Build Event-Sourced Projections and Read Models

Defines a projection architecture (event store to projector to read model) and four projection types - live (real-time from subscription), catchup (processing historical events), persistent (storing a checkpoint to resume after restart), and inline (same transaction as the write, for strong consistency) - then ships five runnable Python implementations built on an abstract Projection base class and a Projector runner that reads events in batches, dispatches to registered projections, saves checkpoints, and supports rebuilding a projection from scratch. The templates are: an order summary projection updating a PostgreSQL order_summaries table across OrderCreated, OrderItemAdded, OrderItemRemoved, OrderShipped, OrderCompleted, and OrderCancelled events; an Elasticsearch search projection indexing, updating, and deleting product documents on ProductCreated, ProductUpdated, ProductPriceChanged, and ProductDeleted events; a daily-sales aggregation projection that upserts per-day order counts, revenue, and item totals on OrderCompleted and reverses them on OrderRefunded; and a multi-table customer-activity projection that writes to customers, customer_activity_summary, and customer_order_history tables inside one transaction across CustomerCreated, OrderCompleted, ReviewSubmitted, and CustomerTierChanged events.

Use it when building CQRS read models, creating materialized views from events, optimizing query performance, implementing real-time dashboards, building search indexes from events, or aggregating data across streams. Skip it for anything unrelated to projection patterns or outside this scope.

Input is an event stream and the read-model shape needed - a summary table, a search index, an aggregate, or a multi-table view; output is a Projection subclass implementing name, handles, and apply, ready to register with the Projector runner. The best-practices list covers idempotency for safe replay, transactional multi-table updates, checkpoint storage, lag monitoring, and designing for full rebuilds from scratch.

Built on asyncpg for PostgreSQL read models, with a fifth template targeting Elasticsearch's AsyncElasticsearch client for full-text product search. Assumes an event store and a checkpoint store as pluggable dependencies rather than naming a specific event-sourcing framework - the Projector class takes both as constructor arguments and reads events in configurable batch sizes, so it can sit in front of any store that exposes a read_all(position, batch_size) method. References Microsoft's CQRS pattern documentation and an external projection-building-blocks article.

Engineers building event-sourced systems who need concrete, checkpointed projection code for CQRS read models rather than a conceptual explanation of projections - particularly teams standing up their first search index or reporting rollup from an existing event stream.