Next, someone should build a unusably slow but completely accurate transistor-level simulation of the 386 (as they did of the admittedly much simpler 6502) and validate this against it to ensure that it actually is 100% accurate.