Why trust it
Built to be checked, not believed.
AI research is only as useful as your ability to check it. So every call here comes with a date, its sources and the price that would prove it wrong, the portfolio is public, and the scoreboard grades every pick against the S&P 500.
It says what would prove it wrong
Every thesis comes with a kill line: the price, set in advance, at which the call is wrong. It is published before the market decides and frozen when a pick opens, so it cannot be moved after the fact.
See a kill line →Every pick is graded against the market
A pick counts as a win only if it beats the S&P 500 over the same days. Every pick that opens is graded, and every graded pick stays on the scoreboard.
How a pick is graded →Both sides of the argument
The bear case is written with the same care as the bull case, with sources and dates, so you can weigh the whole argument rather than just the conclusion.
How a dossier is built →Everything is dated
Every dossier, journal entry and pick carries its date. You always know how current a read is.
Known limits →Nothing to sell you
No subscription, no newsletter, no affiliate links, and the portfolio takes no outside money. There is no product riding on a call.
The portfolio →Research, not advice
We publish educational research under the BaFin and EU framework. No personal advice, no orders accepted. The reasoning is ours; the decision is yours.
What a model does well, and what it doesn't
A frontier model is excellent at some parts of research and no help with others. Here is where the line falls.
What the model does well
- Reading the news, filings and analyst actions for every stock it covers, on a rolling cycle, without tiring.
- Holding a whole theme in view at once and comparing every name in it side by side.
- Applying the same framework to every stock, with no ego, no fear of missing out and no attachment to yesterday's call.
- Writing the case against as carefully as the case for.
What it cannot do
- Knowing anything non-public. It reads the same public record you do.
- Guaranteeing an outcome. Markets can turn in a session; that is what the kill line is for.
- Catching every fact the moment it changes. Every page shows when it was last updated.
- Predicting the market as a whole. It reads conditions; it does not forecast the index.
How this differs from a traditional analyst
Coverage
FrontierPicks147 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.
Typical analystA short coverage list, updated in spurts.
Accountability
FrontierPicksPublishes the price that would prove each call wrong, and grades every pick against the S&P 500.
Typical analystRarely publishes what would change the call, or a full record.
Consistency
FrontierPicksThe same framework on every stock.
Typical analystIncentives and past calls can bend the read.
Incentive
FrontierPicksNothing to sell, no outside money.
Typical analystBanking, commission or access conflicts are common.
Why a full record matters
Most performance claims can't be checked because of survivorship bias: the winners get screenshotted and the losers quietly disappear. A few good calls, presented as a system, tell you nothing about the calls that were dropped.
The scoreboard closes that gap by rule. The grading rule was published on 1 October 2026, before any pick under it had an outcome. Every medium-or-higher call opens a pick at that day's close, every pick is graded against the S&P 500 when it closes, and every graded pick stays on the board. See how a pick is graded.
Common questions
- Can you trust an AI stock analyst?
- Trust the record, not the voice. Every thesis comes with the price that would prove it wrong, every read is dated, and every pick is graded against the S&P 500 on a public scoreboard. You can check the reasoning and the results instead of taking anything on faith.
- How do I know losing picks are not hidden?
- The grading rule was fixed before any pick had an outcome: a pick opens at the close on the day it is published and is graded when it hits its kill line, when conviction is cut, or after three months. Every pick that opens is listed, and the scoreboard counts all of them.
- Don't language models make things up?
- They can, which is why every claim in a dossier carries its source and date, every edition passes automated checks before it is published, and every call carries a kill line. Nothing here is presented as certain.
- Is FrontierPicks investment advice?
- No. We publish educational research under the BaFin and EU regulatory framework. No personal advice is given and no orders are accepted.
- Why run a model portfolio?
- It puts the research to work in public at real market prices — decided by Claude and measured every day against two fixed, published rules — with every buy and sell recorded, and it leaves no position for us to talk up.
Read the work, check the dates, and watch the picks against their lines. Then decide for yourself.