The only arguable objection I have is the missing possibility to detect configuration errors from the calling code.

If it is intended to detect errors only by reading the output and implicitly fall back to a default or last sucessfully set value, then this approach is fine.
If there is a need to react to configuration errors individually (e.g. print "No, you won't get a car from me") then I see two possible solutions:

1. Use configuration functions that report whether the call succeeded or not. Unfortunately that would make the configuration less convenient than the functions which return self.
2. Set a testable error flag in the builder, optionally have one error flag for each build step to be able to locate or resolve the error.

The only arguable objection I have is the missing possibility to detect configuration errors from the calling code.

If it is intended to detect errors only by reading the output and implicitly fall back to a default or last sucessfully set value, then this approach is fine.
If there is a need to react to configuration errors individually (e.g. print "No, you won't get a car from me") then I see two possible solutions:

1. Use configuration functions that report whether the call succeeded or not, e.g. Option<Vehicle> or Result. Unfortunately that would make the configuration less convenient than the functions which return self as you loose the possibility to directly chain the commands.
2. Set a testable error flag or Result in the builder, optionally have one error flag or Result for each build step to be able to locate or resolve the error.