One System of Record, Many Derived Representations Diagram

Why Not Write Twice

Writing directly to two stores lets them order concurrent writes differently
Neither store decides the order, so they can diverge permanently (Figure 12-4)

One Total Order, Many Derivations

Funnel all input through ONE system that decides a total order (state-machine replication)
CDC or an event-sourcing log both work — the total order is the principle
Derived updates are deterministic and idempotent, so recovery is easy

Loose Coupling Contains Faults

Batch + stream processors maintain derived datasets asynchronously
A slow or failed derived system never blocks the system of record