Event Sourcing

Persistence approach where business-relevant facts ('events') are appended to an immutable event store and aggregate state is reconstructed by replaying events (left-fold) for the relevant aggregate. Read models / projections are built by projectors and can be regenerated into any shape (graph DB, in-memory, partial views). Enables time travel, temporal queries, total self-reconstruction, simpler tests (no updates/deletes), per-event backups via reactors, and auditability. Davydov frames it as storing 'what happened' instead of current state, then computing state on demand via projection functions (reducer, events, initial state) — a pure function. Has a steep learning curve and forces a mind shift away from current-state thinking; only fully makes sense alongside CQRS. Pašalić notes folding ~200k entries (10 years of bank-account events) takes well under a second; snapshots help for high-volume domains like advertising. Davydov emphasizes event sourcing is about state, not about distributed systems or microservice communication.