A stock AI that learns from being wrong.

It learns which scoring factors predict winning picks. Each factor — earnings growth, momentum, low P/E, a strong daily candle, and a dozen more — has a weight. After a pick resolves, the factors behind it are credited or penalized based on the real return, so useful signals get stronger and noisy ones fade.

Yes. The learning is magnitude-weighted: a pick that gained 8% moves the factor weights far more than one that scraped 0.6%. That keeps the model focused on signals that produce meaningful moves, not just barely-positive ones.

It can. Because it re-evaluates outcomes continuously, a factor that stops working gets down-weighted, which can flip a stock from a top pick to an also-ran. The Track Record even logs when a previous call was effectively reversed.

No. Every pick lists the specific factors behind it, and the learning panel shows each factor's current accuracy and weight. You can see exactly what the model believes and how confident it is.

It applies a confidence ramp — a factor needs a handful of resolved picks before it meaningfully shifts the rankings, so a couple of lucky outcomes can't hijack the model. It gets more useful the longer you use it.