All diode tests should be done with the machine off 

I'm thinking along the lines that there is either some sort of short between column 1-7 and column 8, but it would be difficult for column 1 - 7 to all be shorted to column 8.

Seems like a "logic" issue to me, as your test proves that the lamps in column 8 are switched on when the corresponding ROW is switched on.

During a normal game, are all the lamps in column 8 "on" ? I guess they would be