May I also mention that N log N is the total cost of compaction for a DB of size N. You don't perform a compaction on every single write. Amortised per write the cost is more like N log(N)/N == log(N).
Also, N log(N) is nowhere near exponential. O(2^N) would be exponential, and that's not what you have here.
In the case of an architecture diagram being handed down from on high, sure. But in the case of product feature whack-a-mole, this doesn't seem like something that bottom-up decision-making could fix.
Recursive CTEs are a cool feature, but it's a shame the syntax is so awful. You really have to work to understand what the query is trying to do.
Other query languages that make recursion much more bearable are SPARQL (on RDF data), Cypher (Neo4j) or Datalog (Datomic, Cascalog, etc). Would love to see more of those query languages being used.
Ternary (or even higher n-ary) storage is already widely used: multi-level cell (MLC) SSDs store more than one bit per cell by using multiple voltages (most commonly 4, i.e. 2 bits per cell). I believe Ethernet also uses 3 voltage levels.
Does anyone know what they mean when they say "grounded in trust"? (Seems odd to imply that other OSes are not trustworthy.) Referring to code signing, perhaps?
Do you also have stats on the high percentile latencies, besides the average? Sounds like your avoidance of compaction pauses ought to lead to lower latency at the high percentiles.
Also, N log(N) is nowhere near exponential. O(2^N) would be exponential, and that's not what you have here.