# Signal Taxonomy — Recipe · Ingredient · Scheduler

**Canonical URL:** https://stratcraft.ai/alpha-factory/signal-taxonomy

## The Three-Layer Taxonomy

Every quantitative trading operation separates into three layers:

### Layer A: Signal Pack (Recipe)
What produces a read. Four types:
1. **Indicator** — RSI, MACD, Bollinger Bands, Supertrend. Deterministic calculation over OHLCV.
2. **HMM · n-gram** — Hidden Markov Models. Regime detection (trend/mean-revert/chop).
3. **ML** — XGBoost, LightGBM, LSTM, transformers. Learned models predicting return/direction/volatility.
4. **Factor** — Fama-French, momentum, quality, low-vol. Cross-sectional factor exposures.

### Layer B: Data Layer (Ingredient)
What the recipe consumes. Three tiers:
1. **L1 · OHLCV** — price, volume, timestamp. Near-zero cost, fully public.
2. **L2 · 318-factor library** — value, momentum, quality, low-vol, 314 more. Moderate cost.
3. **L3 · Alternative data** — NLP on filings/news, macro feeds, on-chain, satellite, weather. Highest cost, highest noise-per-dollar.

### Layer C: Combinator (Scheduler · Moat)
What fuses the reads into one decision. Five built-in methods:
1. **Equal Weight** — f = (Σ sᵢ) / n
2. **Confidence Weighted** — f = Σ sᵢ·cᵢ / Σ cᵢ
3. **Voting** — f = sign(Σ sign(sᵢ))
4. **Max Confidence** — f = sₖ where k = argmax cᵢ
5. **Min Confidence** — f = sₖ where k = argmin cᵢ

## Architecture Principle

The combinator is the only moat. Signals are commodities — widely available, easily replicated. Data layers are commodities. The specific way you weight, rank, and schedule many signals into one decision is what competitors cannot copy.

## Three Common Category Errors

1. **Calling alt-data "a signal"** — NLP on 10-Ks is a Data Layer (L3). Without a recipe computing on it and a combinator fusing it, there is no signal.
2. **Selling signals as a finished product** — A buyer always brings their own combinator. They are paying for data, not for fusion.
3. **One model = a strategy** — A single recipe with no combinator is a single-regime bet. LTCM is the canonical example.

## Brand

- Product: StratCraft
- URL: https://stratcraft.ai
- This page: light standalone design