Find the leading theme, hold it while trend is healthy, and raise cash when risk breaks. Discipline, not prediction.寻找主导主题，趋势健康时持有，风险破坏时提高现金。这是纪律，不是预测。
Read解读: AI Compute & Optics still ranks first by trailing momentum, but is breaking down now (few members holding up); leadership is narrowing (6/22 themes in uptrend); suggested cash 0%.AI算力与光模块 按趋势动量仍居首位，但近期正在走弱（成分股多数转弱）；领跑面 收窄（6/22 个主题处于上行趋势）；建议现金 0%。
Equal-weight the top-4 themes above their own 200d trend (the validated dual-momentum book), DE-OVERLAPPED to at most 1 per parent super-theme so t...
Equal-weight by design. The rank tells where to look; trend decides whether to hold. Crowding is shown as caution only; it is not a weight cut yet.刻意等权。排名说明看哪里，趋势决定是否持有。拥挤度仅作提示；尚不削减权重。
🧮 De-overlapped for diversification为分散去重叠 — keeps the book from becoming one trade in several labels. Displaced避免组合变成同一笔交易的多个标签。被替换: SemiconductorsConsumer Electronics
Optional de-risk overlay — scale the book toward a ~18% vol target. Backtested on 27y of clean sector history it cut max-drawdown by ~7.7pp (CI [1....可选降风险叠加 — 将组合缩放至约 18% 波动率目标。在27年干净行业历史回测中，最大回撤降低约 7.7pp（区间 [1.0, 7.7, 17.4]），并优于纯趋势刹车；但...
🔬 Measured on the 27y US clean-sector proxy基于27年美国干净行业代理实测: max-drawdown reduction最大回撤降低 7.7pp CI区间 [1.0, 17.4], beats a trend-only brake优于纯趋势刹车: ✓, DSR 0.7705 (80 configs配置). Grade评级: display_only — it cuts drawdown but does not raise Sharpe enough to clear the multiple-testing bar, so it stays a suggestion.降低回撤但夏普提升不足以通过多重检验门槛，故仅作为建议。
What to add, reduce or avoid. Click a theme for its stocks.该加什么、减什么、避开什么。点击主题查看个股。
Suggested cash over time. Higher cash means more defensive.随时间显示建议现金。现金越高越防守。
Book changes, leadership rotation, trend flips and risk-dial swings.组合变化、领涨轮换、趋势翻转与风险刻度变化。
Themes ranked by momentum, trend gate, durability and crowding.主题按动量、趋势门、持续性与拥挤度排序。 The rank is deliberately slow — “Past 2 weeks” shows what each theme just did.排名有意放慢 — 「近两周」显示各主题最近的实际表现。 Click a theme name to open its constituents on the Thematic Baskets page.点击主题名称可在主题篮子页面查看其成分。
| Theme主题 | Score评分 | Past 2 weeks近两周 | Trend趋势 | Durability持续性 | Hurst赫斯特 | Breadth广度 | Crowding拥挤度 |
|---|---|---|---|---|---|---|---|
| AI Compute & OpticsAI算力与光模块↗
LEADER领跑
Technology & AI · 17 names只 |
1.92 | -9.1% | ▲ in trend趋势内 | 58 | 0.59 ↗ | 24% | -0.45σ |
| Semiconductors半导体↗
Technology & AI · 22 names只 |
1.65 | -18.7% | ▲ in trend趋势内 | 63 | 0.648 ↗ | 18% | +0.3σ |
| Innovative Pharma & CXO创新药与CXO↗
Healthcare · 14 names只 |
0.87 | -1.0% | ▲ in trend趋势内 | 91 | 0.653 ↗ | 100% | +1.38σ hot热 |
| Consumer Electronics消费电子↗
Technology & AI · 15 names只 |
0.42 | -16.8% | ▲ in trend趋势内 | 60 | 0.621 ↗ | 20% | -0.58σ |
| Coal煤炭↗
Cyclicals & Resources · 10 names只 |
0.34 | +5.3% | ▲ in trend趋势内 | 55 | 0.399 ↺ | 50% | +0.94σ |
| Brokers & Securities券商↗
Financials & Value · 15 names只 |
0.30 | -4.2% | ▽ off trend趋势外 | 54 | 0.622 ↗ | 87% | +0.42σ |
| Robotics & Automation机器人与自动化↗
Advanced Manufacturing · 11 names只 |
0.27 | -20.9% | ▲ in trend趋势内 | 50 | 0.52 ~ | 9% | -0.02σ |
| Insurers保险↗
Financials & Value · 5 names只 |
0.27 | +5.1% | ▽ off trend趋势外 | 61 | 0.635 ↗ | 100% | +0.31σ |
| Home Appliances家电↗
Consumer & Brands · 8 names只 |
0.15 | +2.5% | ▽ off trend趋势外 | 48 | 0.575 ↗ | 62% | +0.53σ |
| Banks银行↗
Financials & Value · 18 names只 |
0.07 | +6.7% | ▽ off trend趋势外 | 66 | 0.401 ↺ | 100% | +0.57σ |
| Food & Beverage食品饮料↗
Consumer & Brands · 10 names只 |
-0.01 | +2.6% | ▽ off trend趋势外 | 45 | 0.522 ~ | 70% | -0.17σ |
| Medical Devices & TCM医疗器械与中药↗
Healthcare · 14 names只 |
-0.04 | +7.1% | ▽ off trend趋势外 | 63 | 0.77 ↗ | 93% | +0.92σ |
| SOE Blue Chips (中特估)中特估·央企↗
Financials & Value · 16 names只 |
-0.12 | +8.1% | ▽ off trend趋势外 | 58 | 0.681 ↗ | 69% | -0.43σ |
| Industrial Metals有色金属↗
Cyclicals & Resources · 13 names只 |
-0.12 | -2.2% | ▽ off trend趋势外 | 34 | 0.621 ↗ | 23% | -0.43σ |
| Baijiu / Liquor白酒↗
Consumer & Brands · 7 names只 |
-0.35 | +9.9% | ▽ off trend趋势外 | 61 | 0.693 ↗ | 71% | -0.06σ |
| Rare Earth & Magnets稀土永磁↗
Cyclicals & Resources · 9 names只 |
-0.36 | -25.2% | ▽ off trend趋势外 | 16 | 0.521 ~ | 0% | -0.4σ |
| Software & AI Apps软件与AI应用↗
Technology & AI · 15 names只 |
-0.45 | +0.4% | ▽ off trend趋势外 | 29 | 0.554 ↗ | 13% | -0.65σ |
| Defense & Aerospace军工航天↗
Advanced Manufacturing · 13 names只 |
-0.65 | -14.1% | ▽ off trend趋势外 | 26 | 0.57 ↗ | 8% | -1.94σ |
| Autos & NEV Makers汽车整车↗
New Energy & Autos · 16 names只 |
-0.92 | -4.4% | ▽ off trend趋势外 | 42 | 0.671 ↗ | 31% | -1.09σ |
| Gold Miners黄金↗
Cyclicals & Resources · 6 names只 |
-0.96 | -4.0% | ▽ off trend趋势外 | 38 | 0.688 ↗ | 17% | -0.46σ |
| Solar / Photovoltaics光伏↗
New Energy & Autos · 15 names只 |
-1.13 | -14.2% | ▽ off trend趋势外 | 19 | 0.535 ~ | 0% | -1.88σ |
| Battery & Lithium锂电池↗
New Energy & Autos · 16 names只 |
-1.16 | -17.1% | ▽ off trend趋势外 | 37 | 0.664 ↗ | 6% | -1.25σ |
Current leader, challenger and breadth. Handoffs are flagged only after confirmation.当前领跑、挑战者与广度。只有确认后才提示交接。
Broad leadership is healthier. Narrow leadership raises the cash dial.领跑越广越健康；领跑越窄，现金刻度越高。
Theme baskets are short. The algorithm is tested on long sector history below.主题篮子历史较短。算法在下方长期行业历史上检验。 (2017-12-31 → 2026-07-31)
| Strategy策略 | CAGR | Sharpe | Max DD最大回撤 | Crash months崩盘月 | DSR |
|---|---|---|---|---|---|
| Dual-momentum book (this engine)双动量组合（本引擎） | -2.7% | 0.0 | -60.7% | -4.7% | 0.059 |
| Relative-momentum only仅相对动量 | 1.1% | 0.18 | -53.8% | -6.6% | 0.139 |
| Equal-weight buy & hold等权买入持有 | 3.0% | 0.24 | -37.3% | -7.0% | 0.198 |
The validated edge is risk control, not alpha.验证过的优势是风险控制，而非阿尔法。 Above vs below the trend gate, forward return is statistically identical, but Sharpe is在趋势门上方 vs 下方，前瞻收益在统计上一致，但夏普为 0.13 vs对 0.19 and the bad-month tail is且坏月尾部为 -12.0% vs对 -12.5%. Rank shows where to look; trend decides whether to be in.排名说明看哪里；趋势决定是否在场。
The AI may explain the setup, but it may not invent edge beyond the tested rules.AI 可以解释设置，但不能在已测试规则之外编造优势。
Verdict结论: discipline_not_prediction
This ranks where thematic LEADERSHIP and FLOW are concentrating and suggests a trend-following, vol-aware allocation. The ONE validated edge (multi-decade sector backtests) is D...
Directive指令: Permitted: name the prevailing narrative, judge its durability from breadth/cohesion/trend-health, flag crowding as a sizing caution, and describe the one-narrative-vs-everythin...
Desk leans are dated hypotheses, not a buy list or sizing rule.桌面倾向是带日期的假设，不是买入清单或仓位规则。