Pooled releases (release trains)

Release on every merge is great for a fast-moving lib, but sometimes you want fewer, fatter releases — one readable changelog per cut, not a version bump per PR. For the manual and scheduled models it does that without changing how you develop — it only changes the release trigger (the prerelease-channel model in Model 3 does shift where you target work; see there). It builds directly on semantic-release-automation — same plugin pipeline, same Conventional Commits; read that first.

The key idea: decouple merge from release

Trunk-based development still wants frequent small merges to main — that's unchanged, so dev velocity and PR flow don't change. The only thing you move is when a release is published: