An operator can force a rebuild when the hosts changed

Continuous delivery decides whether to build by asking whether the current commit already has a complete set of images. That question is keyed on this repository's commit — but the portal hosts live in MeshWeaver.Plugins. A merge there that edits a file shipping inside the image, such as the host's appsettings.json, changes what the image should contain and changes nothing this check can see.

The consequence was a change that had merged and could not ship: the newest image predated the fix, and no producer existed that would ever rebuild — delivery simply had nothing to notice. It waited for an unrelated commit in this repository to happen along.

Dispatching the workflow by hand did not help, and that is the part worth being precise about: a manual run took the same reconcile path, found the image set complete for the same commit, and decided to re-assert the content bake without building. It was a structural no-op for exactly the case an operator would reach for it.

Running the workflow now takes a Force rebuild option, which builds the full image set for the current commit even when its set is already complete.

This is the escape hatch, not the fix

The durable repair is to make the image's identity honest — record which MeshWeaver.Plugins commit it was built from, and have the completeness check compare that against the current one — so the reconciler notices on its own rather than waiting for a human who happens to know. That is still open.

What changes today is that someone who does know now has a way to say so. Before, they did not.