养比、CAPE；货币变量（实际利率、央行资产）敏感度最低——约束在真实面。结论的排序与结构对扰动稳健；绝对数值不稳健，故给出区间而非点估计。

Applying +/-20% independent perturbations to all 2026 input parameters, N=10,000 simulations: P5/P50/P95 = 58%/65.8%/73%, 90% interval width ~15pp. Tornado ranking: FRAG first (~+/-9pp), then private credit/GDP, household debt/GDP, old-age dependency, CAPE; monetary variables rank lowest—the constraints are real-side. The ordering and structure of conclusions are robust to perturbation; absolute values are not—hence intervals, not point estimates.

5.5 泡沫总量公式 / 5.5 The Aggregate Bubble Size Formula

该式把强度（BII）翻译为规模，是 §6.7 中"1929 年泡沫总量 36.5% GDP / 2021 年 155.1% GDP"两个数字的来源：2021 年泡沫总量达 155% GDP（≈ $45T，史上最大）却只清算 25%；1929 年仅 36% GDP 却清算 86%。规模不决定深度——深度由清算被允许发生的程度决定，即 §2.4 的冻结系数 φ。

This formula translates intensity (BII) into size, and is the source of the two figures cited in §6.7 (1929 bubble 36.5% of GDP / 2021 bubble 155.1% of GDP): the 2021 bubble totaled 155% of GDP (≈ $45T, the largest ever) yet cleared only 25%; 1929 totaled just 36% of GDP yet cleared 86%. Size does not determine depth—depth is determined by how much clearing is allowed to happen, i.e., the freeze coefficient φ of §2.4.

5.6 清算时长与出清速度 / 5.6 Clearing Duration and Pace

其中 MDS 为货币纪律得分（1929：9 → 2026：4）。五轮出清的月均出清速度均值为 2.5%/月（1929 年 2.5、1973 年 2.3、2000 年 1.6、2008 年 3.4、2022 年 2.8）——出清的物理速度惊人恒定，变化的只是政策允许它走多远。这是冻结系数 φ 的实证根基：被冻结的清算并未消失，而以递延利息的形式累积进入后续周期的 δ(S''')——v3.0 的 59.6% 被上修至 65.8% 的深层结构正在于此：不是模型变悲观，是账本变诚实。

where MDS is the monetary discipline score (1929: 9 → 2026: 4). The mean monthly clearing pace across five cycles is 2.5%/month (1929: 2.5, 1973: 2.3, 2000: 1.6, 2008: 3.4, 2022: 2.8)—the physical pace of clearing is remarkably constant; what varies is how far policy lets it run. This is the empirical basis of the freeze coefficient φ: frozen clearing does not vanish but accumulates as deferred interest into subsequent cycles' δ(S''')—the deep structure behind the revision from v3.0's 59.6% to 65.8%: not a more pessimistic model, but a more honest ledger.

六、实证验证：57事件x29国的交叉检验 / 6. Empirical Validation

6.1 验证设计 / Validation Design

模型的核心风险在于过拟合——7个特征拟合57个事件，样本/特征比为8.1:1，处于可接受区间的下沿。为此，我们设计了五层验证体系：（1）留一国交叉验证（LOCO-CV），逐一剔除每个国家的全部事件，用剩余国家训练、预测被剔除国家；（2）分层交叉验证，按发达/新兴经济体分组分别训练和预测；（3）时间分割验证，用2000年前事件训练、预测2000年后事件；（4）Bootstrap置信区间，1000次有放回重采样构建rho的95%置信区间；（5）伪样本外验证（Pseudo-OOS），用2021年前49个事件训练、预测2021年后8个事件。补充：quant-paper 的六时点相空间轨迹（1929/1973/1987/2000/2007/2021/2026）与微观出清证据（IPO崩塌、产业重合度、就业出清系数等）进一步印证了模型的机制判断。

The model's core risk is overfitting—7 features fitting 57 events yields a sample/feature ratio of 8.1:1, at the lower bound of statistical acceptability. We design a five-layer validation system: (1) Leave-One-Country-Out CV (LOCO-CV); (2) Stratified CV by advanced/emerging economies; (3) Temporal split (pre-2000 training, post-2000 prediction); (4) Bootstrap 1000x resampling for 95% CI; (5) Pseudo-OOS (pre-2021 training, post-2021 prediction). Supplementary: the quant-paper's six-point phase-space trajectory (1929/1973/1987/2000/2007/2021/2026) and micro-clearing evidence (IPO collapse, industrial overlap, employment clearing coefficient) further corroborate the model's mechanism judgments.

6.2 分层交叉验证结果 / Stratified Cross-Validation

发达经济体的预测精度（rho=0.902）显著高于新兴经济体（rho=0.800），主要因为新兴市场的危机类型更异质（货币危机、主权违约占比高），且数据质量参差不齐。但即使在跨组预测中，rho仍维持在0.81以上，表明模型捕捉的机制具有跨国普适性。

Advanced economy prediction accuracy (rho=0.902) significantly exceeds emerging markets (rho=0.800), primarily due to greater crisis type heterogeneity in emerging markets and variable data quality. However, even in cross-group prediction, rho remains above 0.81, demonstrating cross-national generalizability.

6.3 Bootstrap 置信区间 / Bootstrap Confidence Intervals

1000次重采样的结果：均值rho=0.9119，95%置信区间[0.8387, 0.9557]，覆盖率94.7%，预测区间宽度36.2pp。置信区间下界0.839仍远高于3F基准的rho=0.607，说明7F模型的增量解释力在统计上显著。

Bootstrap 1000x resampling yields: mean rho=0.9119, 95% CI [0.8387, 0.9557], coverage 94.7%, prediction interval width 36.2pp. The lower bound 0.839 far exceeds the 3F baseline rho=0.607, confirming the 7F model's incremental explanatory power is statistically significant.

6.4 滚动样本外验证 / Rolling Out-of-Sample Validation

7F模型的样本外rho=0.846，IS->OOS衰减仅0.078（5F为0.123），表明增加CRISIS_SOVEREIGN和QE_TAPER两个特征不仅提升了样本内拟合，也改善了样本外泛化。13个滚动窗口中仅1次系数符号翻转，说明模型结构在不同时间段内保持稳定。

The 7F model achieves OOS rho=0.846 with IS->OOS decay of only 0.078 (vs. 0.123 for 5F), indicating that adding CRISIS_SOVEREIGN and QE_TAPER improves both in-sample fit and out-of-sample generalization. Only 1 coefficient sign flip across 13 rolling windows confirms structural stability.

6.5 伪样本外验证：2021年后事件 / Pseudo-OOS: Post-2021 Events

Pseudo-OOS rho=0.810，MAE=7.83pp，方向正确率8/8（100%）。作为对比，3F基准在同组事件上的rho=0.500，MAE=21.32pp。模型对2021年美国QE泡沫的预测精度极高（误差0.6pp），但对小型发达经济体（加拿大、澳大利亚、挪威、新西兰）的房市修正系统性低估，因为这类事件的CRISIS_LEVEL=0（技术回调），模型对低烈度事件的绝对值预测偏保守。

Pseudo-OOS achieves rho=0.810, MAE=7.83pp, direction accuracy 8/8 (100%). By comparison, the 3F baseline yields rho=0.500 and MAE=21.32pp. The model predicts the 2021 US QE bubble with high precision (0.6pp error), but systematically underestimates housing corrections in small advanced economies, as these events have CRISIS_LEVEL=0 and the model is conservative for low-severity events.

6.6 LOCO-CV 分层诊断 / LOCO-CV Stratified Diagnostics

模型在L2级（严重系统性危机）上方向正确率达100%，但在L0级（技术回调）仅62%——这与预期一致：低烈度事件的噪音/信号比更高。值得注意的短板是global_shock类型的相关系数仅为rho=0.267——尽管MAE不高（6.3pp），但模型无法有效区分全球系统性冲击内部的相对严重度排序。这是模型的已知局限，在第十二章中进一步讨论。

The model achieves 100% direction accuracy for L2 (severe systemic crises), but only 62% for L0 (technical corrections)—consistent with expectations: low-severity events have higher noise/signal ratios. A notable weakness is the global_shock type correlation of only rho=0.267—despite a low MAE (6.3pp), the model cannot effectively rank severity within global systemic shocks. This known limitation is further discussed in Chapter 12.

6.7 实证发现的理论意义 / Theoretical Implications

回到理论框架的核心发现：1929年泡沫总量GDP的36.5%（小泡沫），跌了86%（大灾难）；2021年泡沫总量GDP的155.1%（历史最大），跌了25%（温和调整）。决定回撤深度的关键变量不是泡沫规模（BII），而是冻结系数——即央行和政策工具阻止清算发生的能力。7F模型中QE_DUMMY和QE_TAPER两个特征的正入选，从实证角度验证了这一理论判断。

Returning to the framework's core finding: in 1929, a bubble of 36.5% GDP (small) produced an 86% drawdown (catastrophic); in 2021, a bubble of 155.1% GDP (largest in history) produced only a 25% drawdown (moderate correction). The key determinant of drawdown depth is not bubble size (BII) but the freeze coefficient. The inclusion of QE_DUMMY and QE_TAPER in the 7F model empirically validates this theoretical proposition.

6.8 六时点相空间轨迹 / 6.8 The Six-Point Phase Trajectory

§6.1 前向引用的相空间轨迹在此完整给出：把六轮泡沫顶点的三条慢变量（E 采出进度、TFR、Debt/GDP；E 列 2026-09-04 物理重锚，旧 EROI 口径括号留痕）与其合成的 BII、RCS 并置，得到百年泡沫的相空间轨迹。

The phase trajectory forward-referenced in §6.1 is given here in full: juxtaposing the three slow variables (E depletion progress, TFR, Debt/GDP; the E column physically re-anchored 2026-09-04, old EROI caliber retained in parentheses) at six bubble peaks with their composites BII and RCS yields the century-long phase trajectory.

注 1：E 列 2026-09-04 物理重锚——采出进度 Q/D∞（D∞ = 2,782 Gb，成熟期 Hubbert 线性 r = −0.997），括号内为旧 EROI 叙事口径留痕；发现账剩余年限 1929 = 30.0 年 → 2026 = 29.1 年（两时点几乎相同，但存量完整度 98.9% vs 37.5% 相去甚远——流量视角的剩余年限掩盖了存量耗竭深度）。重锚依据见 §2.2 式(4) 实证口径注。

Note 1: the E column was physically re-anchored on 2026-09-04—depletion progress Q/D∞ (D∞ = 2,782 Gb, mature-period Hubbert linearization r = −0.997), with the old EROI narrative caliber retained in parentheses; discovery-account remaining years 30.0 (1929) → 29.1 (2026)—nearly identical flow-side horizons that hide very different stock-side depletion (98.9% vs. 37.5% of the pool intact). See the Eq. (4) empirical-caliber note in §2.2.

轨迹的关键特征：它不是直线。若三条存量独立衰减，相空间轨迹将是线性投影；BII 中的耦合项使轨迹在 2000 年前后弯曲——能源与人口存量跌破临界值后，BII 的增长被物理掐住，即使 Debt/GDP 继续攀升。2000 年后所有泡沫聚集在"高强度 × 高约束"右上象限：信用与人口两线同时收紧、E 线约束持续加深（但按重锚口径从未临界）的定量图示。2026 年是相空间的最远点（BII 6.74 × RCS 7.95）——但按 2026-09-04 重锚口径，其形态为信用×人口双线共振叠加 E 线深耗竭（62.5% 已采、发现账剩余 29.1 年，深但非临界），而非完整三线共振；1929 则相应降为 E 线安全区的对照组（池 98.9% 完整），见 §13 命题一。

The key feature: the trajectory is not a straight line. With independent decay it would be a linear projection; the coupling term inside BII bends the trajectory around 2000—once energy and demographic stocks fell below critical values, BII growth is physically capped even as Debt/GDP climbs. Post-2000 bubbles cluster in the high-intensity × high-constraint quadrant: the quantitative picture of the credit and demographic lines tightening simultaneously while the E line deepens continuously (though never turning critical under the re-anchored caliber). 2026 is the farthest point in phase space (BII 6.74 × RCS 7.95)—yet under the 2026-09-04 re-anchored caliber its configuration is a credit × demographics dual resonance compounded with deep-but-sub-critical energy depletion (62.5% extracted, 29.1 discovery-account years remaining), not a full triple resonance; 1929 is correspondingly demoted to the E-safe control (98.9% of the pool intact). See Proposition 1 in §13.

6.9 微观出清证据 / 6.9 Micro Evidence of Clearing

§6.1 前向引用的微观出清证据在此完整给出：六项微观指标的五轮规律与 2026 读数。

The micro-clearing evidence forward-referenced in §6.1 is given here in full: five-cycle patterns and 2026 readings for six micro indicators.

微观证据与宏观方程互为印证：集中度、重合度与就业系数共同刻画"冻结时代"的出清形态——深度被压缩（冻结）、频率被抬升（形成期加速）、但名义规模被递延（δ 复利）。其中出清完成度 0.76 是现代路径依赖的核心数字：1987 年后"清算 → 低泡沫"被"清算 → 直接再膨胀"取代，R3 低泡沫复利窗口在缩短。

Micro and macro evidence cross-confirm: concentration, overlap, and employment coefficients jointly characterize clearing in the freeze era—depth compressed (freeze), frequency raised (formation acceleration), nominal scale deferred (δ compounding). The completion ratio of 0.76 is the pivotal number of modern path dependence: after 1987, "clearing → low-bubble" was replaced by "clearing → direct reflation," shortening the low-bubble compounding window.

6.10 五轮复苏数据库 / 6.10 The Five-Cycle Recovery Database

清算之后的资产行为是策略层"退出后持有什么"的全部依据。五轮出清的复苏数据库（股票回本年限 / 黄金区间收益 / 货币购买力）：

Post-clearing asset behavior is the entire basis of the strategy layer's "what to hold after exit." The five-cycle recovery database (equity recovery years / gold range return / currency purchasing power):

复苏数据库的三条规律：(1) 黄金在五轮出清中无一例外为正（+60% ~ +700%），是清算期唯一跨制度有效的对冲资产——这构成三档配置中 wait/full_exit 段黄金 55% 权重的实证地基；(2) 股票回本年限从 25 年压缩到 2 年——与冻结系数 φ 的年代编码（0.05→0.30→0.60→0.80→0.95）一一对应，验证的是"政策允许出清走多远"而非"市场自我修复能力变强"；(3) 1973–74 与 2000–02 两轮货币购买力显著为负——通胀型出清中现金不是安全资产，黄金与短债才是。

Three regularities: (1) gold is positive in all five clearings (+60% ~ +700%), the only cross-institutionally effective hedge—the empirical foundation of the 55% gold weight in wait/full_exit tiers; (2) equity recovery years compress from 25 to 2—mirroring the freeze-coefficient era encoding (0.05→0.30→0.60→0.80→0.95), verifying "how far policy lets clearing run," not "markets heal better"; (3) purchasing power is deeply negative in 1973–74 and 2000–02—in inflationary clearings cash is not the safe asset; gold and short bills are.

七、系统性破产风险指数与周期速率模型 / 7. SBRI and Cycle Rate Model

本章把 SBRI 预警系统的全量内容并入论文：既有理论核心——SBRI 构造（连续概率替代二元触发）、时间条件概率、六催化剂追踪、周期速率模型与杠铃定理；也有工程与实证增量——催化剂接近度的操作化读数（§7.3）、SBRI × 退出信号联动矩阵（§7.6，桥接第八章策略引擎）、模型有效性与失效条件的统一声明（§7.7，含 F1–F6）、以及四次历史泡沫清算的 SBRI 回溯（§7.8）。合在一起，回答"会不会破"（SBRI）、"破多深"（§5 方程）、"何时破、多快破"（周期速率）、以及"模型自己何时不可信"（§7.7）四个问题。

This chapter incorporates the full SBRI early-warning system into the paper: the theoretical core—SBRI construction (continuous probabilities replacing binary triggers), time-conditional probabilities, the six-catalyst tracker, the cycle rate model, and the Barbell Theorem—along with engineering and empirical increments: the operationalized catalyst proximity readings (§7.3), the SBRI × exit-signal interaction matrix (§7.6, bridging the strategy engine of Chapter 8), a unified declaration of model effectiveness and failure conditions (§7.7, including F1–F6), and SBRI retrospectives over four historical bubble liquidations (§7.8). Together they answer "will it break" (SBRI), "how deep" (§5 equation), "when and how fast" (cycle rate), and "when should the model itself not be trusted" (§7.7).

7.1 SBRI构造：从二元触发到连续概率 / SBRI Construction

传统的金融预警系统依赖阈值触发（如"当某指标超过 X 时发出信号"），但泡沫周期中许多指标的临界值本身是路径依赖的。SBRI 采用连续概率构造，把 S'''、kappa 和周期长度三个维度合成为一个 0–100 的连续指数。

Traditional financial early-warning systems rely on threshold triggers, but critical values of many indicators are path-dependent in bubble cycles. SBRI uses a continuous probabilistic construction, synthesizing three dimensions—S''', kappa, and cycle length—into a 0–100 continuous index.

Tier 1 的 κ 维度采用式 (23) 的 γ-GSM 修正口径（预警系统内部称 Tier 1.5）：κ'(G, GSM) = 3.92（含法币信心折价 13.1%），使 δ(S''') = 3.35 × 3.92 = +13.13pp，对应 v3.3 预测回撤 65.8%（见 §5.1）。历史对照：仅 1973（GSM = 1.96）与 2026（GSM = 1.31）两次触发黄金压力修正——同样的存量赤字，在法币信心受质疑的时期将触发更深的市场清算。换言之，SBRI 的 κ 维度不是一个静态不平等读数，而是内嵌了"市场对货币制度本身定价"的实时信号。

The κ dimension of Tier 1 adopts the γ-GSM-corrected value from Eq. (23) (Tier 1.5 within the warning system): κ'(G, GSM) = 3.92 (including a 13.1% fiat-confidence discount), giving δ(S''') = 3.35 × 3.92 = +13.13pp and the v3.3 predicted drawdown of 65.8% (see §5.1). Historical comparison: only 1973 (GSM = 1.96) and 2026 (GSM = 1.31) have triggered the gold stress modification—the same stock deficit triggers deeper market clearing when fiat confidence is questioned. In other words, the κ dimension of SBRI is not a static inequality reading but embeds a real-time signal of how the market prices the monetary regime itself.

7.2 时间条件概率 / Time-Conditional Probability

概率函数的指数衰减形式保证了"近处概率低、远处加速增长"，与金融危机"量变→质变"的非线性特征一致。SBRI=64 处于高风险区下沿——不是"即将破产"，而是"条件已就绪，等待催化剂"。

The exponential decay form ensures "low near-term probability, accelerating long-term growth," consistent with the nonlinear "accumulation → phase transition" characteristic of financial crises. SBRI=64 sits at the lower edge of the high-risk zone—not "imminent bankruptcy" but "conditions ripe, awaiting catalyst."

7.3 六催化剂追踪 / Six-Catalyst Tracker

六催化剂追踪不是预测具体引爆器（见 §12 局限），而是对条件概率乘以加速因子：任一催化剂触发即上调破产概率。2026-08 的 P1（估值-信贷背离）= 85，是六项中最高的一项。

The six-catalyst tracker does not predict the specific trigger (see §12 limitations) but multiplies the conditional probability by acceleration factors: any catalyst triggered raises the bankruptcy probability. As of 2026-08, P1 (valuation-credit divergence) = 85, the highest of the six.

式 (31) 的一般乘积形式在操作层按"接近度"连续计量（0–100）：每个催化剂独立更新，接近度越高，距离触发越近。操作化的阈值形式与 2026-08 六项读数如下：

The general product form of Eq. (31) is operationalized through continuous "proximity" scores (0–100): each catalyst updates independently, and higher proximity means closer to trigger. The operational threshold form and the 2026-08 readings of all six:

当前加速效应微弱（A = 1.038）：P1 独自越过 70 的贡献仅 3.8%。但加速因子是非线性的——若 P1 升至 95 且 P2 亦突破 70，A 将升至 1.15 以上，24 个月破产概率从 32.4% 跃升至约 37%。这正是"催化剂不改变结构性风险（SBRI），只改变风险的到达时间分布"的数学表达：Tier 1 决定概率的 ceiling，Tier 3 决定曲线向上弯曲的速度。

The current acceleration is mild (A = 1.038): P1 alone above 70 contributes only 3.8%. But the factor is nonlinear—if P1 rises to 95 and P2 also breaks 70, A exceeds 1.15 and the 24-month bankruptcy probability jumps from 32.4% to roughly 37%. This is the mathematical expression of "catalysts do not change structural risk (SBRI), only the arrival-time distribution of risk": Tier 1 sets the probability ceiling; Tier 3 sets how fast the curve bends upward.

7.4 周期形成期函数与未来六轮预测 / Formation Period Function and Six-Cycle Forecast

周期速率模型从 RCS 的物理约束中推导泡沫形成期的加速定律：RCS 越高，真实约束越紧，形成期越短。这一函数的物理下限为 36 个月——对应"三线共振"状态下系统能维持非平衡态的最短时间。

The cycle rate model derives the acceleration law of bubble formation from RCS physical constraints: higher RCS means tighter real constraints and shorter formation periods. The physical floor is 36 months—the minimum time the system can sustain a non-equilibrium state under "triple resonance."

7.5 杠铃定理 / The Barbell Theorem

杠铃定理：未来周期的相对回撤幅度将收敛（30–45%），但全球经济总量的持续增长使绝对冲击量（美元计）指数化增长。第 6 轮（2049–50）的绝对冲击量约为第 1 轮的 2.7 倍。对资产管理者的含义：相对回撤可控，但绝对损失规模将越来越大——"防守"的回报率在上升。

The Barbell Theorem: future relative drawdowns will converge (30–45%), but continued global GDP growth makes absolute impact (in dollars) grow exponentially. Cycle 6's (2049–50) absolute impact is approximately 2.7× cycle 1's. Implication for asset managers: relative drawdowns are controllable, but absolute loss scale keeps growing—the return on "defense" is rising.

7.6 SBRI × 退出信号联动矩阵 / The SBRI × Exit-Signal Interaction Matrix

SBRI 回答"结构性风险有多高"，第八章的五投票退出信号回答"操作触发到了没有"。两者正交：SBRI 高而退出信号未触发，意味着条件已就绪但市场尚未进入恐慌性出清。联动矩阵把两个维度组成 4 × 3 决策表，是第七章（预警）与第八章（引擎）之间的桥——预警层不直接下单，但为引擎层的档位切换提供先行约束：

SBRI answers "how high is structural risk"; the five-vote exit signals of Chapter 8 answer "has the operational trigger fired." The two are orthogonal: high SBRI with no exit signal means conditions are ripe but the market has not yet entered panic clearing. The interaction matrix combines both dimensions into a 4 × 3 decision table—the bridge between Chapter 7 (warning) and Chapter 8 (engine): the warning layer does not place orders directly, but supplies leading constraints for the engine's tier switching:

2026-08 定位：SBRI = 64 ∈ [60, 80]，退出信号 0/5——落点在"预演退出"格：完成退出路径设计、止损位设定与流动性预检，但暂不执行。这与 §8.4 状态机当前"清算等待（wait）"档的判定一致——两套系统独立读数、相互印证，这是多层架构的价值所在：任何单一系统的误读都会在交叉核对中暴露。

The 2026-08 position: SBRI = 64 ∈ [60, 80] with 0/5 exit signals—the cell is "rehearse exit": complete exit-path design, stop-loss setting, and liquidity pre-checks, but do not execute yet. This matches the §8.4 state machine's current "wait" tier—two systems reading independently and confirming each other, which is precisely the value of layered architecture: any single system's misreading is exposed in cross-checking.

7.7 模型有效性、适用边界与失效条件 / Model Effectiveness, Applicability Zones, and Failure Conditions

预警系统必须自带"何时自己不可信"的声明——这是概率化架构相对二元触发的另一优势：有效性本身也可以被连续计量。模型综合有效性按六个维度量化自评，2026-08 读数为 72（满分 100，"基本生效"区）：

A warning system must carry its own declaration of when it should not be trusted—another advantage of the probabilistic architecture over binary triggers: effectiveness itself can be measured continuously. Model effectiveness is self-assessed on six dimensions; the 2026-08 composite score is 72 (out of 100, the "basically effective" zone):

适用边界按 BII 划分三区。生效区（BII 3.0–6.5，误差带 ±9%）：泡沫具有"估值膨胀 → 信贷扩张 → 货币紧缩 → 杠杆清算"完整四段结构的传统资产泡沫，S 分量主导，SBRI 通常 < 30、概率曲线平缓。过渡区（BII < 3.0 或 > 6.5，误差带 ±15–25%）：2026 美国 BII = 6.74 已突破上边界，轻资产（AI）泡沫的估值特征可能使 κ 低估实际破坏力。失效区：下列 F1–F6 任一触发时，模型须整体重估而非参数微调：

Applicability is zoned by BII. The effective zone (BII 3.0–6.5, error band ±9%): traditional asset bubbles with the full four-stage structure (valuation inflation → credit expansion → monetary tightening → leverage liquidation), S-dominant, with SBRI typically < 30 and a flat probability curve. The transitional zone (BII < 3.0 or > 6.5, error band ±15–25%): the 2026 US BII = 6.74 has breached the upper bound, and the valuation profile of light-asset (AI) bubbles may cause κ to underestimate actual damage. The failure zone: whenever any of F1–F6 below fires, the model must be re-estimated wholesale rather than fine-tuned:

F1–F6 与第十二章的局限性声明互补而不重叠：局限性 1–12 刻画"模型在哪些方面不精确"（误差带问题），F1–F6 刻画"模型在哪些条件下整体失效"（范式适用性问题）。两类边界不可混淆——前者可以通过加宽置信区间吸收，后者意味着全部历史回测同时失效，唯一护栏只剩下仓位结构本身（见 §8.4 的权益 ≤ 85% 约束与风险死线阀门）。

F1–F6 complements rather than overlaps the limitations declared in Chapter 12: Limitations 1–12 describe where the model is imprecise (error-band issues), while F1–F6 describe under what conditions it fails wholesale (paradigm-applicability issues). The two classes must not be conflated—the former can be absorbed by widening confidence intervals; the latter means all historical backtests fail simultaneously, leaving position structure itself (the equity ≤ 85% constraint of §8.4 and the risk deadline valve) as the only guardrail.

7.8 历史验证：SBRI 回溯 / Historical Validation: SBRI Retrospectives

用 SBRI 框架回溯四次历史泡沫清算，检验其区分度与时间条件概率的校准效果：

Four historical bubble liquidations are retraced under the SBRI framework to examine its discriminating power and the calibration of its time-conditional probabilities:

四案例给出三条校准结论。其一，深度排序正确：SBRI 对 1929（82）与 2008（55）的排序与实际回撤（86% vs 57%）一致；1929 在崩盘前约 6 个月即进入临界区，S = −4.2 为历史最深，κ ≈ 4.0 因 Gini > 0.55 而显著放大；2008 年 S 深度适中（−2.1）、κ 较低（2.2），触发后实际崩盘发生在 12–15 个月内，与模型时间窗口一致。其二，范式泡沫是系统性盲区：2000 年 SBRI 仅 48——互联网泡沫并非跨期存量支取型，S 深度有限（−1.5），但板块集中度使 Nasdaq 实际回撤（78%）远超 SBRI 档位所暗示的水平；对集中度极高的范式泡沫，需附加板块集中度乘数修正。其三，出清时长可能被低估：日本 1990 年 SBRI = 70 正确识别了高风险，但时间条件概率低估了清算时长——制度性拖延（央行容忍僵尸贷款、MDS 异常低）使实际出清远超模型预测，这是框架需要补充"制度性拖延系数"的教训。

Three calibration lessons emerge from the four episodes. First, depth ordering is correct: SBRI's ranking of 1929 (82) versus 2008 (55) matches the actual drawdowns (86% vs 57%); 1929 entered the critical zone roughly six months before the crash, with the deepest S on record (−4.2) and κ ≈ 4.0 amplified by Gini > 0.55; in 2008, moderate S depth (−2.1) and low κ (2.2) were followed by an actual crash within 12–15 months of trigger, consistent with the model's time window. Second, paradigm bubbles are a systematic blind spot: in 2000 SBRI read only 48—the dot-com bubble was not a stock-drawdown type and S was shallow (−1.5), yet sector concentration drove Nasdaq's actual drawdown (78%) far beyond what the SBRI tier implied; highly concentrated paradigm bubbles require an additional sector-concentration multiplier. Third, clearing duration can be underestimated: Japan 1990's SBRI = 70 correctly identified high risk, but the time-conditional probability underestimated clearing duration—institutional deferral (central-bank tolerance of zombie lending, abnormally low MDS) stretched actual clearing far beyond the model's prediction, a lesson suggesting the framework needs an "institutional deferral coefficient."

八、策略引擎与状态机 / 8. The Strategy Engine and State Machines

本章把第七章的预警信号转化为可执行的交易规则。策略引擎由四层组成：票源铁律（§8.1，详见 §3.4）、Shiller 回测五票（§8.2）、anchor-vote1-κ0.25 三件套状态机与三档配置（§8.3，v2.2 三档状态机为其沿革、七国沿用）、退出进入模型（§8.4）、E10R5 回撤熔断（§8.5）。核心纪律：退出信号前置——投票读数升至退出阈值即执行，不等价格确认。

This chapter translates Chapter 7's warning signals into executable trading rules. The engine has four layers: the vote-source iron law (§8.1, see §3.4 for details), Shiller backtest votes (§8.2), the anchor-vote1-κ0.25 three-piece state machine and allocations (§8.3; the v2.2 three-tier machine is its lineage and remains in use for the seven-country suite), the exit-entry model (§8.4), and the E10R5 drawdown breaker (§8.5). Core discipline: signals lead prices—execute at the vote threshold without waiting for price confirmation.

8.1 票源铁律（回顾 §3.4）/ 8.1 The Vote-Source Iron Law (Cross-Reference §3.4)

第三章 §3.4 已完整定义票源铁律：四种票源语义（实盘真宏观 ① / 回测 tech5 ② / 实验 macro_proxy ③ / Shiller 真宏观 ④）在任何实证中禁混用。本章全部实证均使用票源 ④（Shiller 真宏观）驱动状态机；票源 ①（实盘真宏观）的规则化五票规格见 §3.2，但因无月度历史序列不可回测。以下各节均指明票源编号。

Chapter 3 §3.4 defined the vote-source iron law in full: four non-interchangeable source semantics (live true-macro ① / backtest tech5 ② / experimental macro_proxy ③ / Shiller true-macro ④). All empirics in this chapter use source ④ (Shiller true-macro) to drive the state machine; source ①'s (live true-macro) rule-based five-vote specification is in §3.2, but it has no monthly historical series and cannot be backtested. Each section below labels its source number.

五投票制（macrovote）以五张独立投票（估值、信贷、能源、人口、技术结构）决定档位切换，构成 8 国模拟盘的引擎；其核心纪律是：投票读数升至退出阈值即执行，不等价格确认——退出信号前置。三阶段（R1 形成期 / R2 清算期 / R3 低泡沫期）的资产 × 阶段年化收益矩阵给出仓位的地基：

The five-vote rulebook (macrovote) drives position-tier switching through five independent votes (valuation, credit, energy, demographics, technical structure) and powers the eight-country simulation book; its core discipline is to execute at the vote threshold without waiting for price confirmation—signals lead prices. The asset × regime matrix for the three stages (R1 formation / R2 clearing / R3 low-bubble) provides the foundation of positioning:

8.2 Shiller 回测五票（票源 ④）/ 8.2 The Shiller Backtest Votes (Source ④)

回测票源来自 Shiller 月度数据集（1881-01 ~ 2026-07，145 年），含 CPI / 长期利率 / RealPrice / PE10 四个核心字段——1996–2026 整 30 年窗口每一个月都有完整五票读数，无窗口外推：

The backtest source draws on the Shiller monthly dataset (1881-01 ~ 2026-07, 145 years) with CPI / long rate / RealPrice / PE10—every month of the 1996–2026 window carries complete vote readings, no extrapolation:

七国无 PE10 月度长史，估值票 m1 改用实际价格分位代理——较 PE10 分位更敏感（不含盈利平滑层）。七国数据集采用 FRED/OECD MEI 国别 CPI + 10y 利率 + 指数/CPI 同比链乘，与 US 同窗跨市场实证（US 366 月、七国 365 月）。m1 的 CAPE 滚动窗口分位判定存在回看校准成分，按 freeze-wait-z20 报告核算前视溢价约 +0.31pp/年（见 §12 局限）。

Seven countries lack PE10 monthly history, so m1 uses the real-price percentile proxy—more sensitive than the PE10 percentile (no earnings-smoothing layer). Their dataset chains FRED/OECD MEI CPI, 10-year rates, and index/CPI, same-window cross-market evidence as the US (366 vs 365 months). The rolling-window m1 percentile embeds a look-back calibration component worth ~+0.31pp/yr of look-ahead premium (see §12).

8.2b ACFR 双口径追加票（ADDVOTE，票源 ④ 扩展，2026-09-03 立项 · 2026-09-04 AN-04 裁定退役）/ 8.2b The ACFR Dual-Caliber Append Votes (ADDVOTE, Source ④ Extension · Retired 2026-09-04)

退役披露（2026-09-04）：anchor-vote1-κ0.25 框架九臂对照（anchor_addon_decomp.json，锚校验 PASS）测得 ADDVOTE 组合级贡献 Lag-1 −0.97pp / SAME −0.67pp 双口径负贡献，MDD 零改善（−22.5% 持平）；v2.2 框架的 +0.92pp 增益系给弱 Lag-1 基线（9.42%/MDD −52.6%）补防御，不可迁移至 anchor 强防御基线。裁定：触发通道退役（wait 降级回归 any1 单条件），ACFR 读数写回保留为观测哨。本节保留为规格历史记录。

规格（2026-09-03 立项，规格五环 M-19）：A 轴分子 rel12 = NDX/SPX 12 月相对收益；B 轴分母 g12 = 实际股息（div）/实际盈利（eps）TTM 12 月增长率，下限截断 −50% 防分母崩塌产生的机械负值。div 链 = Shiller 年化 TTM 实际股息（≤2023-06）接 multpl TTM 股息率 × SP500/CPI（calib=1.0008，五项预注册验证全 PASS）；eps 链 = Shiller Real Earnings + multpl 年度锚点（2022-12: 194.38 / 2023-12: 209.50 / 2024-12: 222.39 / 2025-12: 247.98 / 2026-03: 264.69）。采纳形态 ADDVOTE：仅在 hold 段给 any1 wait 触发追加两个 OR 触发源——不动五票票板、不动 E10/rr5 通道、不动三档配置。经济含义：分子度量「AI/成长相对大盘的兑现亢奋」（价格跑赢大盘），分母度量「基本面兑现能力」（实际股息/盈利增速）——分子涨而分母不涨，即 AI 资本兑现率（AI Capital Fulfillment Ratio）恶化，亢奋脱离兑现，支持降档防御。

Specification (initiated 2026-09-03, five-ring M-19): numerator rel12 = NDX/SPX 12-month relative return; denominator g12 = real dividend (div) / real earnings (eps) TTM 12-month growth, floored at −50% to block mechanically negative denominators. The div chain splices Shiller annualized-TTM real dividends (≤2023-06) with multpl TTM dividend yield × SP500/CPI (calib=1.0008, five pre-registered validations passed); the eps chain uses Shiller Real Earnings plus multpl annual anchors (2022-12: 194.38 / 2023-12: 209.50 / 2024-12: 222.39 / 2025-12: 247.98 / 2026-03: 264.69). The adopted form ADDVOTE appends two OR-trigger sources to the any1 wait trigger in the hold tier only—the five-vote ballot, the E10/rr5 channels, and the three-tier allocations are untouched. Economic reading: the numerator meters AI/growth euphoria versus the broad market (price out-running the index), the denominator meters the fundamental fulfillment capacity (real dividend/earnings growth)—a rising numerator on a stagnant denominator means the AI Capital Fulfillment Ratio is deteriorating: euphoria decoupled from fulfillment, supporting a defensive demotion.

证据与择形（三形态全测 × 双口径 × θ∈{1.15,1.20,1.25} × 六泡沫 LOO）：同月口径（生产）基线 22.68%/−21.7%/Sh 1.650 → ADDVOTE 双票 θ=1.20 达 23.60%/−21.7%/Sh 1.726（卡玛 1.089，MDD 持平——ACFR 只追加 wait 触发不触碰 E10，全形态 MDD 均持平）；Lag-1 口径（研究）基线 9.42%/−52.6%/0.652 → ADDVOTE 9.82%/−44.9%/0.745，其中 eps 票为主力载体（单票 11.54%/−38.4%/0.838），div 票单独追加无增益（8.65%/−52.5%）。LOO 六泡沫同月口径 4正/1负/1零（E1 +2.07 / E2 +4.23 / E6 +0.97 / 唯负 E4 COVID −3.67），θ* 收敛 1.20。否决其余形态：TIER3（any1 AND acfr→full_exit 升档）增益≈0；ONLY 独立臂无优势且 LOO 三案例全负；盈利崩塌守护版（epsg12>−10% 才计 eps 票）两口径均净负——守护条件与增益事件负相关（2008/2020 分母崩塌月恰是分母信号有效月，守护把它们全部屏蔽，Lag-1 下 E1 由 +10.26 毁至 −1.34）。干净过渡：最新信号月 2026-06 读数 ACFR_div = 1.007 / ACFR_eps = 0.934，双票均低于 θ → 上线即无信号翻转。已知代价如实披露：换手 187 → 219（+32 单位，0.15% 单边成本已含）；增益集中于 E1/E2 两案例（样本内拟合风险，前向 12 个月双轨对照对冲）；七国无股息/盈利月度长链，本票为 US 盘专用。过程披露：票贡献分解臂初版实现误为「仅单票」、修复为「any1 OR 单票」后重跑，主形态结果逐位不变。

Evidence and form selection (three forms × dual calibers × θ∈{1.15,1.20,1.25} × six-bubble LOO): same-month caliber (production) baseline 22.68%/−21.7%/Sh 1.650 → ADDVOTE dual votes at θ=1.20 reach 23.60%/−21.7%/Sh 1.726 (Calmar 1.089, MDD unchanged—ACFR only appends wait triggers and never touches E10, so MDD is flat across all forms); Lag-1 caliber (research) baseline 9.42%/−52.6%/0.652 → ADDVOTE 9.82%/−44.9%/0.745, with the eps vote as the dominant carrier (single-vote 11.54%/−38.4%/0.838) while the div vote alone adds nothing (8.65%/−52.5%). Same-month LOO across six bubbles: 4 positive/1 negative/1 zero (E1 +2.07 / E2 +4.23 / E6 +0.97 / sole negative E4 COVID −3.67), θ* converging on 1.20. Rejected forms: TIER3 (any1 AND acfr→full_exit escalation) gains ≈0; the ONLY standalone arm shows no edge with three negative LOO episodes; the earnings-collapse guard (eps vote counted only if epsg12>−10%) is net-negative in both calibers—the guard condition is negatively correlated with gain events (the 2008/2020 denominator-collapse months are precisely when the denominator signal works; the guard blocks them all, wrecking Lag-1 E1 from +10.26 to −1.34). Clean transition: the latest signal month 2026-06 reads ACFR_div = 1.007 / ACFR_eps = 0.934, both below θ—deployment entails zero signal flips. Costs disclosed: turnover 187 → 219 (+32 units, 0.15% per side already included); gains concentrate in E1/E2 (in-sample fitting risk, hedged by a 12-month forward dual-track); the seven countries lack dividend/earnings monthly chains, so this vote is US-only. Process disclosure: the vote-decomposition arms were initially mis-implemented as "single vote only," re-run after repair to "any1 OR single vote," with all main-form results bit-identical.

8.2b.1 双口径数据链构造与单位考据 / 8.2b.1 Data-Chain Construction and Unit Forensics

双链数据构造（可复现口径，全部对账 outputs/acfr_div_chain.json 与回测脚本逐行同源）：

单位考据（M-15 发现，影响所有股息链使用者）：shiller.csv 的 Dividend / Real Dividend 列为年化 TTM 口径而非月度值——2023-06 隐含股息率 1.58% 与 multpl 1.58% 精确一致即为证据。旧口径「12 月求和」等价于恒定 12× 缩放，增速 g12 不受影响，历史 ACFR_div 读数全部有效；补链段按「年化 TTM → TTM=Σ12 月」严格同构构造，避免方法论接缝。

Unit forensics (M-15 finding, relevant to all dividend-chain users): the Dividend / Real Dividend columns of shiller.csv are annualized-TTM caliber, not monthly values—proven by the 2023-06 implied yield of 1.58% matching multpl's 1.58% exactly. The legacy "12-month summation" equals a constant 12× scaling, leaving the growth rate g12 unaffected—all historical ACFR_div readings remain valid; the spliced segment replicates the "annualized TTM → TTM = Σ12 months" construction to avoid a methodological seam.

8.2b.2 全形态 × 双口径择形矩阵 / 8.2b.2 Form-Selection Matrix (All Forms × Dual Calibers)

择形矩阵（θ=1.20 主锚；年化% / MDD% / Sharpe / 换手；含成本 0.15% 单边）：

θ 敏感性（ADDVOTE 同月口径）：θ=1.15 → 23.21%/Sh 1.701 · θ=1.20 → 23.60%/Sh 1.726 · θ=1.25 → 23.11%/Sh 1.663——θ* 收敛于 1.20，且三档全部 ≥ 基线（阈值不敏感区内的单调凸起，非刀锋最优）。全形态 MDD 均持平 −21.7%：ACFR 只追加 wait 触发、不触碰 E10 兜底通道，回撤结构由 E10 决定，与票源无关——这是「最小侵入」设计的结构性保证。

θ sensitivity (ADDVOTE, same-month): θ=1.15 → 23.21%/Sh 1.701 · θ=1.20 → 23.60%/Sh 1.726 · θ=1.25 → 23.11%/Sh 1.663—θ* converges on 1.20, with all three tiers ≥ baseline (a monotone hump inside an insensitive band, not a knife-edge optimum). MDD is flat at −21.7% across all forms: ACFR only appends wait triggers and never touches the E10 backstop, so the drawdown structure is set by E10 and is vote-source independent—a structural guarantee of the minimal-intrusion design.

8.2b.3 六泡沫 LOO 留一交叉验证全表 / 8.2b.3 Six-Bubble Leave-One-Out Cross-Validation

LOO 协议：对每个泡沫集 E_k，θ* 仅在其余五集上校准（目标 = 预注册的集内得分），再在留出集上评测——θ 无接触留出集信息。d_score = 留出集（形态 − 基线）集内得分差（年化 pp + MDD 改善 pp 之和）。

读表要点：① 同月口径六集 4 正 / 1 负 / 1 零，净 +3.60pp——正贡献集中于 E1/E2（分子亢奋 + 分母失速的典型双杀役），如实在上文披露样本内拟合风险；② E4 COVID 为唯一负集（−3.67pp）：疫情崩塌月分母 epsg12 同步崩塌、ACFR 被地板截断后失真，属票设计已知的分母失效场景；③ 守护版「epsg12>−10% 才计 eps 票」本意是屏蔽 E4，结果 E4 只从 −13.97 收窄到 −6.61，却把 E1 从 +10.26 毁到 −1.34、E2 归零——守护条件与增益事件负相关（2008/2020 分母崩塌月恰是分母信号有效月），这是守护版被否决的核心证据；④ E3/E5 零贡献：两役 wait 已由 any1 触发，ACFR 无增量信息（冗余而非有害）。

Reading notes: ① same-month caliber scores 4 positive/1 negative/1 zero, net +3.60pp—gains concentrate in E1/E2 (classic dual-kill episodes of numerator euphoria plus denominator stall), with in-sample fitting risk disclosed above; ② E4 COVID is the sole negative episode (−3.67pp): during the pandemic crash the denominator epsg12 collapsed in sync, and ACFR was distorted by the floor—a known denominator-failure scenario of the vote design; ③ the guard form ("count the eps vote only if epsg12>−10%"), intended to shield E4, only narrowed E4 from −13.97 to −6.61 while wrecking E1 from +10.26 to −1.34 and zeroing E2—the guard condition is negatively correlated with gain events (the 2008/2020 denominator-collapse months are precisely when the denominator signal works), the core evidence for its rejection; ④ E3/E5 contribute zero: any1 had already triggered wait in both episodes, so ACFR adds no incremental information (redundant, not harmful).

8.2b.4 触发统计与生产接入披露 / 8.2b.4 Trigger Statistics and Production-Integration Disclosure

触发统计（同月口径，θ=1.20，ADDVOTE vs 基线）：

生产接入披露（2026-09-03 上线）：ADDVOTE 闸已接入 US 真宏观盘（sim-us-macro）生产管道——仅在 hold 段给 wait 降级追加 OR 触发源；闸优先级 E10（full_exit）> A1/A2 预警闸 > ADDVOTE > any1；wait 段内退出路径唯一为 rr5（闸不干预恢复）；rr5 恢复当日不干预（镜像回测 rr5 优先语义）。信号链月度刷新：Shiller 更新 → 股息链构建（_build_acfr_div_chain.py）→ 信号构建（_build_acfr_addvote_signal.py，与回测逐行同源）→ outputs/acfr_addvote_signal.json → 每日管道自动读取（读取失败 = 闸不触发，保守降级）。上线当日信号 2026-06 读数 ACFR_div=1.0069 / ACFR_eps=0.9336，双票均低于 θ=1.20，trig=0——干净接入，零信号翻转；12 项单元测试全 PASS（触发/不触发/非 hold 段/非目标盘/rr5 恢复/幂等/旧行为回归）。θ 变更须重走规格五环。

Production integration (deployed 2026-09-03): the ADDVOTE gate is live in the US true-macro sim pipeline—appending OR-trigger sources to the hold-tier wait demotion only; gate priority E10 (full_exit) > A1/A2 warning gates > ADDVOTE > any1; within wait the sole exit remains rr5 (the gate never interferes with recovery), and on an rr5-recovery day the gate stands down (mirroring the backtest's rr5-first semantics). Monthly signal chain: Shiller update → dividend-chain build → signal build (line-by-line identical to the backtest) → outputs/acfr_addvote_signal.json → auto-read by the daily pipeline (read failure = gate inactive, conservative degradation). On deployment day the 2026-06 signal read ACFR_div=1.0069 / ACFR_eps=0.9336, both below θ=1.20, trig=0—a clean integration with zero signal flips; 12 unit tests all pass (trigger/no-trigger/non-hold tier/non-target page/rr5 recovery/idempotency/legacy regression). Any θ change must re-enter the five-ring process.

8.3 实盘真宏观五票规格（票源 ①，规则化定版）/ 8.3 The Live True-Macro Five-Vote Specification (Source ①, Rule-Based)

实盘真宏观五票的完整规格已在 §3.2 给出。此处重申关键纪律：自 2026-08-27 定版起，污染度 P 改为纯规则化计算（§3.3 式 8），不再依赖 AI 主观研判。五票为：① 污染度 P（规则化公式，P = 0.35×I_level + 0.30×I_trend + 0.35×I_money）；② IPO 温度计（12 月滚动 z-score）；③ BII 连续两季（BII >= 阈值 AND 两季）；④ Top10 集中度（前十大权重分位）；⑤ 居民债务-GDP（杠杆加速度）。

The full specification of the live true-macro five votes is given in §3.2. Key discipline restated: from the 2026-08-27 specification onward, pollution P is replaced by a purely rule-based calculation (§3.3 eq 8), no longer relying on AI subjective judgment. The five votes: ① pollution P (rule-based formula); ② IPO thermometer (12-month rolling z-score); ③ two-quarter BII (BII >= threshold AND two quarters); ④ Top10 concentration (top-10 weight percentile); ⑤ household debt/GDP (leverage acceleration).

诚实披露：规则化 P 的首次回算尚未执行，因此实盘当前无法给出票源 ① 的完整票值读数。在首次回算完成前，实盘状态机仍由票源 ④（Shiller 真宏观）驱动，票源 ① 的规则化票值作为参考并行记录。票源 ① 无月度历史序列，因此不可回测——这是票源铁律的核心约束。

Honest disclosure: the rule-based P's first recalculation has not yet been executed, so live source ① cannot currently produce a complete vote reading. Until the first recalculation, the live state machine remains driven by source ④ (Shiller true-macro), with source ①'s rule-based votes recorded in parallel as reference. Source ① has no monthly historical series and therefore cannot be backtested—this is the core constraint of the vote-source iron law.

8.4 anchor-vote1-κ0.25 三件套状态机与三档配置（现行） / 8.4 The anchor-vote1-κ0.25 Three-Piece State Machine and Allocations (Current)

自 2026-09-03 起，US 生产盘（sim-us-macro）由 anchor-vote1-κ0.25 三件套状态机驱动（E15 + RR5′ + κ0.25，月末 tick 判定，票源 MV2 金融五票严格 Lag-1）；v2.2 三档状态机（any1/E10/rr5）为其沿革，七国 sim 仍沿用 v2.2 同构。四条通道：

From 2026-09-03 the US production sim (sim-us-macro) runs the anchor-vote1-κ0.25 three-piece machine (E15 + RR5′ + κ0.25, end-of-month tick, driven by MV2 financial five votes under strict Lag-1); the v2.2 three-tier machine (any1/E10/rr5) is its legacy, retained by the seven-country sims. Four channels:

三档配置沿袭 v2.2 的设计逻辑：① hold 段指数晋升 NDX（最高 β 资产）；② wait / full_exit 段指数降级 SPX → DJI（成熟期防御资产）；③ 黄金阶梯 15 → 55 → 55（五轮出清黄金全正的实证落地）；④ 防御 + 能源各 15% 双缓冲。相对 v2.2 的三处关键升级：① 判定频率从日频 tick 改为月末 tick（每月首个交易日封上月月末价，杜绝日频噪声）；② rr5 的段内低点反弹语义升级为 RR5′ 的锚价水平回归（恢复判据不依赖段内路径，仅依赖净值水平 vs 锚价，消除低点反弹的路径敏感性）；③ κ0.25 锚衰减给深度回撤后的恢复留出时间衰减通道。E15 优先于 any1 保留：净值距 12 月滚动峰 ≤ −15% 时即使宏观票为 0 也立即跳 full_exit——防御最末线的客观价格锚定。诚实披露：回测 KPI（§9.4）未含 L2-k20 黄金腿调制与 ACFR ADDVOTE 生产叠加件；生产端 any1 判定已并入 ACFR ADDVOTE（ACFR_div/eps > 1.20 追加 OR 触发源）。七国注意：防御 / 能源 / 黄金采用美国代理数据，权益袖 = 该国单一指数，三指数降级 β 未建模——回测结论成立，实盘部署需补充国别 def/ene ETF。

The three-tier allocations inherit v2.2's design: (1) hold promotes to NDX (highest-β); (2) wait/full_exit demote SPX → DJI (mature defensive); (3) gold ladder 15 → 55 → 55 implements the finding that gold is positive in all five clearings; (4) defense + energy at 15% each double-buffer. Three key upgrades over v2.2: (1) daily ticks become end-of-month ticks (first trading day seals the prior month-end price, eliminating daily noise); (2) rr5's intra-segment rebound semantics upgrade to RR5′'s anchor-level re-entry (recovery depends only on the net-value level vs the anchor, not the intra-segment path); (3) the κ0.25 anchor decay opens a time-decay channel for recovery after deep drawdowns. E15-over-any1 is retained: a ≤ −15% fall from the 12-month rolling peak jumps straight to full_exit even at zero macro votes. Honest disclosure: the backtest KPIs (§9.4) exclude the L2-k20 gold-leg modulation and the ACFR ADDVOTE production overlay; in production, any1 additionally includes the ACFR ADDVOTE OR-trigger (ACFR_div/eps > 1.20). Seven-country caveats: defense/energy/gold use US proxies; the equity sleeve is the single country index with β demotion unmodeled—valid for backtest conclusions, live deployment requires local def/ene ETFs.

8.5 退出进入模型：五票退出投票与复苏重入 / 8.5 Exit-Entry: The Five-Vote Exit and Re-Entry

退出进入模型是所有模型的顶层决策框架——追踪模型管仓位，退出进入模型管时机。退出投票与宏观五票（§3.2、§8.3）是两套不同票板：退出投票面向事件驱动，五项及其触发条件：

The exit-entry model is the top-level decision framework—tracking manages positions, exit-entry manages timing. The exit votes are a separate ballot from the macro five votes (§3.2, §8.3), event-driven with five items:

决策矩阵：0–1.5 票 = 持有期（全仓位，追踪模型管动量仓）；2.5–3 票 = 退出决策区（启动退出计时、动量仓缓减、现金升至 20%、进入观察名单种子）；≥ 3.5 票 = 全档退出（动量降至 10%、现金 + 黄金 ≥ 60%，仅保留幸存者与对冲仓位）。退出后的持仓构成（清算等待期）：

Decision matrix: 0–1.5 votes = hold (full position, tracking manages momentum); 2.5–3 votes = exit-decision zone (start the exit clock, taper momentum, raise cash to 20%, enter watch-list seeds); ≥ 3.5 votes = full exit (momentum to 10%, cash + gold ≥ 60%, survivors and hedges only). Post-exit holdings (clearing-wait period):

重新进入需四项复苏信号 ≥ 2：① HY OAS 回落（清算期 ＞ 550bp 后回落至 ＜ 400bp）；② 失业率峰值确认（Sahm 触发后回落）；③ 复苏速度方程（回本预测 ＜ 3 年提前布局，＞ 5 年仅保留幸存者）；④ 形成期重置（下一泡沫形成期启动）。2026-08 四项全部未触发——仍在退出计时阶段。

Re-entry requires ≥ 2 of four recovery signals: (1) HY OAS normalization (from > 550bp in clearing back below 400bp); (2) unemployment-peak confirmation (Sahm triggered then receding); (3) the recovery-speed equation (predicted recovery < 3 years → early positioning; > 5 years → survivors only); (4) formation-period reset. As of 2026-08, all four are untriggered—still in the exit-clock phase.

8.6 E10R5 回撤熔断状态机 / 8.6 The E10R5 Drawdown-Breaker State Machine

熔断层与投票层（基本面）互补：投票负责"提前减仓"，熔断负责"价格确认兜底"。状态机规格：

The breaker layer complements the vote layer (fundamentals): votes "de-risk early," the breaker "floors on price confirmation." Machine specification:

右侧确认（反弹 5% 恢复）是其关键设计：若改用"回撤收窄即恢复"的左侧规则，2009 年 V 型反弹等恢复期的踏空损失会明显放大——三资产口径 32 年逐年拆解显示全部年份无负贡献（2008 +10.05pp、2002 +9.03pp、2022 +7.75pp、2001 +7.72pp 为最大增益年），而恢复期年份（2003 / 2009）熔断贡献仍为正（+1.36pp / +3.65pp）。

Right-side confirmation is the key design: a left-side "drawdown-narrowing" recovery rule would sharply amplify opportunity cost in V-shaped recoveries like 2009. A 32-year year-by-year decomposition shows no negative-contribution year, with recovery years (2003 / 2009) still positive (+1.36pp / +3.65pp).

九、策略回测全集 / 9. The Backtest Suite

9.1 退出—接力 vs 买入持有 / 9.1 Exit-Relay vs Buy-and-Hold

55.5 年（1971–2026）、200 种子蒙特卡洛对照：退出—接力（清算顶部退出转现金，清算中段 4–6 季播种新叙事，形成期享 1.25 倍新泡沫弹性）对比纳指 Top50 买入持有（市值吐故纳新、无切换）：

A 55.5-year (1971–2026), 200-seed Monte Carlo duel: exit-relay (exit at clearing top to cash, seed new narratives 4–6 quarters into the clearing, harvest 1.25× new-bubble elasticity in formation) versus buy-and-hold Nasdaq Top50 (market-cap auto-turnover, no switching):

73 倍终值差距的全部来源是清算期的 6–10 个季度：纳指 Top50 在非清算段的复利能力与接力几乎不拉开差距，但三次清算（1973–74、2000–02、2008/2022）每次把净值拦腰砍断——复利的乘法结构决定 −77% 需要 +335% 才能回本。"不作新旧产业区分"的代价不是选错股票，而是放弃了"何时不持股"的权利。

The entire 73× terminal gap comes from the 6–10 quarters of clearing: Nasdaq Top50 compounds on par with relay outside clearings, but each of the three clearings cut cumulative wealth in half—multiplicative compounding means −77% requires +335% to recover. The cost of "not distinguishing new vs old industries" is not stock selection but surrendering the right to decide when not to hold.

9.2 泡沫追随 A–E：执行纪律的五档阶梯 / 9.2 Bubble-Follower A–E: The Five-Step Discipline Ladder

阶梯的两个刻度：(1) 滞后 1 季仅损失 1.4pp 年化（A→B），滞后 2 季损失 6.8pp（A→C）——执行延迟的代价是超线性的；(2) 永不退出比 60/40 基准的年化还低 3.2pp 且回撤 −57% vs −7%——"看对不做"比"不看"更差。

Two calibrations: (1) a one-quarter lag costs only 1.4pp CAGR (A→B) while two quarters cost 6.8pp (A→C)—the price of execution delay is superlinear; (2) never exiting underperforms even 60/40 by 3.2pp with −57% vs −7% drawdown—"seeing right but not acting" is worse than not looking.

9.3 八模型 × 三情景汇总 / 9.3 Eight Models × Three Scenarios

9.4 真宏观状态机：8 国 30 年含成本回测（US 现行 anchor-vote1-κ0.25） / 9.4 True-Macro Machines: Eight Countries, 30 Years, Cost-Inclusive (US now on anchor-vote1-κ0.25)

US 现行口径为 anchor-vote1-κ0.25 三件套 B1 主臂膀（2026-09-04 切换：信号锚=纳指综合（生产端 ONEQ 月末代理）× hold 持仓腿=NDX100 TR；Shiller 数据源，MV2 金融五票严格 Lag-1，月末 tick，§8.4；决策直觉：泡沫期尽量持有受泡沫膨胀影响但本身质量较高的股票）；A 臂双综合语义（14.31%/Sh1.241）于 2026-09-03~09-04 短暂生产后降为沿革。v2.2 三档状态机（any1/E10/rr5，同月决策口径）自 2026-09-03 起退役为沿革，七国仍以 v2.2 同构回测。0.15% 单边滑点 × 月度换手已计入全部 KPI（B1 换手/成本与 A 臂逐位一致——ALLOCS 键结构决定换手）。US 全表（1996-01 ~ 2026-06，366 月）：

The US current standard is the anchor-vote1-κ0.25 three-piece machine in its B1 primary arm (switched 2026-09-04: signal anchor = Nasdaq Composite (ONEQ monthly proxy in production) × hold-leg = NDX100 TR; Shiller data source, MV2 financial five votes under strict Lag-1, end-of-month tick, §8.4; rationale: during bubble episodes hold stocks influenced by the bubble yet of intrinsically higher quality); the A-arm dual-composite caliber (14.31%/Sh1.241) served briefly in production 2026-09-03~09-04 and is now lineage. The v2.2 three-tier machine (any1/E10/rr5, same-month decision convention) retired to legacy on 2026-09-03, with the seven countries still backtested on the v2.2 isomorph. 0.15% one-way slippage × monthly turnover is embedded in all KPIs (B1 turnover/cost identical to the A arm—turnover is determined by the ALLOCS key structure). US full table (1996-01 ~ 2026-06, 366 months):

anchor 现行触发统计（B1 与 A 臂共享同一 smap，阶段分布逐位一致）：总切换 76 次；阶段分布 hold 102 月（28%）/ wait 121 月（33%）/ full_exit 143 月（39%），锁定段（hold + wait）264 月占 61.7%。口径警示：v2.2 行为同月决策口径（当月票值驱动当月决策，含前视乐观偏置），anchor 行为严格 Lag-1 无前视口径——同框架 Lag-1 对照下 v2.2 仅 9.42%/Sh 0.652（V3rr）/ 11.23%/Sh 0.825（MV2），anchor B1+F1m 16.12%/Sh 1.309（A 臂 14.31%/Sh 1.241）显著占优；年化自 22.68% 降至 16.12% 是剔除前视偏置的口径换血代价，且换手 122.2 不足 v2.2 生产版六成。κ 对照：B1+F1m κ0 15.81% → κ0.25 16.12%（+0.31pp）；A 臂 κ0 14.00% → κ0.25 14.31%（+0.31pp）。B1−A 增量 +1.77pp 的单段依赖披露：终值差 68% 来自 1996-2000 泡沫期，2001 年以来年化差 +0.50pp（M-06 探针 V 归因法）。

Anchor-current trigger statistics (B1 shares the same smap as the A arm—tier distribution bit-identical): 76 total switches; tier distribution hold 102m (28%) / wait 121m (33%) / full_exit 143m (39%), locked segments (hold + wait) 264m at 61.7%. Convention caveat: the v2.2 rows use the same-month decision convention (same-month votes driving same-month decisions, an optimistic look-ahead bias), while the anchor row is strict Lag-1 with no look-ahead—under the same Lag-1 convention v2.2 reads only 9.42%/Sh 0.652 (V3rr) / 11.23%/Sh 0.825 (MV2), so anchor B1's 16.12%/Sh 1.309 (A arm 14.31%/Sh 1.241) dominates; the fall from 22.68% to 16.12% is the price of purging look-ahead bias, with turnover 122.2 under 60% of v2.2's production version. κ controls: B1 κ0 15.81% → κ0.25 16.12% (+0.31pp); A arm κ0 14.00% → κ0.25 14.31% (+0.31pp). Single-segment dependence disclosed: 68% of the B1−A terminal gap accrues in 1996-2000, with +0.50pp/yr from 2001 onward (M-06 probe V attribution).

收益率口径披露（2026-09-04 B1 重算，全表适用）：本表及全站所有「年化」一律为时间加权复合年化 TWR-CAGR（终值^(12/N)−1，几何复利口径；期初单一现金流、月度几何连乘、收益全额再投资——该设定下 TWR = MWR/IRR 同值）。拆分对照（US B1 净口径 vs 持仓腿基准 NDX100 TR）：算术年化（月均×12）15.76%、历年收益算术平均 16.7%、NDX100 TR 三口径 14.90%/16.89%/19.62%——算术口径给高波动基准虚增 +1.99pp「波动补贴」（vol drag 反向），跨口径比较无效；anchor 低波动（12.2% vs 24.1%）使 CAGR（16.12%）反超算术年化（15.76%），复利凸性增益大于波动损耗（A−C=−0.36pp）。A 臂沿革对照（基准纳指综合）：净三口径 14.31%/14.09%/14.81%、综合 12.12%/14.00%/15.63%（+1.88pp 波动补贴）。完整说明见主线〇报告 returns-caliber-report.html；数据 outputs/anchor_arith_disclosure.json。

Returns-caliber disclosure (2026-09-04 B1 recomputation, applies to all tables): every "annualized" figure in this paper and across the site is time-weighted CAGR (final^(12/N)−1, geometric compounding; a single initial cash flow, monthly geometric chaining, full reinvestment—under which TWR = MWR/IRR). Split reference (US B1 net vs the holding-leg benchmark NDX100 TR): arithmetic annualization (monthly mean × 12) 15.76%, arithmetic mean of annual returns 16.7%, NDX100 TR three-caliber 14.90%/16.89%/19.62%—the arithmetic caliber inflates high-volatility benchmarks by +1.99pp of "volatility subsidy" (the inverse of vol drag), invalidating cross-caliber comparison; anchor's low volatility (12.2% vs 24.1%) lets CAGR (16.12%) exceed the arithmetic rate (15.76%), compounding convexity gain outweighing the volatility loss (A−C = −0.36pp). A-arm lineage reference (benchmark Nasdaq Composite): net three-caliber 14.31%/14.09%/14.81%, composite 12.12%/14.00%/15.63% (+1.88pp volatility subsidy). Full statement in the Main-Line-0 report returns-caliber-report.html; data at outputs/anchor_arith_disclosure.json.

七国全表（1996-02 ~ 2026-06，365 月，夏普降序）：

Seven-country table (1996-02 ~ 2026-06, 365 months, Sharpe-descending):

横向洞察：IN/DE/JP 构成夏普三强梯队；KR 年化高但波动 17.43% 拉低夏普（韩股 β 高）；UK 是七国夏普下界（FTSE 长期低 β、防御属性强、票源敏感度低）；七国 Pain 减轻 56%~72%、MDD 普遍从 −48%~−71% 收窄到 −19%~−24%。回测研究未实盘：七国 v2.2 画像已并入 strategy_profiles.json（共 8 份，US 档案已于 2026-09-03 切换为 anchor-vote1-κ0.25 现行口径），但 sim-cn/jp/kr/de/fr/uk/in 实盘仍跑各 sim 的 tech5（票源 ②）与本节 Shiller 回测（票源 ④）不同源；七国沿用 v2.2 三档状态机（§8.4 沿革口径），迁移至 anchor 三件套的研究见 academic/模型优化记录_anchor替换上线_2026-09-03.html。

Cross-sectional insights: IN/DE/JP form the Sharpe top tier; KR's high CAGR is dragged by 17.43% volatility (high-β KOSPI); UK is the seven-country floor (low-β FTSE, defensive, vote-source insensitive); pain relief runs 56%~72% with MDD compressing from −48%~−71% to −19%~−24%. Backtest-not-live caveat: the seven v2.2 profiles are merged into strategy_profiles.json (8 entries total; the US profile switched to the anchor-vote1-κ0.25 current standard on 2026-09-03), but sim-cn/jp/kr/de/fr/uk/in run live on tech5 (source ②)—a different source from this section's Shiller backtest (source ④); the seven countries retain the v2.2 three-tier machine (§8.4 legacy), with the anchor-migration study documented in academic/模型优化记录_anchor替换上线_2026-09-03.html.

9.5 E10R5 熔断：八国验证 / 9.5 The E10R5 Circuit Breaker: Eight-Country Validation

E10R5 规则：动量指数自 12 个月高点回撤 ≥10% → 降级退出档（动量仓位 70% → 45%）；自熔断期低点反弹 ≥5% → 恢复。8 国 30 年月度回测全部正贡献：年化增益 Δ +1.87 ~ +5.97pp，回撤同步收窄，高波动市场增益最大。

The E10R5 rule: momentum index down ≥10% from its 12-month high → downgrade to the exit tier (momentum exposure 70% → 45%); rebound ≥5% off the circuit-breaker low → restore. Across eight countries and 30 years, contributions are positive everywhere: annualized gain Δ +1.87 ~ +5.97pp with simultaneous drawdown reduction.

双档 = 追加第二档（回撤 ≥20% → 全档退出，现金+黄金 ≥65%），增益进一步放大至 +3.25 ~ +11.19pp。全部国家在收益-回撤象限中向右上方移动（年化抬升 + 回撤收窄同时成立），无任何一国的负样本。三国一致单调：8/16 > 10/20 > 12/24——越敏感的熔断增益越大，10/20 是趋势上的保守点而非过拟合尖峰。

Dual tier = adding a second tier (drawdown ≥20% → full exit, cash+gold ≥65%), further amplifying gains to +3.25 ~ +11.19pp. Every country moves up-and-right in the return-drawdown quadrant; zero negative samples. Monotone across three countries: 8/16 > 10/20 > 12/24—more sensitive breakers earn more; 10/20 is a conservative point on the trend, not an overfit spike.

阈值稳健性完整表 / Threshold Robustness (Full Table)

口径注记：本验证采用「恒持基准」对照（base 始终满配动量 70%），隔离出熔断层的纯增量，与 §9.4 的 v2.2 多资产框架（22.68%）是不同框架，E10R5 在此作为独立可叠加的补充层验证。

Calibration note: this validation uses an always-invested baseline (momentum 70% throughout) to isolate the breaker's pure increment—a different framework from §9.4's v2.2 multi-asset machine (22.68%); E10R5 is validated here as an independently stackable layer.

9.6 CN 实盘 QDII：ESC3_025 升坡门控 / 9.6 CN Live QDII: The ESC3_025 Ramp Gate

现行 QDII 模拟盘执行规格（cn-qdii 自 2026-08-22 起执行）。信号层与执行节奏与固定 5% 版完全相同——月末评估 + T1 次交易日收盘 + pending 取消，唯一差异是买入腿门控从「固定 5%」改为「升坡 ESC3_025」：

The live QDII execution spec (in force for cn-qdii since 2026-08-22). Signal layer and cadence are identical to the fixed-5% version; the sole change is the buy-leg gate from fixed 5% to the ESC3_025 ramp:

ESC3_025 全谱最优：比固定 5% 高 0.85pp/年，比无门控高 2.50pp/年；累计溢价成本 17.14pp（固定 5% 为 20.12pp，无门控为 30.55pp）。回测窗口 2017-01 ~ 2026-06（114 月，2,291 交易日），成本 0.15% 单边。

ESC3_025 is the full-spectrum optimum: +0.85pp/yr over fixed 5%, +2.50pp/yr over no gate; cumulative premium cost 17.14pp. Backtest window 2017-01 ~ 2026-06 (114 months, 2,291 trading days), cost 0.15% one-way.

9.7 未来版：4.6 年高频稳态下的重估 / 9.7 The Future Version: Repricing Under the 4.6-Year Steady State

把 §9.1–9.3 的模型投入预测的未来 24 年（6 轮 4.6 年周期、闪电清算 12.4 月、清算幅度 −33~−44%，参数来自 §5.6 时间方程与周期速率模型），200 种子蒙特卡洛：

Running the §9.1–9.3 models over the predicted next 24 years (six 4.6-year cycles, 12.4-month clearings, magnitudes −33~−44%, parameters from the §5.6 timing equations and the cycle-rate model), 200-seed Monte Carlo:

未来版第一定律："持有"的期望收益 ≈ 0，收益全部转移到"切换"。纳指 Top50 的终值中位是 1x（形成期 +105% 与清算 −55% 每 4.6 年对冲一次）；高频稳态是切换者的提款机与持有者的绞肉机。若本轮清算如预测在 12–14 个月内完成且出清完成度仅 0.75–0.8，则"清算 → 直接再膨胀"的高频稳态即告确认。

The future version's first law: expected return of "holding" ≈ 0; returns migrate entirely to "switching." Nasdaq Top50's median terminal value is 1× (formation +105% and clearing −55% offset every 4.6 years); the high-frequency steady state is the switcher's ATM and the holder's grinder. If the current clearing completes in 12–14 months at completion 0.75–0.8, the "clearing → direct reflation" steady state is thereby confirmed.

9.8 追踪模型 2026 实时读数 / 9.8 Tracking Model: 2026 Live Readings

追踪模型（骑泡沫）的当前面板，展示引擎层诸模块在实盘口径下的合成输出（数据截至 2026-07/08）：

The tracking model's (bubble-riding) current panel—how the engine-layer modules synthesize into live output (data as of 2026-07/08):

据此执行的动作：P≈7 上升 + 退出投票 2.5/5 → 不清仓但启动退出计时，动量仓降至半仓——QQQ/科技龙头（减仓中，污染度最高的纯叙事、负现金流、靠再融资标的先减）50%；观察名单 VST/CEG/SMR/防务（下轮角色种子：电力与能源瓶颈 + 产业重合度最高的下轮叙事）20%，待清算中段（4–6 季）播种；现金 30%（五票制退出期权金）。三情景回检：A 估值熊（45%）−8% / B 信用熊（30%）−12% / C 滞胀熊（25%）对应防御模型接棒。

Actions taken: P≈7 rising + exit vote 2.5/5 → no full liquidation, but exit-clock started and momentum cut to half—QQQ/tech leaders (de-risking; highest-pollution pure-narrative, negative-cashflow, refinancing-dependent names cut first) 50%; watchlist VST/CEG/SMR/defense (next-cycle role seeds: power and the energy bottleneck, the highest industrial-overlap next narratives) 20%, to be planted mid-clearing (4–6 quarters); cash 30% (five-vote exit option premium). Three-scenario replay: A valuation bear (45%) −8% / B credit bear (30%) −12% / C stagflation bear (25%), with the defensive model taking over.

9.9 理论前置：国际收支第六票（B4）/ 9.9 Theoretical Foundation: The Sixth Balance-of-Payments Vote (B4)

B4 提议把「国际收支脆弱度」升格为投票层第六票（07 优化清单 · 待办 · 不急），B4 是把同一脆弱度升为与 P/IPO/BII/Top10/居民债务并列的投票输入，走 06 定版规格变更流程（先历史回放再回测对照）。截至本文，B4 尚无回测数字，本节给出理论支持矩阵与数据内核（信贷脉冲）的框架性论证，作为测量层 §3 的候选扩展。

B4 proposes promoting "balance-of-payments fragility" to a sixth vote in the voting layer (roadmap item 07 · pending · not urgent), elevating the same fragility to a vote input on par with P/IPO/BII/Top10/household debt, following the 06 spec-change process (historical replay before backtest comparison). As of this paper B4 has no backtest numbers; this section provides the theoretical support matrix and the credit-impulse data kernel as a framework-level argument and a candidate extension of the §3 measurement layer.

理论支持矩阵（奥派 / MMT / 三存量）：(1) 奥地利学派——开放经济 ABCT 把资本流入视为利率信号的国际传导：外资涌入压低国内融资成本 → 信贷扩张 → 不当投资；外资是可逆转的抵押品，一旦全球风险偏好反转（sudden stop），抵押品被抽走，清算被外部强制。1997 亚洲危机正是这一机制的教科书案例（资本流入驱动的信贷扩张 + 可逆外资抵押品）。

Theoretical support matrix (Austrian / MMT / three-stock): (1) Austrian—open-economy ABCT treats capital inflows as the international transmission of the interest-rate signal: inflows depress domestic funding costs → credit expansion → malinvestment; foreign capital is reversible collateral—when global risk appetite reverses (sudden stop), the collateral is withdrawn and liquidation is externally forced. The 1997 Asian crisis is the textbook case of capital-inflow-driven credit expansion plus reversible foreign collateral.

(2) MMT——部门平衡恒等式提供核算基础：

(2) MMT—the sectoral-balances identity provides the accounting foundation:

经常账户赤字 = 私人部门储蓄缺口或财政扩张的外部镜像。但 MMT 的分层约束是关键：储备货币发行国（US）无外部硬约束，外部赤字由本币融资，第六票在 US 上弱化；非储备货币 + 外币债务国家（新兴市场）则受 hard external constraint（原罪假说 original sin），第六票在七国上强化——国际收支票的信息量集中在非储备货币国。

A current-account deficit is the external mirror of a private saving gap or fiscal expansion. But MMT's tiering constraint is the key: reserve-currency issuers (US) face no hard external constraint—external deficits are funded in domestic currency, so the sixth vote weakens for the US; non-reserve-currency countries with foreign-currency external debt face a hard external constraint (original sin), strengthening the sixth vote for the seven countries—the balance-of-payments vote's information content is concentrated in non-reserve economies.

(3) 三存量框架——当前 E/P/M 三轴没有外部存量轴：B4 是「第四存量（NIIP/外债/储备）」的候选，把交叉抵押机制国际化（外部可逆抵押品与国内三存量互相掩盖）。实证背书：FRAG 在回撤方程（式 25）中系数 5.62、Tornado 敏感度第一（±9pp）——同一脆弱度的投票层表达顺理成章。

(3) Three-stock framework—the current E/P/M axes have no external stock axis: B4 is the candidate "fourth stock (NIIP/external debt/reserves)," internationalizing the cross-collateralization mechanism (reversible external collateral masking the three domestic stocks). Empirical endorsement: FRAG enters the drawdown equation (Eq. 25) at coefficient 5.62 with #1 Tornado sensitivity (±9pp)—making a voting-layer expression of the same fragility natural.

数据内核：信贷脉冲的理论框架——信贷脉冲 (credit impulse) 由 Biggs-Mayer-Pick (2010) 提出：定义为信贷存量流量的变化（2yrΔ 私人信贷/GDP），领先 GDP 1–2 季。其理论根有四支：Fisher 债务通缩（债务存量收缩的破坏性）、Minsky 三阶段（脉冲见顶 ≈ 投机→庞氏转换期）、奥地利 ABCT（扩张的边际动能衰竭）、Borio 金融周期（信贷/GDP 缺口是 Basel III 逆周期资本缓冲的基准，与 BIS WS_TC 数据同源）。在三存量框架中，信贷脉冲 = Δμ = M 存量轴的二阶导，是「M 轴加速度」的观测投影——这也是 B4 若落地的统计基础：国际收支脆弱度 × 信贷脉冲的组合，将同时捕捉「外部抵押品可逆性」与「国内信用动能衰竭」。

Data kernel: the theoretical framework of the credit impulse—the credit impulse (Biggs-Mayer-Pick 2010) is defined as the change in the flow of credit stocks (2yrΔ private credit/GDP) and leads GDP by 1–2 quarters. Its four theoretical roots: Fisher's debt-deflation (the destructiveness of debt-stock contraction), Minsky's three stages (impulse peak ≈ the speculative→Ponzi transition), Austrian ABCT (the exhaustion of expansion's marginal momentum), and Borio's financial cycle (the credit/GDP gap as the Basel III countercyclical buffer benchmark, same source as BIS WS_TC). Within the three-stock framework the credit impulse equals Δμ—the second derivative of the M axis—the observable projection of "M-axis acceleration," and the statistical foundation of B4 if enacted: a balance-of-payments fragility × credit impulse combination would capture both "external collateral reversibility" and "domestic credit momentum exhaustion" simultaneously.

诚实边界：B4 仍是「待办 · 不急」项——无回测数字、未进规格变更流程；其理论前置（外部存量轴第四慢变量论证 + MMT 分层国别启用范围）已齐备，下一步是 1997/2008/2011 场景回检与七国回测对照。若落地，将把测量层从「五票」扩展为「六票」。

Honest boundary: B4 remains a "pending · not urgent" item—no backtest numbers, not yet in the spec-change pipeline; its theoretical prerequisites (the fourth-stock external-axis argument plus the MMT-tiered country enablement scope) are complete, and the next step is 1997/2008/2011 scenario replay and seven-country backtest comparison. If enacted, the measurement layer would expand from five votes to six.

十、票源对照实验：tech5 vs macro5 vs no_vote / 10. Vote-Source Control Experiment

本章是一组控制实验，直接回答一个关键问题：在同一状态机框架（E10 + rr5 + any1）下，不同票源驱动的差异有多大？实验设计：固定状态机和资产配置不变，仅切换 any1 通道的票源——V1 = tech5（票源 ②，技术指标）、V2 = macro5（票源 ④，Shiller 真宏观）、V4 = no_vote（纯 E10 + rr5，无 any1 通道）。BASE = E10 + rr5 熔断基准（无 any1 但含价格通道）。

This chapter is a set of control experiments directly answering a key question: within the same state machine framework (E10 + rr5 + any1), how much difference does the vote source make? Design: fix the state machine and allocations, switch only the any1 channel's vote source—V1 = tech5 (source ②, technical indicators), V2 = macro5 (source ④, Shiller true-macro), V4 = no_vote (E10 + rr5 only, no any1 channel). BASE = E10 + rr5 breaker baseline (no any1 but with price channels).

10.1 US SPX 控制实验 / 10.1 US SPX Control Experiment

以美国标普 500 为测试基准（1996-02 ~ 2026-06，365 月），四个变体的含成本结果如下：

Using US S&P 500 as the benchmark (1996-02 ~ 2026-06, 365 months), the four variants' cost-inclusive results:

关键发现：(1) V1 tech5 在年化和夏普上均优于 V2 macro5——技术信号在 SPX 上比宏观信号增益更大（+1.51pp 年化、+0.099 夏普）；(2) V2 macro5 与 V4 no_vote 几乎无差异（13.47% vs 13.46%）——宏观票在 any1 通道的增量贡献接近零；(3) V4 no_vote（纯价格通道）仍达 13.46%/1.210——状态机的结构性优势独立于票源。

Key findings: (1) V1 tech5 outperforms V2 macro5 in both CAGR and Sharpe—technical signals deliver more increment than macro signals on SPX (+1.51pp CAGR, +0.099 Sharpe); (2) V2 macro5 and V4 no_vote are nearly identical (13.47% vs 13.46%)—macro votes' incremental contribution via the any1 channel is near zero; (3) V4 no_vote (price channels only) still reaches 13.46%/1.210—the state machine's structural advantage is vote-source-independent.

10.2 九国对照：tech5 增量 vs macro5 增量 / 10.2 Nine-Country Comparison

将同一控制实验扩展到 9 个国家/指数（1996-01 ~ 2026-06，366 月），tech5_incre 和 macro5_incre 分别表示 V1 和 V2 相对于 BASE 的年化增量（pp）：

Extending the same control experiment to 9 countries/indices (1996-01 ~ 2026-06, 366 months), tech5_incre and macro5_incre denote V1 and V2's annualized increment (pp) over BASE:

口径披露：本表为 tech5 对照实验的历史档案快照（数据时点 2026-08-21，产物 v22_macrovote_vs_tech5_11b.json；tech5 状态机已于 2026-08-30 退役）。七国权益数据此后经历 2026-08-27 intl 月份错位修复与 2026-08-31 孟买SENSEX 口径勘误，IN 序列实为 BSE SENSEX；本表数字保留实验时点原值不追溯重算，与现行生产回测（intl_macro_v22_bt.json）存在数据版本差异。

Data caveat: This table is an archived snapshot of the tech5 control experiment (data vintage 2026-08-21, artifact v22_macrovote_vs_tech5_11b.json; the tech5 state machine was retired on 2026-08-30). The seven-country equity data subsequently underwent the 2026-08-27 intl month-misalignment fix and the 2026-08-31 SENSEX relabeling (the IN series is in fact BSE SENSEX); figures here are preserved as-of the experiment date and are not retroactively recomputed, hence a data-vintage gap versus the current production backtest (intl_macro_v22_bt.json).

九国对照的三个核心发现：

Three core findings from the nine-country comparison:

发现 1：tech5 全面优于 macro5。9 国中 tech5 增量全部为正（+0.47 ~ +2.49pp），而 macro5 增量在 3 国为负（CN −3.70pp、DE −0.12pp）。tech5 增量最高的是 FR（+2.49pp），最低的是 KR（+0.47pp）。US 纳斯达克 tech5 +1.69pp vs macro5 +0.02pp——技术信号在该市场几乎完全压倒宏观信号。

Finding 1: tech5 comprehensively outperforms macro5. All 9 countries show positive tech5 increments (+0.47 ~ +2.49pp), while macro5 increments are negative in 3 countries (CN −3.70pp, DE −0.12pp). The highest tech5 increment is FR (+2.49pp), the lowest is KR (+0.47pp). US Nasdaq: tech5 +1.69pp vs macro5 +0.02pp—technical signals almost completely dominate macro signals in this market.

发现 2：CN 是 macro5 有害的唯一显著案例。中国 A 股 macro5 增量 = −3.70pp（年化从 19.70% 降至 14.64%），是 9 国中唯一的显著负值。可能原因：中国宏观周期与货币/能源/人口票的 Shiller 数据重建口径存在结构性错配——Shiller 真宏观票的 CAPE 分位、CPI、实际利率等指标在 A 股市场的信号方向可能与美股相反。这一发现严格遵循票源铁律：票源 ④（Shiller 真宏观）的结论不可外推至票源 ①（实盘真宏观），也不可外推至票源 ②（tech5）。

Finding 2: CN is the only significant case where macro5 is harmful. China A-share macro5 increment = −3.70pp (CAGR drops from 19.70% to 14.64%), the only significant negative among 9 countries. Possible cause: structural mismatch between China's macro cycle and the Shiller dataset's reconstructed metrics—Shiller true-macro vote signals (CAPE percentile, CPI, real rate) may have opposite directional implications in the A-share market. This finding strictly follows the vote-source iron law: source ④ (Shiller true-macro) conclusions cannot be extrapolated to source ① (live true-macro) or source ② (tech5).

发现 3：V4 no_vote（纯价格通道）在所有国家表现稳健。9 国 V4 年化范围 9.04% ~ 19.38%，均显著优于纯指数持有。这印证了 §9.4 的"诚实下限"结论：状态机本身具有结构性优势，any1 通道的票源选择是精化而非必需。

Finding 3: V4 no_vote (price channels only) is robust across all countries. V4 CAGR ranges 9.04% ~ 19.38% across 9 countries, all significantly outperforming buy-and-hold. This corroborates §9.4's "honest floor" conclusion: the state machine itself has structural advantage; the any1 channel's vote source is refinement, not necessity.

10.3 历史真宏观回测的不可行性披露 / 10.3 Infeasibility of Historical True-Macro Backtest

本章实验使用票源 ④（Shiller 真宏观）作为 macro5 的实现。票源 ①（实盘真宏观，§3.2 规格的五票）无法用于历史回测——其五个票项（规则化 P / IPO 温度计 / BII 连续两季 / Top10 集中度 / 居民债务-GDP）需要 Ritter IPO 库、BIS 信贷统计、FINRA 保证金数据等月度长史序列，而这些数据要么不存在 1996 年以前的月度序列，要么口径不一致无法重建。因此，本章的 macro5 结果是 Shiller proxy 的结果，不是真宏观的结果——两者口径差异通过前向双轨验证弥合（§3.5）。

This chapter's experiments use source ④ (Shiller true-macro) as the macro5 implementation. Source ① (live true-macro, the §3.2 specification's five votes) cannot be used for historical backtesting—its five vote items (rule-based P / IPO thermometer / two-quarter BII / Top10 concentration / household debt-GDP) require monthly long-history series from the Ritter IPO database, BIS credit statistics, FINRA margin data, etc., which either lack pre-1996 monthly series or have inconsistent calibers preventing reconstruction. Therefore, this chapter's macro5 results are Shiller proxy results, not true-macro results—the caliber gap is bridged through forward dual-track validation (§3.5).

十一、成本与诚实边界 / 11. Costs and Honest Boundaries

成本披露 / Cost Disclosure：以上策略回报均为不含交易成本的模型估算。L1 滑点（买卖合计 0.15–0.30%）、L2 佣金税费（美股 ~0.008%、A 股 ~0.08%、港股 ~0.28%）、L3 市场冲击（大资金 0.05–0.20%，平方根模型估计）；年化成本拖累：低频策略 0.2–0.5pp、高频策略 0.8–2.0pp，含成本真实回报约低 1–3pp。合成回溯使用量级口径的阶段收益与事后划分的 regime，滞后惩罚只模拟了 6 个月识别延迟。结论的结构（清算期防守贡献大部分超额、R3 是积累引擎、周期在加速）稳健；绝对倍数不稳健。
All strategy returns above are frictionless model estimates. L1 slippage (0.15–0.30% round trip), L2 commissions/taxes (~0.008% US, ~0.08% A-share, ~0.28% HK), L3 market impact (0.05–0.20%, square-root model); annual drag 0.2–0.5pp low-frequency, 0.8–2.0pp high-frequency—real returns roughly 1–3pp lower. Synthetic backtests use magnitude-calibrated regime returns with ex-post regime splits and only a 6-month recognition lag. The structure of conclusions is robust; the absolute multiples are not.

十二、模型边界与局限性 / 12. Model Boundaries and Limitations

建模的核心原则是：模型回答"如果……就会……"，而非预测"一定会……"。基于 57 事件 × 29 国的实证验证结果与全部回测证据，以下局限性必须明确声明。注意本章与 §7.7 的分工：本章刻画"模型在哪些方面不精确"（误差带型局限），§7.7 的 F1–F6 刻画"模型在哪些条件下整体失效"（范式适用型边界）。

The modeling principle: the model answers "if... then..." rather than predicting "will definitely happen." Based on empirical validation across 57 events × 29 countries and all backtest evidence, the following limitations must be explicitly declared. Note the division of labor with §7.7: this chapter describes where the model is imprecise (error-band limitations), while §7.7's F1–F6 describe under what conditions it fails wholesale (paradigm-applicability boundaries).

局限性 1 / Limitation 1 · n = 4 插值性质（2026-09-02 已由 LOEO 替代验证）/ Interpolation Nature (superseded by LOEO on 2026-09-02)：回撤方程权重原为四次历史出清的精确拟合（4 方程 4 未知数，R²=1.0 的代数必然），样本外验证缺失。2026-09-02 以 57 事件库机器可读化（outputs/crisis_db57.json，论文 §5.3 分布对账 12/12 一致）为底座，用 leave-one-event-out（LOEO，57 折）重做验证，结论三层：① n=4 权重在 LOEO 下对原四事件 3/4 系统性低估（US 2000 预测 28.9 vs 实际 49.0，误差 −20.1pp）——「精确」确系插值假象；② 式 (26) 的符号约束（β₂>0）在 n=4 下空转、在全 57 无截距 OLS 下直接违反（RCS 系数 −1.19 < 0，FRAG 膨胀至 9.60），n=4 的系数结构在全样本不存在；③ 诚实样本外精度：v3.3 四因子形式 LOEO ρ=0.582 / MAE 15.08pp，7F 论文模型 LOEO ρ=0.915 / MAE 8.37pp——回撤预测力主要来自危机分类特征（CRISIS_LEVEL，Δρ=+0.247），而非三存量连续值本身。因此系数精确值（如 8.54 vs 8.50）确实不应被解读，且现在有了数字级理由；详见 academic/35-loeo-drawdown-57.md。

局限性 2 / Limitation 2 · 2026 为进行中状态 / 2026 Is In Progress：本框架对 2026 的全部读数（BII 6.74、RCS 7.95、S''' = −3.35、预测回撤 65.8%）是事前预测而非事后拟合，其最终检验只能由市场完成——这既是弱点，也是框架的可证伪性所在。

局限性 3 / Limitation 3 · 引爆点不可预测 / Trigger Unpredictability：模型预测泡沫膨胀的总量和清算时的回撤深度，但无法识别具体引爆器——是 AI 财报不及预期、私募信贷爆仓、还是地缘冲突。引爆器是随机变量 ξ，模型不试图拟合它。

局限性 4 / Limitation 4 · 全球系统性冲击预测力薄弱 / Weak Global Shock Prediction：LOCO-CV 显示，global_shock 类型事件的相关系数仅 ρ=0.267——尽管 MAE 不高（6.3pp），模型无法有效区分全球冲击事件内部的严重度排序。

局限性 5 / Limitation 5 · 低烈度事件方向预测率低 / Low Direction Accuracy for L0：L0 级（技术回调）的方向正确率仅 62%——模型对"小修正还是大危机"的区分能力在低烈度区间显著下降。

局限性 6 / Limitation 6 · 不预测时点 / No Timing Claims：模型刻画"到期时的清算深度"，时点信息仅来自形成期方程的寿命百分比（当前 96%）与 SBRI Tier 2 的条件概率。

局限性 7 / Limitation 7 · 口径依赖 / Calibration Dependence：Ω 的两种分解粒度（0.15 / 0.38）、平均周期与终态周期（4.6 / 4.5 年）、κ 的版本谱系（2.0 / 3.47 / 3.92）在本文中已统一为主口径并逐处标注，但跨版本比较时必须显式声明版本。

局限性 8 / Limitation 8 · 票源不可互换 / Vote Sources Non-Interchangeable：宏观五票存在四种语义（实盘真宏观 / 回测 tech5 / 实验 macro_proxy / Shiller 真宏观），本文全部实证已逐处指明票源；跨票源结论不可相互印证，实盘票源 ① 无月度长史、不可回测。第十章的九国对照实验进一步证实：tech5 在 9 国全面正贡献，macro5 在 CN 有害（−3.7pp）——票源选择对结论有决定性影响。

局限性 9 / Limitation 9 · 前视溢价 / Look-Ahead Premium：七国估值票 m1 的 CAPE 滚动窗口分位判定含回看校准成分，按 freeze-wait-z20 报告核算约 +0.31pp/年。

局限性 10 / Limitation 10 · AI 可能改写剧本 / AI May Rewrite the Script：若 AI 的 TFP 贡献在未来三年内兑现为年化 +1.5—2% 的持续提升，S''' 的读数需要大幅修正。三笔支取框架的核心命题是"存量红利已经吃完"，但技术创新可能开出新的增量空间。当前判断：此事在发生，但速度不够快、广度不够深——尚未改写"存量出清"的剧本。

局限性 11 / Limitation 11 · 尾部风险不可对冲 / Unhedgeable Tail Risk：L3 场景——大国直接冲突或全球结算体系分裂——在任何风险管理框架内都无法对冲。这不是模型设计的问题，是概率分布的尾部根本不服从正态分布。

局限性 12 / Limitation 12 · 模拟盘性质 / Simulation Nature：策略回测含理想化假设（零摩擦或 0.15% 单边滑点近似、事后 regime、量级口径收益），绝对倍数不可外推；八国 E10R5 与 CN QDII 门控回测为研究性质，八国 v2.2 画像尚未在 sim 实盘执行（实盘仍跑 tech5 票源）。

十三、结论 / 13. Conclusion

三笔跨期存量支取框架把两百年的泡沫周期压缩为可操作的量化链条：三条慢变量（E 采出进度、TFR、债务/GDP；E 轴 2026-09-04 物理重锚，旧 EROI 口径留痕）→ 四个指标（BII、RCS、FRAG、S'''）→ 一个回撤方程（v3.3，含 Gini 三通道与 γ-GSM 修正）→ 一族时间方程（形成期加速、冻结系数、4.5 年高频稳态）→ 一个全量并入的 SBRI 预警层（Tier 1–4 四层架构 + γ-GSM 修正 + 催化剂接近度 + SBRI × 退出联动矩阵 + F1–F6 失效条件 + 四案例历史回溯）→ 一套完整披露的引擎层（票源铁律四源 + v2.2 三档状态机 + 五票退出进入 + E10R5 双档熔断 + ESC3_025 升坡门控）→ 一部回测全集（A–E 泡沫追随阶梯、八模型三情景、v2.2 真宏观 8 国含成本、E10R5 八国验证、CN QDII 门控全谱、票源控制实验九国对照）。2026 时点的判定：BII 6.74 × RCS 7.95 的信用×人口双线共振叠加 E 线深耗竭（62.5% 已采、非临界）、S''' 首次深负、κ'(G, GSM) = 3.92 的双重放大，指向中位 65.8%（58–73%）的出清深度与已在当下的切换窗口。

The Three-Stock Drawdowns framework compresses two centuries of bubble cycles into an operable quantitative chain: three slow variables (E depletion progress, TFR, debt/GDP; the E axis physically re-anchored 2026-09-04, old EROI caliber archived) → four indicators (BII, RCS, FRAG, S''') → one drawdown equation (v3.3, with Gini three-channel and γ-GSM corrections) → a family of timing equations (formation acceleration, freeze coefficient, 4.5-year steady state) → a fully disclosed engine layer (four vote-source iron rules + v2.2 three-tier state machine + five-vote exit/re-entry + E10R5 dual-tier breaker + ESC3_025 ramp gate) → a complete backtest suite (A–E bubble-riding ladder, eight models × three scenarios, v2.2 true-macro eight countries cost-inclusive, E10R5 eight-country validation, CN QDII gate spectrum, vote-source control experiment nine-country comparison). The 2026 verdict: a credit × demographics dual resonance of BII 6.74 × RCS 7.95 compounded with deep-but-sub-critical energy depletion (62.5% extracted), first deep-negative S''', and double amplification at κ'(G, GSM) = 3.92 point to a median clearing depth of 65.8% (58–73%) and a switching window already open.

核心发现总结为四个命题：

1. 双线共振假说（原"三线共振假说" · 2026-09-04 实证修正）：理论上，当 $S_E \to 0 \land S_P \to 0 \land S_M \to 0$ 同时成立（三线共振，理论极限态），系统性危机的概率从独立事件乘积跃升至近 1。实证上，百年样本中完整触发的仅有信用×人口双线（$S_P \to 0 \land S_M \to 0$）——1929 与 2026 两次；能源线在两个时点均未临界：1929 池 98.9% 完整（E 线安全区，降为对照组），2026 已采 62.5%、发现账剩余 29.1 年（深耗竭但远非零）。修正后的可操作含义：出清触发 = 信用×人口双线共振；出清深度由清算/冻结系数决定；E 线深耗竭为放大器（压低长期增长天花板）而非独立触发器。
2. 清算深度由冻结系数主导：$\delta(S'')$ 包含了 Gini 分配不均、底层空心化和法币信心折价的三重放大效应。2026 年 $\delta=+13.13$pp，使预测回撤从 50.2% 的基础值跃升至 65.8%。7F 模型中 QE_DUMMY 和 QE_TAPER 两个政策变量的入选，从数据角度验证了"冻结系数"的理论预期。
3. 冻结系数的持续升高正在改变危机形态：从"自然振荡"推入"积累-灾变"制式——形成期被拉长、清算期被压缩，未来危机的相对回撤幅度将收敛，但绝对冲击量将指数化增长。被推迟的清算以复利形式积累，政策制定者需认识到冻结不是免费延期，而是滚了利息的贷款。
4. 票源铁律是方法论的底层纪律：四种票源语义（实盘真宏观 / 回测 tech5 / 实验 proxy / Shiller 真宏观）在任何实证中禁混用。第十章的九国对照实验证明：tech5 增量在 9 国全部为正（+0.47~+2.49pp），macro5 增量在 CN 为负（−3.7pp）——票源选择对结论有决定性影响。实盘真宏观五票的规则化规格（§3.2-3.3）已定版但首次回算未执行，回测使用 Shiller 真宏观 proxy，两者口径差异通过前向双轨验证弥合。

Four core findings:

1. Dual Resonance Hypothesis (formerly the "Triple Resonance Hypothesis," empirically revised 2026-09-04): In theory, when $S_E \to 0 \land S_P \to 0 \land S_M \to 0$ hold simultaneously (triple resonance, a theoretical limiting state), systemic crisis probability jumps from the product of independent events to near-unity. Empirically, only the credit × demographics pair ($S_P \to 0 \land S_M \to 0$) has fully triggered in the century sample—1929 and 2026; the energy line was sub-critical at both points: 98.9% of the pool intact in 1929 (E-safe, demoted to control), and 62.5% extracted with 29.1 discovery-account years remaining in 2026 (deep depletion, yet far from zero). Operational implication after revision: the clearing trigger is the credit × demographics dual resonance; clearing depth is governed by the freeze coefficient; deep energy depletion acts as an amplifier (capping long-run growth) rather than an independent trigger.
2. Clearing depth is dominated by the freeze coefficient: $\delta(S'')$ incorporates triple amplification from Gini inequality, bottom-hollowing, and fiat confidence discount. At +13.13pp in 2026, it raises predicted drawdown from a base of 50.2% to 65.8%. The inclusion of QE_DUMMY and QE_TAPER in the 7F model empirically validates the "freeze coefficient" proposition.
3. Rising freeze coefficient is transforming crisis morphology: The system is pushed from "relaxation oscillation" into an "accumulation-catastrophe" regime—formation phases lengthen, clearing phases compress. Future relative drawdowns will converge, but absolute impact will grow exponentially. Frozen liquidation accumulates at compound interest; freezing is not free deferment but a loan with accruing interest.
4. The vote-source iron law is the foundational methodological discipline: four non-interchangeable source semantics in all empirics. Chapter 10's nine-country control experiment proves: tech5 increments are positive in all 9 countries (+0.47~+2.49pp), macro5 increment is negative in CN (−3.7pp)—vote source choice is decisive for conclusions. The live true-macro five-vote rule-based specification (§3.2-3.3) is finalized but its first recalculation is pending; backtests use the Shiller true-macro proxy; the caliber gap is bridged through forward dual-track validation.

框架的诚实边界同样清晰：n = 4 的插值性质、进行中的 2026、不预测时点、票源四义不可互换、前视溢价 +0.31pp/年、七国代理数据。它的价值不在点估计的精度，而在把"泡沫会不会破"的叙事争论，替换为"破多深、多久、用什么仓位、按什么规则迎接"的可计算结构。当周期变快，纪律的复利也变快。

The honest boundaries are equally explicit: n = 4 interpolation, 2026 in progress, no timing claims, four non-interchangeable vote-source semantics, +0.31pp/yr look-ahead premium, seven-country proxy data. The framework's value lies not in point-estimate precision but in replacing the narrative debate over "whether the bubble bursts" with a computable structure of "how deep, how long, with what positions, and by what rules". As cycles accelerate, so does the compounding of discipline.

附录A：复杂系统视角下的方法论基础 / Appendix A: Methodological Foundations from Complexity Science

A.1 问题的提出 / The Question

前面十三章建立了一个"故事→数字→风险预警→策略执行"的完整链条。但有一个更深层的问题尚未正面回答：凭什么三条简单的存量耗竭规律，能生成全球经济两百年的复杂周期？如果不回答这个问题，读者有理由怀疑三笔账是事后拼凑的叙事。本附录从复杂系统科学的视角论证：三笔框架不是经验拟合，而是对经济系统动力学结构的正确降维——三条存量轴找到了生成宏观复杂性的最小耦合集。

The preceding thirteen chapters establish a complete "story → numbers → risk warning → strategy execution" chain. Yet a deeper question remains: why should three simple stock-depletion rules generate two centuries of complex cycles? Without addressing this, readers may suspect the framework is a post-hoc narrative. This appendix argues from a complexity science perspective that the framework is not empirical fitting but a correct dimensional reduction.

A.2 三层复杂性与最小耦合集 / Three Layers of Complexity and the Minimal Coupling Set

经济系统里有数亿消费者、千万企业、数百政府，每个都在做局部最优决策（组分复杂）。传统宏观经济学处理复杂的方式是"加变量"——DSGE 模型动辄几百个方程。三笔框架走的是相反方向：降维，将结构复杂度压缩到三条正交的存量轴。三笔支取互相抵押、互相掩盖、互相传导——这种耦合反馈意味着系统状态空间是高度扭曲的非线性流形。回撤预测方程中的交互项就是对非线性的数学承认：

The economy contains hundreds of millions of agents, each making locally optimal decisions (agent complexity). Traditional macro handles complexity by "adding variables." The framework takes the opposite direction: dimensional reduction, compressing structural complexity into three orthogonal stock axes. The three drawdowns cross-collateralize, cross-mask, and cross-transmit—coupling feedback means the state space is a highly distorted nonlinear manifold. The interaction term in the drawdown equation is a mathematical acknowledgment:

ABCT 和 MMT 各自发现了一个"局部简洁定律"但都误将局部定律当作全局定律。三笔框架将两家统合为闭合系统：三条存量轴是系统的三个慢变量，在 Haken 协同学框架下决定系统的序参量，而快变量（利率、PMI、VIX）只是慢变量的瞬时投影。这是 Anderson "More is Different" 涌现原理在经济学中的具体实现。

ABCT and MMT each discovered a "local simplicity law" but mistook it for a global law. The framework unifies both: three stock axes are the system's slow variables, determining order parameters in Haken's Synergetics, while fast variables are instantaneous projections. This is Anderson's "More is Different" emergence principle realized in economics.

A.3 冻结系数作为涌现抑制器 / Freeze Coefficient as Emergence Suppressor

复杂系统科学中有一个普适规律：抑制小尺度振荡不会产生稳定，只会把能量积累到大尺度一次性释放。冻结系数做的事情完全一样：1929 年 φ ≈ 0，系统自然振荡，跌了 86% 但 34 个月见底；2021 年 φ ≈ 0.95，系统被锁定在非平衡态，只跌了 25%，但被冻结的清算以复利积累。

A universal principle in complexity science: suppressing small-scale oscillation accumulates energy for large-scale catastrophic release. The freeze coefficient does exactly this: in 1929 (φ ≈ 0), natural oscillation dropped 86% but bottomed in 34 months; in 2021 (φ ≈ 0.95), the system was locked in non-equilibrium, dropping only 25%, but frozen clearing accumulated at compound interest.

冻结系数把经济系统从"弛豫振荡"推入"积累-灾变"制式——形成期被拉长（能量积累），清算期被压缩（能量集中释放）。杠铃形周期不是政策失误，而是复杂系统在外部强制抑制下的必然动力学制式。

The freeze coefficient pushes the system from "relaxation oscillation" into an "accumulation-catastrophe" regime. The barbell cycle is not policy failure but a necessary dynamical regime of a complex system under external suppression.

A.4 认识论定位 / Epistemological Positioning

三笔框架的认识论立场：经济系统是复杂的，但不是不可约的。三条局限需指出：(1) 可能存在第四慢变量（AI 带来的 TFP 提升）；(2) 八组耦合拓扑是从历史归纳的，未来可能需要更新；(3) 经济系统独特之处在于"抑制力量"本身是内生变量——"对抑制的抑制"在自然界没有对应物。

The framework's epistemological stance: the economy is complex but not irreducible. Three limitations: (1) a potential fourth slow variable (AI-driven TFP); (2) coupling topology is historically induced and may need updating; (3) the economy's unique feature—humans can change rules, making the "suppressing force" endogenous.

附录B：AI辅助披露与方法论透明度声明 / Appendix B: AI Assistance Disclosure

B.1 AI辅助范围 / Scope of AI Assistance

本研究在以下环节使用了大语言模型（LLM）辅助：

1. 数学公式形式化：BII、RCS、S''' 的解析表达式、δ(S'') 清算破坏力函数的形式化推导过程中，AI 辅助了 LaTeX 公式排版和符号系统统一。公式的经济逻辑由作者设计。
2. 回归代码生成：Forward Stagewise 特征搜索、LOCO-CV 交叉验证、Bootstrap 重采样、滚动 OOS 验证的 Python 代码由 AI 辅助编写，经作者审查后执行。所有回归结果均由 Python 从实际危机数据库计算得出。
3. 文献检索与格式化：参考文献的格式统一由 AI 辅助完成。文献的选择由作者决定。
4. 语言润色：本文以中英双语撰写，AI 辅助了英文摘要的语法校正和术语统一。

B.2 作者独立完成的部分 / Author-Independent Work

以下工作完全由作者独立完成，未使用 AI 辅助：理论框架设计（三笔跨期存量支取的概念框架、ABCT 与 MMT 的统一逻辑、冻结系数的定义）；数据收集与构建（57 个危机事件 × 29 国数据库）；模型选择决策；结果解读与局限性评估；政策含义推导。

The following was completed entirely by the author without AI assistance: theoretical framework design; data collection and construction (57 crisis events × 29 countries database); model selection decisions; result interpretation and limitation assessment; policy implication derivation.

B.3 可复现性声明 / Reproducibility Statement

本研究的全部回归结果可通过以下方式复现：（1）危机事件数据库可向作者索取；（2）Forward Stagewise 和验证代码的 Python 实现可在附录 C 伪代码框架中找到；（3）所有报告的 ρ、MAE、R²、CI 数值均可通过运行上述代码在相同数据上精确复现。

All regression results can be reproduced: (1) the crisis event database is available from the author upon request; (2) the Python implementation is outlined in pseudocode; (3) all reported values can be exactly reproduced by running the code on the same data.

参考文献 / References

1. Mises, L. von. (1912). The Theory of Money and Credit.
2. Hayek, F. A. (1931). Prices and Production.
3. Rothbard, M. N. (1963). America's Great Depression.
4. Shiller, R. J. (2000). Irrational Exuberance. Princeton University Press.
5. Minsky, H. P. (1986). Stabilizing an Unstable Economy.
6. Reinhart, C. M., & Rogoff, K. S. (2009). This Time Is Different: Eight Centuries of Financial Folly. Princeton University Press.
7. Wray, L. R. (2015). Modern Money Theory. Palgrave Macmillan.
8. Kelton, S. (2020). The Deficit Myth. PublicAffairs.
9. Dalio, R. (2018). Principles for Navigating Big Debt Crises.
10. Hall, C. A. S., & Klitgaard, K. A. (2012). Energy and the Wealth of Nations. Springer.
11. Smil, V. (2017). Energy and Civilization: A History. MIT Press.
12. Cantillon, R. (1755). Essai sur la Nature du Commerce en Général.
13. Chow, G., & Lin, A. (1971). Best Linear Unbiased Interpolation, Distribution, and Extrapolation of Time Series by Related Series. Review of Economics and Statistics.
14. UN World Population Prospects 2024.
15. FAO FAOSTAT / AQUASTAT Databases.
16. EIA / IEA International Energy Statistics.
17. BIS Credit Statistics.
18. Ritter, J. IPO Database. University of Florida.
19. Anderson, P. W. (1972). More Is Different: Broken Symmetry and the Nature of the Hierarchical Structure of Science. Science, 177(4047), 393–396.
20. Mandelbrot, B. B. (1982). The Fractal Geometry of Nature. W. H. Freeman.
21. Haken, H. (1983). Synergetics: An Introduction (3rd ed.). Springer-Verlag.
22. Bak, P., Tang, C., & Wiesenfeld, K. (1987). Self-Organized Criticality: An Explanation of the 1/f Noise. Physical Review Letters, 59(4), 381–384.
23. Holling, C. S. (1973). Resilience and Stability of Ecological Systems. Annual Review of Ecology and Systematics, 4, 1–23.
24. Barabási, A.-L., & Albert, R. (1999). Emergence of Scaling in Random Networks. Science, 286(5439), 509–512.
25. Arthur, W. B. (2014). Complexity and the Economy. Oxford University Press.
26. Jordà, Ò., Schularick, M., & Taylor, A. M. (2015). Leveraged Bubbles. Journal of International Money and Finance, 56, S133–S152.
27. Kindleberger, C. P., & Aliber, R. Z. (2011). Manias, Panics, and Crashes (6th ed.). Palgrave Macmillan.
28. Hall, C. A. S., Lambert, J. G., & Balogh, S. B. (2014). EROI of Different Fuels and the Implications for Society. Energy Policy, 64, 141–152.
29. Borio, C., & Lowe, P. (2002). Asset Prices, Financial and Monetary Stability: Exploring the Nexus. BIS Working Papers, No. 114.
30. Campbell, J. Y., & Shiller, R. J. (1988). The Dividend-Price Ratio and Expectations of Future Dividends and Discount Factors. Review of Financial Studies, 1(3), 195–228.
31. Collin-Dufresne, P., Goldstein, R. S., & Martin, J. S. (2001). The Determinants of Credit Spread Changes. Journal of Finance, 56(6), 2177–2207.
32. De Bondt, W. F. M., & Thaler, R. H. (1985). Does the Stock Market Overreact? Journal of Finance, 40(3), 793–805.