How It Works

The deterministic recommendation engine, end to end.

The engine runs as a pipeline of deterministic stages:

  1. Market data — prices, volume, market cap, dividends and splits are pulled (keyless) and normalized into one price history.
  2. Factors & technicals — per asset, AlphaForgeReco computes momentum, quality, growth, value and risk factor scores plus technicals (RSI, moving averages, volatility, drawdown, trend).
  3. Deterministic scoring — each asset gets a composite 0–100 Score by blending the factor scores under the chosen style (Balanced, Quality, Aggressive Growth, Momentum). The weighting is fixed and reproducible.
  4. Ranking — a shortlist is pre-scored, then ranked by the composite. Optional Themes/Sectors filters narrow the universe first.
  5. Confidence — computed from factor agreement, data completeness and signal support. It measures how well-supported the case is, not the probability of profit.
  6. Signals — separately, rules detect structured events (breakouts, oversold RSI, volume spikes, MA crossovers, rating changes) and emit those above an importance threshold.

Nothing here is hardcoded per ticker and no step uses a language model. The AI narrative is added only at the end.

Why deterministic? Because every figure must be reproducible and gradable after the fact — see Recommendation Tracking.