Comment by makeitdouble

The text based side of protobuf is not base64 or json. We'd be looking at either CSV or length delimited fields.

Many large scale systems are on the same camp as you as their text files flow around their batch processors like crazy, but there's absolutely no flexibility or transparency.

Json and or base64 are more targeted as either low volume or high latency systems. Once you hit a scale where optimizing a few bits straight saves a significant amount of money, self labeled fields are just out of question.