How TradeFlare's AI research works
TradeFlare is an independent research project — not a registered company, investment adviser or broker. Everything it publishes is AI-generated research for education, and it can be wrong. This page explains exactly how the research is made and how every call is scored, so you can judge it yourself.
394 signals published since January 2026
10,324 dated AI calls on 1,656 stocks in the nightly record
1. Where the stock ideas come from
Every market night after the close, prices and fundamentals are refreshed for about 1,700 stocks. Several independent screens then pick candidates for deep research:
- Insider buying — executives and directors buying their own company's stock, from SEC Form 4 filings (see Smart Money).
- Earnings surprise and post-earnings drift — companies that beat expectations and keep moving afterwards.
- Analyst revision momentum and broker ranking changes.
- 52-week breakouts — stocks making new highs on rising volume.
- Quality and long-run strength screens, plus news-driven screens such as layoff announcements.
Members can also ask for research on the stocks they own or watch. Those are researched privately for that member and are never published on public pages.
2. Four AI research teams, one verdict
Each shortlisted stock is analysed by 15 AI agents (large language models, currently Google Gemini) organised in four teams:
- Technical — price trend, momentum, support and resistance, volume and chart structure.
- Fundamental — revenue, earnings, margins, cash flow and debt, read from the company's own SEC filings where available and checked against market data.
- Sentiment — news coverage, analyst tone and how the market is reacting.
- Risk — volatility, valuation, concentration, upcoming events, and a risk score from 1 to 10.
A final "fund manager" step reads all four reports and writes the call: BUY, HOLD or SELL, a conviction from 0 to 100, and for a BUY an entry price, a target and a stop. The full reasoning is shown on each stock's page, and every night's call is kept on the stock's history page (for example NVDA) with how the price moved 5 and 20 trading days later.
3. When a call becomes a published signal
Most calls are not signals. A BUY is published to Signal History only if it passes every rule below. The rules are enforced in code, not left to the AI:
- Rank: in the top 30% of that night's list, and conviction of at least 60 (80 when the market is in risk-off).
- Market regime, from the S&P 500 (SPY): risk-on when it is above its 20-day average and has not fallen more than 2% in five days; caution when one of those fails; risk-off when both fail. In caution and risk-off, positions are half size; in risk-off, the momentum screens (breakouts, post-earnings drift, revision momentum) are paused.
- Risk veto: a risk score of 9 or 10 out of 10 is never published.
- No chasing outside risk-on: skipped if the stock is at its 60-day high after a run of more than 12% in 20 days, or if its volatility is above 40%.
- Trend: skipped if the price is below its 200-day average.
- Entry and payoff: the entry must be within 15% of the current price, and the target's reward must be at least as large as the stop's risk.
- Earnings: skipped if the company reports within 5 days.
- No repeats or crowding: no new signal on the same stock within 30 days, no more than 5 open signals in one sector, and only one share class per company.
4. How every signal is scored
Each signal is judged against its own published target and stop, on daily closing prices:
- A close at or below the stop → hit stop, at the stop price.
- A close at or above the target → reached target, at the target price.
- From day 45, the first close that is up 1% or less → closed after 45 days, at that close.
- Otherwise the signal stays open and is marked at the latest close.
The first rule to trigger wins, and a closed result is locked — it is never changed afterwards. The scoring rules were tightened on September 2, 2026 (the 45-day exit and daily-close basis); signals closed before then keep the result recorded under the rules in force at the time, so results before and after that date are not strictly comparable. All results are hypothetical: no real money is traded.
5. What is published, and what is not
- Every published signal, win or loss, stays on Signal History.
- Every night's AI call on a stock stays on that stock's history page, including the misses.
- Members' own stocks, positions and prices are private and never appear on public pages.
- TradeFlare is not paid by any company whose stock appears here.
Limits. AI models make mistakes, and source data can be wrong or late. Past results do not predict future results. Nothing here is personalised advice or a recommendation to buy or sell any security — see the
Terms.