How the analysis works

1. News

Headlines from financial news feeds, company press releases and SEC 8-K filings are collected every 15 minutes. Each headline is matched to the company it is about, and FinBERT, a language model trained on financial text, scores its tone as positive, negative or neutral.

2. Technicals

Momentum (14-day RSI, MACD), trend (price versus the 50- and 200-day moving averages) and participation (volume versus its 20-day average).

3. Trend structure

"Smart money concepts": breaks of structure, changes of character, order blocks and fair-value gaps, which describe where large buyers and sellers have been active.

4. Fundamentals

Valuation (PEG ratio), profitability, revenue growth, the latest earnings result against expectations, and the upcoming earnings date.

5. Language-model research

A financial language model reads the recent news and research for the stock and gives its own positive / negative view.

6. Composite verdict

The signals are combined into one score from −1 to +1. Strong positive scores become BUY, mildly positive WEAK BUY, near zero SKIP, and the negative side WEAK SELL or SELL. The confidence figure is the size of that score. The model also proposes an entry, a target, a stop-loss and a date by which to reassess.

Limitations

Models are wrong often. Data can be late or incomplete, headlines can be misread, and markets react to things no model sees. Ratings are not updated continuously and can go stale. Use this site to learn and to generate ideas, never as the only reason for a trade.