How it works

What this model does, how the rankings are made, and what to trust.

What this model is

It's a risk model for US AI and semiconductor stocks and funds. It reads the whole economy first (growth, interest rates, credit, market fear), then asks those two questions for each name by simulating about 150,000 possible 7-year futures and counting what happens. A "probability" here is a frequency over those futures — like a rain forecast, not a promise about one outcome. It never gives a single price target.

Read the full methodology report →

The numbers you see, in plain English

What decides the comeback

A fall is set by the market. The recovery is set by the business — and this is the part the model now judges most carefully. It weighs five things:

• Room to grow — how much of its market is still open, and how fast that market is growing.
• Balance sheet — cash cushions a fall; debt deepens it and slows the climb back.
• Competition & moat — can it keep its lead and its pricing power over time?
• Earnings cyclicality — in a downturn, some companies' profits collapse far faster than their share price. That makes them look "cheap" right before earnings roll over — a trap the model flags.
• Profit sensitivity — how much higher interest rates (on debt) or input costs eat into profit.

Is the price "earned" or "floating"?

The model splits each stock's price into three parts: the base (what an ordinary business is worth at today's interest rates), the earned premium (justified by real growth, margins, market size and moat), and the floating premium (the speculative part resting on hope). Only the floating part is treated as fragile. This is not a price target — it shows how much of today's price stands on substance vs sentiment.

How a ranking is made

The result is a living league table: the best ~125 names by reward-for-risk, with the Top 50 shown by default. The model re-screens the whole AI/semis market and a name only enters when it genuinely out-ranks one already on the list — so the board stays stable unless something real changes. You can click any column heading to re-sort the view (a "default rank" badge marks the model's own order, with a one-tap reset).

The verdicts

The market-stress reading

The dial at the top of the dashboard is a 0–100 score for how dangerous the overall market is right now — not any one stock. It has two parts:

• Structural — slow-building pressure (how expensive stocks are, debt levels). The "dry powder."
• Acute — immediate stress (market fear, credit spreads). The "lit match."

Higher stress makes the model assume bigger, more frequent drops — so it raises everyone's crash odds.

ETF horizons & the stress ladder

Below the main table the dashboard has two more tools:

• ETF horizons — the best fund depends on how long you hold. A basket that heals in two years ranks low for a 1-year hold and high for a 3–5 year hold. Switch the horizon to watch the order change.
• Stress ladder — how the chance of a big fall widens as the overall market goes from benign to critical, with today's reading highlighted.

Where the data comes from & how often it updates

• Every day: live market data — interest rates, market fear (VIX), credit spreads, oil — from the U.S. Federal Reserve (FRED). The model re-runs and re-ranks, and the date at the top updates.
• Every week: an AI (Claude) re-evaluates each company's outlook — growth, business strength, valuation — grounded in fresh financials pulled from Polygon. See exactly how below.

So day-to-day you'll see the stress reading, prices and odds move; the deeper rankings drift slowly. That's intended — a risk model shouldn't flip its whole list every day.

How the weekly AI review decides

Once a week the model pulls each company's latest revenue and price from Polygon, then asks Claude to re-score that company's outlook on a fixed scale. It's told to make small, well-reasoned changes only — keep a number unless fresh data or a clear reason justifies moving it — so the rankings stay stable. Every score it touches comes with a one-line written reason, shown below.

These are the exact inputs the AI re-scores each week:

Your portfolio's risk

Add your holdings on the Portfolio page and the model reads your whole basket, not just each stock alone. The key number is effective bets: because these AI names move together (~91% in a crash), holding five of them isn't five bets — it can be close to one. So the page also shows the chance most of your basket drops 25%+ at the same time, plus your weighted growth, recovery and long-term-loss odds.

The takeaway it keeps repeating: spreading money across these names barely lowers a big drop — real diversification means owning things outside the sector. Beginners see plain cards; Expert mode adds the exact math and a per-holding breakdown.

Options

The Options page opens on Top picks — a ranked list of specific call contracts showing the live market price next to the model's fair value, a buy / watch / skip read, and the best spread (the legs) for each name. Switch to Look up a stock to search any company and see its single calls or spreads with a plain-English take.

Calls vs spreads (legs): a call bets the stock rises by a deadline (cheaper, but can go to zero). A spread buys one call and sells a higher one — the two strikes are the "legs"; it costs less but caps the gain at the higher strike. The board covers the whole AI/semis league's optionable names and re-prices daily.

These are thesis-conditional values — what the move is worth if the model is right — and the model is blind to implied volatility unless you connect a live broker feed. "Passes the gates" is a conviction test, not a green light. Calls and call-spreads only — no puts.

Tap any Top-picks row for a fuller pop-up — simple or detailed depending on your Beginner/Expert toggle — with the contract, the chance it pays, and the suggested spread. Two live buttons sit there too: live check compares the model's value to what the market is actually charging in volatility, and spread cost measures how much the bid-ask would eat when you enter both legs (🟢 cheap, 🟡 meaningful, 🔴 costly — at which point a plain ETF is often the smarter way to express the same view).

Honest limits

• The backtest (7 historical crashes) checks how deep falls go — not the recovery timing. So the depth/permanent-loss side is the most validated; recovery speed and how far earnings collapse are the least certain parts.
• The company inputs (growth, balance sheet, moat) are informed judgments — "moat" is the single most sensitive one, and if it's wrong the output is wrong.
• This whole group of AI stocks tends to move together, so spreading across them is less diversification than it looks — the model measures this directly.
• It can't predict surprises — a single-company blow-up or a black-swan event.