Okay, I seem to have solved my own problem--the second channel, 21231 is wrong, it should have been 21240. Just to clear any confusion for anyone who comes across this posting:
My task was to generate an arbitrary waveform, and then record that waveform at the exact same times that I record other physical channels. As such, I decided to generate the waveforms on analog output 0 and 1, and then literally wire those outputs back into two of my analog input channels--that way, I can record those two channels as physical channels along with my other physical channels. I simply was looking for a way to bypass the wiring part--in effect, look for an internal routing of the output at the analog channels to the analog input channels. After searching on the NI website,
I found out that there are diagnostic virtual channels--21230 and 21240--that represent analog outputs 0 and 1. So, instead of literally wiring output to input (and wasting 2 extra analog input channels), I simply added 21230 and 21240 to my analog input scan list. If this represents a case of mixing physical and virtual channels, then I guess doing so is okay, contrary to what the knowledge base article says. My physical data and my arbirtrary waveform data are being collected perfectly, at the same time, under hardware control.