Size tiered compaction vs leveled compaction for Akka Cassandra persistence

What is the recommended compaction strategy for Akka Cassandra persistence? The default is SizeTieredCompaction and there is a note suggesting to look in the tests for more config. There is config in a test for size tiered, leveled, time window and date tiered compaction. It seems to me that size tiered isn’t a very good fit for event sourcing because the messages can very easily be spread out across many, many sstables. Leveled compactions is much better for reads in particular which would be important after restarting services and it is better at keeping the number of sstables required to be read to a low number. Any thoughts regarding which compaction strategy is recommended and why?

Also, I noticed in the test settings for size tiered compaction that bucket_low is set to 2.5. This seems very odd. Does anyone have an explanation for this reasoning or are the settings in the test really not recommended settings?