We generally accept that emulators are innaccurate without giving it too much thought. If a certain TAS happens to work on console, or a game happens to be particularly stable and easy to emulate, great, but just as a matter of practicality we don't spend much time thinking about it. If we had to wait for perfect emulation, we'd have hardly any TASes, so we settle for good enough.
What's good enough? Generally speaking if it 'looks right' no more attention is paid to it. This works most of the time, maybe you did a jump in a TAS that doesn't quite line up on console, but if it looks like it could work with only minor adjustments then it's probably close enough. Or maybe some RNG manipulation doesn't quite work out, but if slightly different manipulation could still make it work, well close enough.
But, how do you know? Can we know which TASes are fixable, and which ones are impossible, even in principle? I recently realized that
[2599] A2600 Seaquest by morningpee in 01:39.80 relies on a glitch that doesn't even exist in reality. The majority of the encode is showing something that will literally never happen. I like to focus on accuracy, so for me this was a bit un-settling. But originally I thought, 'Well it's only A2600, who really cares? It's only a 1:45 long TAS, I'll make a replacement and move on.'
Yesterday though I remembered that this wasn't the first time I saw something like this!
[749] NES Paperboy by Randil in 11:30.78 Paperboy is just as bad as Seaquest. The customer houses are determined by the number of cycles from power on until the code reaches a certain point. FCEU is not even close on this, so it gives an arrangement that you would never see in a real console. Now we're in NES territory, with hundreds of runs dating back over a decade.
So now I'm wondering how many more TASes are out there like this. How many LOOK right but in fact rely on emulator bugs? How many impossible TASes are there? I think this is more then a philisophical question. We spend hours and hours working on a TAS, I think most people wouldn't want to work on something that is inherently flawed on the tool they are using.
Can anyone else think of any examples? Are there any runs you suspect have something fundamentally wrong with them?
(I'm currently looking at
[1686] NES Mega Man by Shinryuu & finalfighter in 12:23.34 The NMI race condition at the start of Ice Man stage is known not to work on console (and desyncs in the same manner on BizHawk.) I don't know if it can be fixed or not.)
