That exact same thing happened to me, except that I was using a smaller error rate, and a Hamming(7,4) code for error correction.
After the ordeal was over, I looked at the code that the other guys had been writing. They hadn't started on the error detection and correction -- I'd heard much wailing and gnashing of teeth earlier about how mathematical it was -- and their code (all in one big main function, with no indentation) wouldn't compile. I watched as they spent about an hour randomly permuting it, to no avail.
I'm not sneering at these guys. I'm baffled by them.
I was about done implementing the first draft of my side and asked the other side how it was going so we could test some actual communication. The response I got was "it's about done, we just need to split it up into functions". I was initially shocked and then naively impressed that someone could actually reason about the problem without breaking it down.
The end result was of course that I just had to give up and implement both sides of the communication. This was eventually a much better learning experience. I ended up abstracting out the serial port and allowing the two sides to communicate through a unix pipe with random bit errors introduced in packets to test the recovery. I could then run much longer testing without depending on the lab or someone else. I think I eventually tested it enough that I was up against the fundamental problem that the cheap checksum we were using let errors pass way too easily.