Wow, I'm so glad you brought this thread back. Looking forward to reading the paper. Just flipped through it though - separate transmit and receive stations, each with diodes - that is the way to do it.
By the way, I was in Helsinki in October. It's an amazing place.
main stumbling block I have come to is how to verify that the outgoing packets, leaving the computer you are typing on, have truly been encrypted with the intended public key.
You're right, without the private key you really can't. TFC's OTP encryption can be manually verified using ASCII conversion table (DEC values) and simple clock arithmetics and it's more secure than public key crypto. There are various downsides in convenience regarding OTP but easier auditability and not having to worry about algorithm security makes up a lot of it.
Hashes are used to verify no data errors were present during data diode transmission. From Pidgin you can check what type of message was sent to friend, but in theory a backdoor in both systems could of course send other type of data through serial and into network.
Quoting wikipedia:
Realtime spectrum analyzers are able to see signals hidden behind other signals. This is possible because no information is missed and the display to the user is the output of FFT calculations.
Moreover, you can use a Logic analyzer to store and view raw digital signals.
I'm really glad another person came up with almost the exact same implementation to improve security.