I didn't understand this sentence: it means that, looking at the code and calculating all possibilities of gear changing moments, it would be impossible to achieve 5.51 ? If so, why would any other evidence be necessary ? Or, if it wasn't such analysis (and I think that this one would be definitive), why wasn't this one done ?
Todd Rogers however has multiple impossible scores which are still considered "records" even though they can't possibly be obtained, some of them are stupidly obvious, like games in which the score increases in steps of five, yet he has a number not evenly dividable by five and such stuff.
I guess nobody really knows why he still holds these records.
The score has also been achieved by humans on real hardware.
An emulator with preprogrammed inputs doesn't mean it plays perfectly. It just means that it plays reproducibly. Whether or not it plays perfectly depends on the preprogrammed inputs.
How did they choose the inputs?
Looking at the game manual, it comes down to deciding when to shift, and deciding how to use the gas pedal.
Shifting looks like it is easy to deal with in emulated testing. You have neutral, 1, 2, 3, and 4. You start in neutral. So, an optimal run will start with you shifting once at t=0. You cannot down shift, so your only choice when shifting is up, and once you reach 4 you can no longer shift.
From another comment, it seems the game works in ticks of 3/100th second. That means that for a 6 second race, there are around 1.3 million possible shifting timings, assuming that you cannot skip gears.
For using the gas pedal, it looks like if you are not red lined, there are no circumstances where it would be better to not give it the gas. When red lined but at a time when your shifting plan calls for staying in the current gear, I can't tell from the manual what you should do. It says if you go past that you risk blowing the engine. What it doesn't say is if you get more power when going past the red line, so that it may be worth it.
Assuming that red line is max power, then when at red line but not shifting optimum would be to modulated the gas so as to keep it right at the red line.
For each shift pattern, it should be relatively easy to figure out when the gas has to switch between steady and modulated. Putting it all together that would give 1.3 million candidate inputs, one of which should be optimum.
If going past the red line does give more speed (at increased risk of engine blow out), it is more difficult. If engine blow out is probabilistic, you have to worry about the possibility that someone could get insanely lucky.
At that point, you probably need to analyze how the random number generator works, and how it is used, and how user inputs affect it. One particularly worrisome possibility would be that the timing of user input affects it. For example, if whenever the user presses or releases the gas it uses a random number to pick from several sounds associated with that action, that could be bad news. It would mean that the earlier analysis that the gas would optimally in each gear start out as constant and then switch modulated at the red line is no longer applicable. It could be that there is a run where if someone releases briefly the gas during the "constant" phase, that happens to get the random number generator to a position where they will get very lucky on the engine blow out rolls later.
If that meant we had to try every possible gas pattern, instead of just "constant followed by modulated", we'd have a combinatorial explosion.
Probably the best to do in this case (the "past the red line gives more speed, and the random number generator has a lot of user provided entropy" case) would be to tweak the emulator so that the engine cannot blow out, change the gas pedal input to simply be full gas all the time, and see what the results are. That would at least give us a lower bound on what the luckiest person in the universe could conceivably achieve.
Based on that code, the alleged record was not possible.
The reason that Twin Galaxies didn't accept that analysis is because they're run by members of a clique of old-school gamers who stayed behind when the speedrunning community abandoned TG, and Todd Rogers is part of that clique. The analysis was largely done by Omnigamer, who is a well-respected speedrunner outside of TG but isn't part of the clique. TG's staff chose to believe their own in-group rather than a evidence-based challenge from an outsider until their hand was forced by a viral Youtube video[1] calling them out.
This kind of behavior isn't new for Twin Galaxies. They have a long history of mismanagement and corruption. Most of the times/scores posted on their site lack any publicly-viewable proof because until very recently, proof only had to be shown to TG staff "referees" so that people could keep their special techniques secret from the competition. Modern speedrunning eschews secrecy and requires publicly-viewable video proof for all runs.
"I seem to recall Walter Day had sold the rights to TG a couple years back. I used to work for TG some years ago, but hated it. My brother I and were constantly having to fight with what I call "the old guard" comprised of record holders from the early 80's. First, they resisted every one of our innovative new ideas like speed-gaming records for consoles, and then later we discovered a lot of the records from the '80s were bogus because video proof was not required back then. The worst was "Mr. Activision" (aka Todd Rogers), where I and several others were able to prepare a case and conclusively prove Todd had been completely fabricating world records for various Atari games. To name a few:
Barnstorming (2600): Todd's record, which stood for many years, was proven to be impossible once we broke down the game code and stripped the stage of any obstacles. With the stage completely blank, flying a straight line to the finish was slower than Todd's record. When we presented this evidence, we were attacked by fans and supporters of Todd, and eventually an excuse was cooked up that I lovingly refer to as "the coffee stain excuse". Yes, after being attacked and told we were clueless about how good Todd was, one of the referees covered for him and claimed the 'document' detailing his record had a coffee stain on the part where the record time was listed. Instead of throwing the record out and forcing Todd to do a legit one on video tape, they just simply adjusted the record to be MAYBE possible by adding a half-second to the time."
Source: http://nintendoage.com/forum/messageview.cfm?catid=31&thread...