What We Learned Moving Search to the Edge

The performance improvement was real, but not where we expected it. The biggest gain came from avoiding repeated work before the query reached the search service.

Caching common query shapes helped. More important was reducing the amount of metadata required to render the first result set.

The migration also exposed an architectural smell: too much presentation logic had leaked into the search response.

The useful outcome was not simply lower latency. It was a cleaner boundary between retrieval and rendering.