So, since Capridog's run just tied the world record, I figured I would make this forum post explaining my timing methodology which led me to determine that these runs most likely tied exactly to the frame.
Some things to note: Brandon's run was using the autosplitter. The way that works is the autosplitter actually looks at the memory values in the emulator itself to determine when to start/split. Due to this, it is impossible for the autosplitter to split early, since the value in memory would have had to already change (meaning the action has occurred) before it will split.
For the start of the run, buzz's first frame of movement will cause his animation to change, showing his left leg lifting up, which is how we can determine the first frame of movement.
Also note that both videos are in 30fps, meaning that one video frame is roughly 33 milliseconds (0.033 seconds).
We'll start by looking at the start of Brandon's run.
Examining this, since the first video frame of movement shows the timer at 0.01, the worst case scenario of the autosplitter splitting late would mean that it theoretically could have started up to 0.02 seconds late.
Now lets look at Capridog's run.
Unfortunately, Capridog is not using the autosplitter, and the way that the video frames line up, we have a difference of 0.05 seconds here, effectively meaning that the real start could have occurred somewhere between 0.32 and 0.37 seconds on the timer.
Now lets look at the final frames of the runs. Brandon's video first again.
You can tell this is the last frame because the blacksmith in the last frame has started his first frame in his death animation (most easily determined by the fact that the blacksmith has started to "jump" up (part of his death animation)).
Fortunately here, there is only a 0.02 second window in the timer between video frames, meaning that this run could have ended any time between 28:44.93 and 28:44.95
Now lets look at Capridog's run.
You can tell this is the last frame because the health bar started to pan up (which means the boss has been finished), even though the blacksmith's death animation didn't start on the same frame.
Again, fortunately here, there is a relatively small 0.03 second window in the timer between video frames, meaning that this run could have ended any time between 28:45.26 and 28:45.29
Now looking at edge case scenarios:
The shortest that Capridog's run could've been is: 28:44.89 (assuming that the first video start frame was the same as the actual game start frame, and that the actual ending game frame occurred less than 0.01 seconds after the previous video ending frame)
Given this information, we have about a 1 frame window that Brandon's run could actually be, and a 2-3 frame window that Capridog's run could have ended.
Given the general accuracy of the autosplitter, it is VERY likely that the recorded time for Brandon's run is accurate to the 0.01 seconds, meaning that his ending time is VERY likely between a 44.94 and a 44.96. All the other possible times are very unlikely
In general, when comparing timing of video vs actual game frames, it is statistically most likely that the actual time will fall in the middle of the possible window (this would follow a bell curve distribution).
Based on this fact, it is most likely that Capridog's actual ending time sits somewhere around 44.94, however due to the nature of this timing, it could fairly easily be +/- about 0.02 seconds (due to evaluating the standard deviations of video vs actual game frame timings)
Given the most likely scenarios of 44.94-44.96 and 44.92-44.96 being the windows of time, the most likely combination of these ranges gives us an outcome there is about a 75% chance that the runs end with less than a 0.03 second gap (which is less than one frame).
Given all of this, it is most likely that the runs likely ended on the same game frame, meaning that these runs are an exact tie.
However, given all of this, the next most likely (and not at all far-fetched) scenario is that Capridog's run was actually 1 frame faster than Brandon's run. The likelihood of Brandon's run being faster than Capridog's run is very unlikely.
So while I will record these runs as an exact tie, it is impossible to know 100% whether or not it was an actual tie.
Since an exact tie is the most statistically likely scenario, that's how I have officially record the run.
Edit: I wasn't super happy with all of this since it wasn't a definitive answer, so I did a lot more testing that I'm not going to try to get into here (timing based soley on video fps, rather than tracking using the timer on the video, do some test recordings and actually getting some standard deviations for game frame times vs video frame times to make a better mathematical model, etc, etc). Depending on how you measure it, Brandon could have won by up to 2 frames or Capri could have won by up to 2 frames, however I still stand behind the idea that the most likely scenario is an exact tie.
Edit 2: I've spent over 5 hours simply timing this different ways, creating a few different statistical models, etc to try and get a better definitive answer. Unfortunately, it's impossible to know for sure because of the videos themselves. I am officially declaring this a tie, and that's (basically) final. It would take quite a bit of testing and showing better methods than what I've already tried to convince me that a different outcome is more likely at this point. Everything seems to point to the most likely probability of an exact, frame per frame tie.
|CapridogCapridog likes this.|
I uploaded a side-by-side comparison of the 2 runs in their entirety, so now anyone curious can actually watch and see how this whole thing unfolds:
Interesting to watch, for sure.
Side note: Amazing.
|CapridogCapridog and GameguySDGameguySD like this.|