See https://github.com/quarkusio/quarkus/issues/41429: to take full advantage of the index layout strategy in Quarkus, and in particular to use a different layout depending on the environment, the layout strategy should be retrieved and used on index start() rather than during model building.

It makes perfect sense, by the way, since the index names are only used to send requests to Elasticsearch, so after start() gets executed.

I think there is no equivalent limitation to be removed in Lucene, as index names there are fully defined by entity names with no concept of layout strategy, but we might want to check.