What the Deterministic Rules Engine Does

Once facts are validated, the rules engine evaluates the current workflow state and chooses the permitted next step.

A simplified example:

def next_step(state, facts):
    if not state.customer_verified:
        return "request_verification"

    if facts.get("transaction_id") is None:
        return "ask_for_transaction"

    if not state.dispute_reason_confirmed:
        return "ask_dispute_reason"

    if not state.action_confirmed:
        return "request_confirmation"

    return "submit_approved_dispute"

This is illustrative pseudocode, not a complete production dispute policy. Real implementations should also enforce authorization, idempotency, connector error handling, and audit requirements.

The key difference is that the next step comes from explicit rules and state, not from an open-ended LLM instruction.