Omnigamer

If there was any doubt, yes, that video is not real. It just overlaid values over the HUD. This is most obvious from the first few points of the match where it curiously starts in 2nd without going to Neutral, then proceeds to shift twice while staying in 2nd... You can tell when a shift occurs by watching the jumps in the tachometer. The number of frames the game is in second 5 (17) also matches with a 5.57.

There are a couple of things with that question. If the race were to play out exactly as normal - ie you don't gain any distance until countdown hits 0 - then yes, 5.51 is the theoretical limit with perfect play, and assuming you are in motion (not clutched) on frame 1. If you assume that on the very first frame you clutch into 2nd, as Todd describes, then at best 5.54 is possible. However, the game wouldn't operate that way for this reason. If you somehow manage to be in gear during the countdown and you have some amount of tachometer, you will gain distance. Even if you were only able to achieve this state in 1st gear, times of 4.xx would be possible. 5.51 would be a pretty bad time with this level of exploit. If you had some means to achieve it for subsequent gears without triggering the early start, I think I calculated that something like 2.xx is possible. There is no way for this to occur within the code as written, however.

Hi. This thread has been very busy since I last saw it. Just going to quickly address some of the major points that have come up throughout: -The most likely scenario is a clerical error in entering the initial 5.51s, which were likely misread 5.61s. -I have no way of knowing how many players legitimately accomplished a 5.57 back in the day, but I would venture very few to none. Todd, at least, could not have mathematically achieved a 5.57 using the strategies he showcased in every bit of historical footage. -Since Todd had a working relationship with Activision, he may not have even needed to fabricate evidence at all. It's entirely possible that they simply trusted him at his word. -Activision never officially removed the other alleged 5.51 players. This is true both in the publications and from discussions with one of the players. This claim only comes from Todd. -The easiest, most compelling way to show 5.51 as an impossible time is through the simplified model that Thomas provided. I made a video going through a similar exercise; it's very straightforward, and just uses some basic arithmetic. -The *true* search space, using only inputs and time reference is significantly large. However, there are other ways to break down the problem. Benoit Esnard cleverly mapped the number of state combinations of critical internal variables, and used that to simplify the search space. You can perform an exhaustive test in a very short amount of time now. You can find his code here: https://github.com/esnard/dragster -The above code confirms that the best achievable time is a 5.57 with remaining distance of 98; the same as the TAS I submitted to TASVideos. -Another user had previously developed a dynamic programming approach to do something similar, but its code and results are unfortunately no longer available. It also supported that 5.57 with 98 distance was optimal. -Starting in 2nd gear would allow a 5.51 if and only if you were already in gear on the first frame of activity; ie no speedfreeze. If you are instead shifting from 1st to 2nd on that first frame, the best is still only a 5.54. -There is no benefit to being the 2nd player, or on being the car that has a "0.00" drawn for 1 frame. Both players are capable of achieving the same times.

As far as executing known tool-assisted inputs on console, at least two individuals have done this so far. Both noticed no differences between the end state and the expected/emulated state, and ran the experiment several times. As for building a neural net, I think the current models do more than enough to show the impossibility of faster times. Even for brute-forcing optimal patterns, there are more concise ways to cover the search space than to try to back up potential state from an assumed ending value. Formal methods would probably handle that aspect better as well.

The win condition is checked before any of the tachometer logic, so yes I suppose you can win on the same frame your engine would otherwise blow-out. It doesn't affect any speed stuff though because of this order of operations.

I don't know the exact effects from repeated power cycling, but like I stated earlier it seems likely to me that it's leaving the program counter or other registers in a dirty state. This could do a lot of things to program flow and execution, but if you're at that point then you're already not playing the same game as far as I'm concerned. I lump that in with my blanket case of "hardware failure."

ACTUALLY... I was browsing through the archived Activisions here on AtariAge (thanks for having them in full! link: https://atariage.com/magazines/activisions.html) and it seems like the timelines for things is... off. Up until Volume 5 (Dec 1982), the top listed time for Dragster was always 5.57, shared by 3-4 people, Todd among them. In Volume 5, however, is the first time 5.51 is listed. What I hadn't seen before was on the next page, where it shows that only 2 individuals had claimed the time - Todd not among them. In Volume 6 (Spring 1983), Todd is also added to the list. Volume 7, which to my knowledge is the last issue, doesn't include Dragster at all. So yes, D.Yancey is right, their names were never printed in the newsletter again. For that matter though, neither was Todd's (with respect to Dragster). I don't see any notices about disqualifications or otherwise though, so again we only have Todd's word to go on for that. EDIT: Also, to provide some more context, Todd claims he achieved the 5.51 in his home on March/April of 1982. He also claims he achieved it at the Electronic Thing show in ~August of 1982. It seems like a pretty significant time discrepancy for the publications if he wasn't credited until Spring 1983...

The only information we know on that front comes from Todd himself. Even so, it would all still be speculative. No matter who had what arrangements or not, the limitations of the game and system provide a more objective piece of evidence than personal accounts.

5.51 is impossible even for "auto-shifted" gameplay, as discussed earlier in the thread. Even an idealized version of Dragster without acceleration hazards would not be able to achieve a time of 5.51. As yet, no bugs have been found that would allow such either.

25.51, 55.51, and 75.51 are all valid times. We can speculate all we want about what Activision may or may not have accepted, but there's little chance we'll ever get to see whatever evidence they had for ourselves. You can jump down the conspiracy hole too, since Todd and Activision had something of a working relationship, but really that's not something that's going to be solved. The best we can do now is give a fair assessment of what the game could possibly have allowed, or some scenarios which could have possibly caused faster times to appear.

Thomas was referring to how the two separate players are handled on alternating frames. It's not that gas or shift are read at 60/30 Hz, it's that P1 inputs are read on (for example) even frames, while P2 inputs are read on odd frames. (this is from memory, don't treat it as exact) Frame 20: -logic- read P1 gas, use immediately use stored P1 Left from frame 18 -more logic- -video logic- -overscan logic- read P1 Left, store in memory for frame 22 Frame 21: -logic- read P2 gas, use immediately use stored P2 Left from frame 19 -more logic- -video logic- -overscan logic- read P2 Left, store in memory for frame 23 Frame 22: -logic- read P1 gas, use immediately use stored P1 Left from frame 20 -more logic- -video logic- -overscan logic- read P1 Left, store in memory for frame 24 ... and so on. This is why, from the perspective of only one player, the game operates at 30 Hz. The order of calculations in the spreadsheets matches the order used in the original programming, where necessary. When you "finish" a game, it continues as normal, displaying the graphics based on whatever state they were left in, including timer and car position. It stops from doing any further calculations though, until you force a reset.

5-11under that was one of my initial thoughts as well, but alas, the game only checks specific inputs once in a given frame. When the input is needed for multiple sets of calculations, like the clutch, it is latched into memory and utilized from there.

That type of brute-forcing also doesn't consider any effects that may come from dirtied state after extended use, only power-on. We pretty well understand those effects by now (according to the code), but if the point is to understand the realm of all possibilities, you would have to cast some of the knowledge aside and extend it out for many more (possibly infinite) frames.

There is no existing material evidence for the score beyond a text listing in Activision's newsletter and Todd's own historical testimony. He has not made any statements whatsoever about this score since the game code was modeled, despite being aware that it was ongoing. A 5.51 has never been reproduced, and in fact nobody has even verifiably reproduced a 5.57 (which is known to be possible) or a 5.54 (believed to be impossible).

That description for "frying" suggests a form of glitching (the electronics kind) or fault injection. This can lead to any amount of dirty states if the program counter or other critical registers aren't fully reset between power cycles. If you allow this kind of activity in some competition or optimization problem, though, then your base medium is inconsistent. The program is no longer functioning in the way it was designed - you've imposed external influences to directly affect its execution and flow. It might be interesting in and of itself to find optimality with those conditions, but it also makes the problem less interesting. For example, I imagine there may be a way of affecting Dragster to trigger the win condition with a time of less than 1.00 with some precisely applied faults or glitches. It's something that you can do, but it's no longer relying on human skill, which is hopefully what the competition aims to measure.