Questions, answered plainly

A research system for US stocks. It collects market facts every night, builds technical and fundamentals cards with code, and lets Claude give you a verdict on a stock when you ask. Each verdict is recorded and graded later against what the price did.

MFA is invite-only. Request an invite and tell us what you would use it for. Requests are read by a person. If we can offer a place, you will get an email with a link to set your password.

US-listed stocks, about 4,000 of them. Other markets are not covered.

Your own Claude account, where you add the MFA plugin and connect it to your MFA account, and a Telegram account if you want alerts. The web dashboard works in any browser.

MFA will give you real opinions when you ask for one, such as "I would start small here", with the evidence, the case against and what would change its mind. Those answers carry one line: "Not financial advice — this is research, and the decision is yours." It is research, and you make the decision.

It is shown as missing. MFA does not estimate, average in or default a missing value to zero, and a stale price is shown with its date. A verdict lists every part of the dossier that was missing or out of date, and whether it matters.

In two places. Small models read news and filings into structured reports, and code drops any figure that does not appear in the cited source. Claude makes the judgment when you ask for one. Data collection, the cards, alerts, digests and grading are all plain code.

Each night, open calls are checked against the stock's own daily prices. A call ends when its stop or target is touched or its horizon is reached, and its return is compared with SPY over the same days. If prices are missing, the call waits; it is never given an invented result. Hit rates appear only once 20 calls of a kind are graded.

Because the ones we tested did not work. Nineteen popular entry setups, including breakouts, MACD, golden crosses and Supertrend, were backtested over five years on about 4,000 US stocks against random entries on the same days, after costs. None passed the bar set before the test. You can still set alarms on indicators; they are a heads-up, not a reason to buy.

The one portfolio rule MFA runs: a core in SPY plus the 20 US stocks with the strongest 12-month momentum (skipping the last month), equal weight, reviewed at each month end. It passed a pre-registered backtest. The momentum part on its own is volatile and has had deep drawdowns, and past results do not promise future ones.

It receives alerts, a weekday note and a Monday scorecard, plus trade confirmation codes if trading is enabled. The bot only sends; questions go to Claude. You link your Telegram account in the dashboard with a one-time code.

No. Members who are enabled for it can place orders on their own Binance account through Claude, and each order needs their explicit yes plus a code sent to their Telegram. Orders have per-order and daily limits. Without that, MFA only records the trades you tell it about.

No. The book, the alerts and the profit and loss all assume long holdings, so MFA does not pretend to model anything else.

Yes. Paste or attach your broker's trade history in Claude and MFA records the fills. Average cost is worked out from the fills, so you can check it by hand.