How MFA works

Every night MFA collects daily prices, company fundamentals and analyst estimates for about 4,000 US stocks. For the stocks members hold or watch it also collects hourly prices, news and SEC EDGAR filings, including the text of the material ones, and refreshes them through the day. Any other stock can be refreshed on request before an analysis. Reddit is tracked as mention counts and links to sources, not as sentiment.

Every stored fact keeps its date and its source. When a fact is missing or out of date, it is reported as missing or stale, never filled in.

From those facts, plain code with no AI builds two cards for each stock after every US session.

• Technical card: trend, moving averages and momentum indicators, relative strength against the rest of the market, the market regime, and support and resistance levels.
• Fundamentals card: valuation, growth, margins and estimate revisions, compared with the company's industry. A ratio that cannot be computed honestly, such as a P/E with no earnings, is marked not available.

You can screen the whole universe on these cards from Claude, for example for names near their 52-week high with accelerating revenue.

News, filings and earnings releases are read by small, inexpensive models into structured reports: what happened, how material it is, whether it is new. The reports follow a fixed schema, and code checks them: any figure a report quotes has to appear in the source it cites. If it does not, the figure is dropped and the drop is counted.

Text from news and filings is treated as quoted material from third parties, never as instructions.

Judgment happens in Claude, through the MFA plugin, using your own Claude account. You sign in to MFA once when you connect it. After that you just ask:

• Analyse HUBS.
• What should I do with my portfolio this month?
• How much should I buy if my stop is here?
• Tell me if this closes below its 200-day average.

For an analysis, Claude asks MFA to refresh the stock, waits for the fresh facts and reports, reads the full dossier, and answers. A verdict is one of buy, start small, hold, trim, sell or pass, and it comes with:

The verdict is then recorded, with the price taken from MFA's own store at that moment, not from the model. Claude only runs when you write to it; nothing is analysed on a schedule behind your back.

Alerts are written by code from stored facts and sent to your own Telegram. There is no AI in the alert path, so the same facts always produce the same message.

• Stops and targets on your positions, and price alarms you set.
• Indicator alarms, such as a Supertrend flip, an RSI cross or a close above the 200-day average, checked on the daily close.
• Automatic alerts on what you hold: an unusual move, a material filing, high-impact news, or earnings in the next couple of days.

On weekday afternoons you get a short note on your holdings and watchlist, and on Monday mornings a scorecard of your recent calls. Both stay silent when there is nothing to report. The number of pushes per day is capped, so a busy market does not bury you. Every message sent is logged, and Claude can show you exactly what went out.

Every night, open calls are checked against the stock's own price history. A call closes when its stop or target is touched, or when it reaches its horizon, and its result is measured against SPY over the same period. A "pass" counts as right only if the stock then did no better than the market.

When the prices needed to grade a call are missing, the call stays open. It is never given a made-up result. Hit rates for a kind of call are shown only after at least 20 graded calls; below that they read not available.

MFA runs one portfolio rule, momentum-1. It keeps a core in SPY and adds the 20 US stocks with the strongest momentum over the past twelve months, skipping the most recent month. Candidates must trade at $5 or more, with at least $100 million changing hands a day. The 20 are held in equal weights and the list is reviewed on the last trading day of each month.

It is the only rule that passed a test whose pass marks were written down before the test was run. Adding stops or a market-timing exit made it worse in testing, so it has neither. On its own, the momentum sleeve is volatile and has had deep drawdowns; the SPY core is there to soften that, not remove it.

At each month end you get a review for your own book: what to sell, what to buy and what to keep. How much goes into the strategy is your choice.

Members who are enabled for it can place orders on Binance with their own account. You add your own API key in the dashboard; it is restricted to MFA's server address and stored sealed, so only the trading service can open it.

Claude can prepare an order, but it cannot place one on its own. Every order needs your explicit yes and a short code that is sent to your Telegram, never to Claude. Each order and each day has a spending limit. Fills are added to your book automatically.

• It does not promise returns or present any result as guaranteed.
• It does not model shorts, options or leverage; the book assumes long holdings.
• It does not treat chart patterns as buy signals; none passed its backtest.
• It does not mix currencies or convert them; amounts always say their currency.
• It does not trade on its own, and it does not chat on Telegram.