[vklmn]

That's what the website missing indeed. We have a few words about that in docs, but it's still not enough https://jitsu.com/docs/internals/jitsu-server#mapping-step

Overall, Jitsu tries to decompose (aka flatten) JSON as deep as possible. E.g. {a: {b:1, c:2}} will become a_b=1, a_c=2. If column is missing, it will be created. We don't decompose arrays so far

[atombender]

Can one disable the flattening? BigQuery, for example, supports nested objects just fine, and flattening them for no particular reason seems counter-productive.

I work on an application where we already have a schema in BQ, but we'd like to start moving events through something like Jitsu or Rudderstack. This uses nested objects extensively. Looking at Jitsu, it looks like we wouldn't be able to keep our existing table schema.

PS. Whoever wrote your BigQuery code does not understand Go contexts. Only functions should take a context argument; you should almost never store contexts in structs!