Why we run CQRS and event sourcing entirely on Workers and Durable Objects, and how the pieces fit together.
This is a placeholder article. The content layer parses this frontmatter against a
Zod schema, so a missing title or a malformed pubDate fails the build rather
than shipping a broken page.
Every article route also emits a .md sibling (/articles/hello-cqrs.md) and is indexed
in /llms.txt, so crawlers that do not execute JavaScript get the source directly.
Because the current state is derivable and the events are not. Once a fact is recorded it stays recorded, so a change to how we interpret those facts is a recomputation over data we already hold rather than a fresh round of collection.
It gives a single stream one authoritative writer. Commands for an aggregate route to its own Durable Object, which decides them in order against the stream it owns — which is what makes an append-only log safe to write to concurrently.
No. The browser holds a local event store and projects its own read models, then syncs the log. Client-tier commands execute optimistically and replicate; server-tier commands are decided authoritatively on the server.