Watch AI assemble a new factor from raw signals

Every factor in StrategyNet starts as a raw signal — a single cross-sectional measurement computed per symbol, per day. The interesting work is combining those signals into something that carries return. This walkthrough shows how the Signal Designer uses AI to do that composition, from a one-line idea to a validated, tracked factor.

1. The idea, in plain language

The analyst doesn't write code. They describe the intuition:

"Find me stocks that are quietly being accumulated — improving fundamentals, but the price hasn't caught up yet, and the tape isn't crowded."

The Signal Designer parses that into three ingredients it already has in the library, and pulls the candidate raw signals:

2. The AI proposes a recipe

Rather than average them blindly, the AI weights each signal by its recent information coefficient and proposes a composite. It deliberately takes the negative of crowding — quiet names, not crowded ones.

Weights are z-scored per signal and blended cross-sectionally, then re-ranked into a single exposure per symbol.

3. From factor to portfolio

A factor is only worth keeping if it survives construction. The candidate is run through the same two-stage pipeline every production factor uses: build a long/short, market-neutral factor-mimicking portfolio (FMP), then measure it.

4. Validation before it earns a slot

The walk-forward backtest reports the same stats the analyst sees for every factor in the library. The candidate clears the bar the AI set for promotion:

The low correlation to the existing book is the point: the AI isn't just finding a good factor, it's finding one that adds something the library doesn't already have.

Where this lives in the product

This is the Signal Designer stage of the pipeline. The same composition mechanism works from raw signals (as above) or from combinations of existing factors — a factor-of-factors — when you want to blend styles rather than signals. The factor catalog lists what's available to build from today.