Signal families — why these four

The engine's core claim is that confluence across independent evidence types beats any single indicator. The four families are chosen to be as orthogonal as TA allows:

Family	Question it answers	Evidence type
Trend	Is there a directional regime?	price level/slope vs its own history
Momentum	Is the move being pushed right now?	rate-of-change of price
Volume	Is real participation behind it?	non-price data (the only family not derived from close)
Structure	Does the map of levels leave room?	spatial position vs historical reversal points

Each component within a family casts a bounded vote (the point tables), the family clamps to [−1, +1], and every component's vote is preserved as a written detail — this is what makes the family cards in the UI possible and is a hard requirement: no component may influence the score without producing a human-readable reason string.

Design rule: no single component may exceed 0.5 family points, so no single indicator reading can dominate a family, and no family can trade alone (II.3).