This can be one of the hardest problems to fix. It basically comes down to a dirty data or addressing problem. If you're getting a RAM error message but you tried to replace the RAM chip make sure your RAM socket in question is good. You can pull them up with a small flat head screwdriver and checking to make sure the pins are clean and not broken. Just make sure you take great care in replacing the socket or you'll be soldering in a new socket. After reinserting the original socket do a continuity check with your VOM to make sure all the pins are still connected. The best way is to replace the socket in question in the first place. Sometimes you'll get a 1-2-1 or 1-1-1, this will help you to isolate the problem to the bank that is failing. If you're getting a 1-1-1 error look at the chip located at 1H, if you get a 1-2-1 look at chip 3H and so on.
Make sure your power supply is good. If you're still using the original PS make sure to rebuild it. Once the original PS is rebuilt it's as good as any switching PS.
You can also try replacing all RAM chips. In some cases I had a problem with a bad RAM chip on the board but was giving me the wrong error code.
Replacing the CPU (6809EP) chip and/or decoder chips and decoder sockets on the CPU board is a long shot but has been known to work.
After you tried all the steps above but are still getting the same error it's time to get out the test equipment. Check to see if your chips at location 1H, 3H and 4H are getting the proper signals (on Defender these are located at 1M, 3M and 4M) Also make sure the 4.7pf cap is in place on each chip (Defender did not use them). The Rev.B boards actually had one soldered to pin 1 and ground on each chip. Later on Williams had a location for the caps on the CPU board, one under 1H and the other two under 5H. Also the four 74153 (74LS153 will not work) between the decoders could be part of the problem. But if these are bad the rug test will often look distorted.
Here is a lay out of the RAM map for the Williams CPU board: