How the research is made.

In one paragraph

Code ranks the liquid US market by trend every trading day and groups stocks that move together into themes. A frontier AI model then reads the news, company releases, filings and analyst actions for each stock we cover and writes a dossier: the thesis, the bull and bear case, the catalysts ahead, a conviction level and a kill line — the price that would prove it wrong. Medium-or-higher conviction calls become picks, graded against the S&P 500. Every trading morning Claude decides the public model portfolio from that research and from two fixed rules' proposals, and code places the orders within fixed limits.

The steps

1. Scan. Code ranks 3,488 US stocks by three-month and one-month trend and by how fast the trend is accelerating. This is the momentum board.
2. Group. Claude Opus groups the strongest stocks, and the ones we hold, into themes from their news and filings, and tags each theme accelerating, maturing or saturated. A weekly scan of stocks whose prices move together, the radar, looks for new ones. See themes in motion.
3. Read and write. For each covered stock, Claude Opus reads the last two weeks of news and the price history, searches the web for company releases, SEC filings, analyst actions and earnings-call coverage, and writes or updates its dossier.
4. Commit. Every dossier ends with a conviction level and a kill line. A medium-or-higher call opens a pick on the scoreboard.
5. Invest. Every trading morning Claude decides what the portfolio holds, from the research and from two fixed rules' proposals; code places the orders within fixed limits (how the portfolio works).

Two different universes appear on this site: the momentum scan covers 3,511 names, and the market-breadth reading on macro is computed over the 1,000 most liquid US stocks by dollar volume (978 with enough history for a 50-day average).

The schedule

• Every trading day — a fresh read of the stocks we hold or just sold, every open pick, every stock in an accelerating theme, and any covered stock with an earnings report or other dated event around that day or a move of 7% or more on the last close; the market regime from that close; the themes regrouped; the portfolio decision in the morning and its orders before the close; after the close, the daily journal and the macro view.
• On a rolling cycle — a fresh read of the rest of the coverage, including the largest US companies. 147 stocks are re-read every trading day: every stock in the portfolio or with an open pick, plus those in accelerating themes, reporting soon or moving sharply. The other 78 are re-read on a rolling cycle, and none is more than 6 sessions old.
• Every week — the theme radar scan; Claude's forecast of every stock in the portfolio's opportunity set and its web research on the best new ideas; the sector and theme cards; the stop rules re-checked.
• Twice every weekday, once a day at the weekend — the site is rebuilt in the morning, hours before the US market opens, with the research written overnight, and again every evening after the close, with the day's results. Each edition is checked automatically for stale facts and broken links before it goes live.

Schedules follow the New York market calendar, so holidays and daylight-saving shifts are handled.

What's in a dossier

• Thesis — the call and the reason for it, in a paragraph.
• Kill line — the condition that would prove it wrong, set in advance.
• Bull case and bear case — the strongest arguments on each side, each with its source.
• Price setup — where the price stands: trend, key levels, momentum.
• Catalysts — dated events over the next 30 days.
• What would change the view — the evidence that would raise or cut conviction.

Conviction levels

Four levels, from strongest to weakest: supreme, high, medium and low. Conviction says how strongly the model backs the thesis, not how far the price will go. A medium, high or supreme call is a pick and goes on the scoreboard. A low call is research coverage: tracked and kept current, but not graded.

Dossiers are re-read regularly, and conviction moves on new facts: results, guidance, a filing, analyst or industry data, company or regulatory news, or the thesis's kill condition being met. Since 10 October 2026, a change that would open or close a pick without such a fact counts only when the next read, after another trading day, reaches the same level, so one day's price swing does not open or close a pick.

How picks are graded

One rule, fixed on 1 October 2026 before any pick under it had an outcome, and applied to every pick:

1. Open. A pick opens at the closing price on the day a dossier first states the thesis at medium conviction or higher, as long as that price is still above the thesis's own kill line. The conviction, the kill line and the thesis text are frozen at that moment.
2. Close. It closes at the first of three things: a close through the frozen kill line (daily or weekly, as the line was written); conviction cut below medium; or 63 trading days, about three months.
3. Grade. It counts as a win if its total return over that window beat the S&P 500 (SPY) over exactly the same days, and a loss otherwise. Prices are adjusted for splits and dividends on both sides.

There is no discretion and no exception: every pick that opens is graded, and every graded pick stays on the scoreboard. The full record is machine-readable at /track-record.json. More detail, with examples, on how a pick is graded.

How the portfolio works

The portfolio is a model portfolio run at real market prices, from 5 October 2026. Each trading morning Claude Fable, Anthropic's most capable model, decides what it holds: it reviews every holding and every proposal from two fixed rules, with the research on file, the latest filings and news, and the weekly forecasts and web research Claude Opus prepares on the market. It accepts, resizes or rejects each proposal and adds companies of its own, including ones the rules cannot see, such as foreign or pre-revenue companies. Code places the orders before the close and enforces the limits: no borrowing, no short sales, no new money into a stock that is already a quarter of the portfolio, and no stock sold on the day it was bought.

The two rules were chosen after testing them on US stocks from 2016 to 2026, including most of the companies that later went bankrupt or were bought:

• Growth inflections — about a fifth of the rules' portfolio. Every trading day the system reads each quarterly and annual report filed with the SEC the day before. A company qualifies when its revenue for the quarter grew at least 25% on the year and faster than in the quarter before, while its shares trade within 15% of their 52-week high and their six-month gain ranks in the top fifth of the market. The rule buys every qualifying company the next trading day, in equal amounts, and holds it for six months. Each new report that qualifies again renews the six months; the position is sold six months after the last qualifying report.
• Momentum leaders — the rest. At the end of each month: the ten US stocks with the strongest twelve-month gain among the thirty with the most trading, each company counted once, in equal amounts, re-chosen the next month.

The research behind it found that, across the broad market, strong prices alone did not beat the S&P 500, but strong prices backed by accelerating revenue in a filed report did, and that holding every such company beat picking a few of them. The rules keep running alongside, and every day Claude's decisions and the rules are valued the same way, at the closing prices with the same costs, so the comparison measures the decisions rather than the timing of trades. Every forecast Claude makes is scored against what happened; the stop rules page says when the owner reviews Claude's decisions. There are no stop-losses and no index funds. We publish each holding's entry date and percentage return, closed trades, and the portfolio against the S&P 500 — never share counts or dollar amounts.

Archetypes

Each stock gets an archetype: a profile of how it tends to behave, which shapes how its setup is read. Full definitions on the archetypes page.

• Theme leader. The stock a theme is traded through.
• Picks and shovels. Sells what a theme consumes.
• Second-order beneficiary. Supplies what AI demand uses downstream: power, cooling, the grid.
• Business pivot. An established company re-aimed at a new source of demand.
• Event-driven. A dated event decides it: a readout, a deal, a ruling.
• Crowd-driven. Moved more by trading flows than by the business.
• Distinct case. Fits none of the others and is read on its own terms.

Market regime

Every trading day fixed rules classify the market backdrop — for example risk-on, choppy or risk-off — from the S&P 500's trend against its 200-day average, market breadth and volatility. The regime is stated on every journal entry. It informs how the research reads a setup; it does not size positions. See regimes.

Known limits

• Facts can lag. A dossier is as current as its last update date, which every page shows.
• Setups change fast. A price pattern can fail within a session; the kill line is how a call that stops working gets closed.
• Themes can be misread. A stock can be grouped with peers whose price driver is different.

None of this is a recommendation to buy or sell. The research shows its reasoning so you can judge it yourself.

Common questions

What is FrontierPicks?

FrontierPicks is free stock research written by a frontier AI model. It publishes a dossier on every stock it covers, a daily market journal, a public model portfolio, and a scoreboard that grades every pick against the S&P 500.

Which AI models write the research?

Anthropic's Claude Opus writes every published dossier, the daily journal and the macro view, and groups stocks into themes; its conviction calls are the picks on the scoreboard. Claude Fable, Anthropic's most capable model, decides the portfolio's trades. We always run the latest version of each.

What is a kill line?

The price, stated in advance, that would prove a thesis wrong — usually a daily or weekly close below a specific level. When a pick's kill line breaks, the pick closes and is graded.

How often is the site updated?

Every US trading day, with the research through that day's close. 147 stocks are re-read every trading day: every stock in the portfolio or with an open pick, plus those in accelerating themes, reporting soon or moving sharply. The other 78 are re-read on a rolling cycle, and none is more than 6 sessions old. Every page shows when it was last updated.

Is this investment advice?

No. We publish educational research under the BaFin and EU regulatory framework. No personal advice is given and no orders are accepted.