Most behavioral scoring treats each interaction as independent. Temporal engines treat them as a sequence — a trajectory through behavioral state space with momentum, phase transitions, and predictable future states.
The Engines
- State Machine Runtime — Maintains a behavioral state for each actor (e.g.,
exploration → commitment → negotiation → close) and detects phase transitions. - Causal Graph Engine — Builds a causal model of behavioral patterns: what actions cause what responses. Distinguishes correlation from causation using do-calculus.
- Counterfactual Reasoning Engine — Answers "what would this actor have done if the context had been different?" Used for stress-testing decisions.
- Long-Horizon Prediction Engine — Projects behavioral trajectories 30, 60, and 90 days forward. Confidence degrades sharply beyond 30 days; the uncertainty band expands accordingly.
Code Walkthrough
// Behavioral state machine: track phase transitions
class BehavioralStateMachine {
constructor(actorId, initialState) {
this.actorId = actorId;
this.state = initialState;
this.history = [{ state: initialState, ts: Date.now() }];
}
transition(newState, evidence) {
if (!VALID_TRANSITIONS[this.state]?.includes(newState)) {
return false; // Invalid transition — log as anomalous
}
this.state = newState;
this.history.push({ state: newState, ts: Date.now(), evidence });
eventBus.emit("state.transition.detected", {
actorId: this.actorId,
from: this.history.at(-2).state,
to: newState,
evidence,
});
return true;
}
}
What to Watch For
- Long-horizon predictions should always display their confidence decay curve, not just the point estimate.
- Causal graphs built on observational data contain confounders. Use instrumental variables or randomisation where possible before trusting causal claims.