We track 96 stocks across the 10 layers of the AI supply chain — from uranium miners to foundation model companies. Every trading day, our scanner detects technical breakout patterns, ranks relative strength, and tracks each signal through its full lifecycle — from entry to exit.
No predictions. No opinions. Just structured signals backed by price data.
What We Do
Breakout Scanner — We run 5 proven technical setups against every stock in our universe, every day:
- VCP (Volatility Contraction Pattern) — Tight consolidation after a strong prior move. Compression before expansion.
- Episodic Pivot — Gap-up on heavy volume that holds for 3+ days. Institutional accumulation in action.
- Power Breakout — New 52-week highs with elite relative strength. The strongest stocks making new highs.
- O'Neil (Cup-with-Handle / Flat Base) — Classic growth patterns: 25+ day consolidation near highs with proper moving average alignment.
- Base Recovery — Stage 1-to-2 transitions. Stocks recovering 20%+ off lows with SMA structure turning bullish.
Each setup is scored 0-100 based on prior gains, pullback depth, tightness, volume, moving average alignment, and relative strength rank. Only signals scoring 40+ make the cut.
Signal Lifecycle — We don't just flag entries. We track every breakout through four stages:
| Signal | Meaning | Trigger |
|---|---|---|
| BREAKOUT | Entry detected | Pattern fires, score >= 40 |
| STRONG | Follow-through confirmed | Score improved, price above entry and 21 SMA |
| TRIM | Take profits warning | Price drops below 10 SMA after 5%+ gain |
| EXIT | Hard stop | Price below 50 SMA (-2% buffer) or -7% from entry |
Every trade has a clear entry, clear exit, and a recorded P&L. No ambiguity.
How We Measure Alpha
Most stock-picking services show you wins and hide the losses. We measure everything.
The problem with semiconductor sources: Everyone covering AI chips is bullish on the sector. That's not edge — the sector goes up. The real question is: which specific stocks outperform the sector?
We measure two things:
-
Stock-picking alpha (vs SMH) — Did the pick beat the semiconductor ETF? If you called NVDA and NVDA went up 5% but SMH went up 8%, that's not a good pick. You would have been better off buying the ETF.
-
Market alpha (vs QQQ) — Did the pick beat the broader tech market? This separates genuine insight from sector momentum.
Edge Score — For every signal source, we calculate:
Edge = (hit_rate x avg_win) - (miss_rate x avg_loss)
Positive edge means the source picks winners within semis more than losers, adjusted for magnitude. This is measured at multiple horizons (5, 10, 20, 40, 60 days) to match different thesis types — catalyst calls are evaluated at 5-10 days, trend calls at 20-60 days.
Signal Track Record
Our agents don't just detect signals — they catch major moves early. Here are selected breakouts our scanner identified before the crowd:
| Ticker | Entry Date | Setup | Entry Price | Current | Return |
|---|---|---|---|---|---|
| SNDK | Sep 16 '25 | Episodic Pivot | $91.55 | $649.97 | +610% |
| LITE | Aug 25 '25 | O'Neil | $123.42 | $667.77 | +441% |
| WDC | Aug 25 '25 | O'Neil | $79.22 | $285.52 | +260% |
| MU | Aug 28 '25 | VCP | $122.00 | $428.17 | +251% |
| SK Hynix | Sep 8 '25 | VCP | 276,802 KRW | 949,000 KRW | +243% |
| CIEN | Aug 28 '25 | O'Neil | $97.89 | $334.95 | +242% |
| BE | Aug 27 '25 | O'Neil | $50.85 | $147.55 | +190% |
| TER | Aug 25 '25 | VCP | $117.07 | $324.85 | +178% |
| Samsung | Aug 25 '25 | Power Breakout | 70,838 KRW | 190,100 KRW | +168% |
| COHR | Sep 5 '25 | Base Recovery | $97.84 | $248.18 | +154% |
| STX | Aug 25 '25 | O'Neil | $164.00 | $411.11 | +151% |
| LRCX | Aug 25 '25 | O'Neil | $101.28 | $244.92 | +142% |
| ASML | Aug 25 '25 | Base Recovery | $754.46 | $1,469.59 | +95% |
| CAMT | Aug 28 '25 | Base Recovery | $87.82 | $148.10 | +69% |
| PSTG | Aug 26 '25 | Base Recovery | $58.56 | $73.93 | +26% |
| TSM | Dec 29 '25 | O'Neil | $293.28 | $370.54 | +26% |
SNDK (SanDisk) is a standout case. Our scanner flagged it as an Episodic Pivot on September 16, 2025 — right after the Western Digital spin-off — at $91.55. The stock is now above $649. Our Discovery Engine simultaneously picked up multiple high-edge KOLs identifying SNDK as a pure-play HBF (High Bandwidth Flash) opportunity before the broader market recognized the thesis.
MU (Micron) was caught as a VCP breakout on August 28, 2025 at $122. Memory was entering a super-cycle driven by HBM demand for AI training, and our relative strength ranking had MU in the top 5 of the entire universe. The stock confirmed with a STRONG signal within a week and never looked back.
SK Hynix (000660.KS) triggered a VCP on our Korea exchange coverage — something most US-only scanners would miss entirely. The HBM leader broke out at 276,802 KRW and is now above 949,000 KRW (+243%). Our multi-exchange coverage caught the move that US investors couldn't see.
Not every signal wins. We show our exits too — SMCI (-32%), ORCL (-37%), NOW (-45%). The difference: losers get stopped out at -7% or at the 50 SMA. Winners run. That asymmetry is the edge.
Returns shown are from signal entry date to February 20, 2026. Past performance does not guarantee future results.
Portfolio Management — Two Strategies
We don't just find breakouts — we manage them through their full lifecycle using two independent portfolio strategies, each with a distinct philosophy:
V5.3 (Milestone System) — A flat portfolio structure that treats every position equally at entry, then scales aggressively into winners. Small probe entries (1.5-3% of portfolio) grow into full positions through milestone adds at +8%, +20%, +40%, and +80% gain. Winners also receive pullback adds when they bounce off the 21-day moving average, and momentum-doubled adds when price overshoots milestones by 3%+. Losers are culled through time stops (15-day trim, 30-day exit) and hard stops (-7% probe, -10% confirmed, -20% trailing from high-water mark). The V5.3 breakthrough: filtering re-entries by exit reason — stocks shaken out by probe stops are worth re-entering, but stocks that exited on trailing stops or exit signals (true trend reversals) are not.
4-Pool (Premier League System) — A tiered structure inspired by football league promotion and relegation. Every stock enters as a Newcomer (P4) or Developer (P3) and must earn its way up through a composite score (40% profit, 30% 20-day momentum, 20% SMA50 health, 10% confirmation status). Top performers advance to Contender (P2) and eventually Star (P1), receiving progressively larger capital allocations — P1 gets 66% of capital for just 8 slots. When pools are full, the system forces swaps — strong candidates replace the weakest incumbents, ensuring the best stocks always get the most resources. Six milestone add levels (+8% through +180%) with aggressive per-milestone sizing (3/7/7/6/7/6% of portfolio) compound winners through their entire run.
Backtested Performance (Jan 2022 — Feb 2026, $100K starting capital):
| Metric | V5.3 | 4-Pool |
|---|---|---|
| Final Value | $950,748 | $1,442,808 |
| Total Return | +850.7% | +1,342.8% |
| CAGR | 73.1% | 91.6% |
| Sharpe Ratio | 2.12 | 2.13 |
| Max Drawdown | -24.7% | -30.4% |
| Win Rate | 39% | 34% |
Both strategies run simultaneously on live signals. V5.3 offers better risk control (lower drawdown, higher win rate). The 4-Pool system captures more upside through aggressive tiered allocation to proven winners. Toggle between them on the Portfolio page.
The AI Supply Chain Universe
We cover 96 stocks across 10 layers of the AI infrastructure stack — the most comprehensive AI supply chain coverage available:
| Layer | What It Covers | Examples |
|---|---|---|
| Raw Materials | Uranium, lithium, copper, rare earths | CCJ, MP, FCX, UEC, UUUU |
| Semiconductor Equipment | Lithography, deposition, inspection, testing | ASML, LRCX, KLAC, AMAT, TER, CAMT |
| Foundries | Chip manufacturing, wafer fabrication | TSM, UMC, GFS, INTC |
| Memory & Storage | DRAM, NAND, HBM, HBF, SSDs | MU, SK Hynix, Samsung, WDC, SNDK, PSTG |
| Processors | GPUs, CPUs, custom silicon, IP | NVDA, AMD, ARM, AVGO, QCOM, MRVL, CRDO |
| Networking | Optical, switches, data center interconnects | ANET, CIEN, COHR, LITE, CSCO |
| Energy Infrastructure | Nuclear, solar, batteries, power management | CEG, BE, FLNC, ETN, VST |
| Data Centers | REITs, cloud infra, bitcoin mining | DLR, EQIX, AMZN, CIFR, IREN, WULF |
| Software & Models | Enterprise AI, SaaS, cloud platforms | PLTR, NOW, CRM, ADBE, MSFT, GOOGL, META |
| LLM | Foundation model companies | Chinese AI labs, frontier model builders |
Our coverage spans US exchanges (NYSE/NASDAQ), Hong Kong (HKEX), and South Korea (KRX) — giving us visibility into the global AI supply chain that US-only trackers miss.
Every stock gets a daily Mansfield Relative Strength score (252-day SMA-based, benchmarked against QQQ), ranked against the full universe. The heatmap on the homepage shows the entire supply chain at a glance — green means outperforming, red means underperforming.
Research Intelligence
Discovery Engine — We monitor 100+ X/Twitter KOLs (Key Opinion Leaders) across semiconductors, AI infrastructure, energy, and data centers using Grok AI — the only LLM with native X Search access. Every day, our agents scan posts, long-form articles, and Substack newsletters to extract structured signals:
- Every signal gets a direction (bullish/bearish), signal type (inflection, early trend, catalyst, confirmation), and conviction level
- After 5-60 trading days, we measure actual performance vs SMH and QQQ — no hiding
- Sources are scored weekly on a rolling 90-day window using our Edge Score formula
- Only sources with 5+ measured signals get scored — no small-sample noise
- Low-edge sources get automatically deprioritized; high-edge sources get amplified
The result: we know which KOLs actually generate alpha, and which ones are just loud. Follow count means nothing — P&L means everything.
Research Memos — Deep-dive analyses on specific stocks and themes. Each memo includes a conviction rating (high/medium/low), a specific thesis, and supporting evidence. Available in English and Chinese.
Flash Notes — Real-time alerts for breakout signals, RS score changes, and new stock discoveries. Color-coded by severity (critical/warning/info).
How Signals Produce Returns
The core thesis is simple: buy stocks with strong relative strength that are breaking out of sound technical bases, and sell when the structure breaks.
Here's why this works in the AI supply chain specifically:
-
Concentrated universe — 96 stocks, not 5,000. We know every name, every layer, every supply chain relationship. When one layer moves, we catch the ripple.
-
Relative strength as a filter — RS ranking eliminates the bottom half of the universe before we even look at patterns. The strongest stocks in the strongest sector — that's where institutional money flows.
-
Defined risk on every trade — Every breakout has a hard stop: -7% from entry or close below 50 SMA. No "diamond hands" — if the thesis breaks, we're out. Average loss is bounded; average win is unbounded.
-
Lifecycle management — STRONG signals confirm follow-through (let winners run). TRIM signals protect gains (price weakness after profit). EXIT signals enforce discipline (cut losses short). The system manages the position — not emotions.
-
Multi-source validation — When a technical breakout aligns with a high-edge X source calling the same name? That's confluence. We score sources by actual alpha, not follower count.
-
Sector expertise — AI infrastructure has secular tailwinds (CapEx supercycle, inference scaling, memory bandwidth). Within that, winners and losers diverge sharply. Our job is identifying which companies capture the most value at each supply chain layer.
What We're Not
- Not a trading bot — We generate signals, not trades. You make the decisions.
- Not financial advice — We're a research tool. Past signals don't guarantee future returns.
- Not a sentiment tracker — We measure alpha, not vibes. A stock can have perfect sentiment and still underperform the sector.
- Not a black box — Every signal has a quality score breakdown. Every source has a measurable edge score. Every trade has an entry, exit, and P&L. Full transparency.
The Stack
Built for speed and reliability:
- Real market data from Polygon.io — no simulated prices, no delayed feeds
- Daily automated scans after market close (every weekday)
- Grok AI for X/Twitter signal extraction — the only LLM with native X Search access
- Next.js 16 frontend with server-side rendering — fast, SEO-friendly, bilingual (EN/ZH)
- PostgreSQL for durable storage — every bar, every signal, every outcome recorded
- Public API at
/api/claw/breakouts— integrate our signals into your own tools
CrazyRich Agents provides AI-generated research for informational purposes only. This is not investment advice. Always do your own research and consult a qualified financial advisor before making investment decisions.