The Elite Forum
The Big Three Plus One => GoldenEye 007 => Topic started by: Imperfect Clark on June 27, 2020, 09:27:40 pm
-
I saved Boss's predictions from a post almost 15 years ago. Just encountered it by chance and decided to do some data analysis.
If anyone was qualified to make predictions, it was the dominant champ of the time. I remember these to be bold... he was supposing 3:20 in total cuts... over 3 seconds per WR! Despite his foresight, 58/60 have been outright surpassed.
(https://i.imgur.com/wPAMm5R.jpg)
Link: https://imgur.com/gallery/1CzB2YF (https://imgur.com/gallery/1CzB2YF)
The moral of the story is always the same -- do not base predictions off of understanding of the levels or the game itself. Rather, use statistics. That said, it would be fun to try and do a data-driven prediction of limits using the GE WR database.
PS - Boss gave his permission for me to share this. Please be respectful.
-
Boss could not have known about 2.x or some of the other innovations that have occurred, but that's kinda the point I want to make... true forecasting should account for the unknown.
1:01 Surface will happen as long as people keep trying... any attempt to disprove this is moot to me. On an infinite timeline all WRs are 0:00 :)
-
I wonder what he imagined the strat would be on S2 00. Amazing post btw
-
Seems like this is from 2004 or 2005. Or he wasn't very optimistic about Frigate SA...
-
Even back in 2005, there was always talk that eventually, Surface 2 00 Agent would incorporate use of like 2-3 or more nades. I think it was also vastly overestimated just how much time extra nades would save. It's perhaps the biggest example in GE history of the general populace actually overestimating just how much time could be cut from the level... on most levels, it's more traditional to be less generous and underestimate how low the records can go.
Aztec 00A is the other interesting one, but it's always been at least known/understood that in theory it could go equally as low as Secret Agent, maybe even Agent, on paper.
-
Seems like this is from 2004 or 2005. Or he wasn't very optimistic about Frigate SA...
I think you're right, actually
-
I think it was also vastly overestimated just how much time extra nades would save
Exactly... Fernando's run demonstrates that a full skip of 3rd camera still wouldn't yield 1:16.
I remember talk about firing through the window to take out the last camera, but to save time this would require an early/hard nade throw on an already 1-in-1000 run. But Boss used the word "true limit", so I don't mind 1:16 in that context.
$20 toward bounty on Aztec 00 1:33 =)
-
Despite his foresight, 58/60 have been outright surpassed.
Only 54 have been surpassed, 58 have been matched or bettered you mean.
1:16 was the TAS record set by Rising Tempest's full run released 2007-11-26, so it seems a pretty solid estimation for a true limit in 2005. Now the record is 1:09 by Eliminatexenia (using the modern fence shot) and that should go lower.
do not base predictions off of understanding of the levels or the game itself. Rather, use statistics.
Lol how is that going to work, especially for predicting times on the individual levels, in 2005. Even recently there have been a good few new strats, it's not like we're 100% in the "just optimise existing runs" era. Plus you'll get some narrowing of focus in the "final days" as more times become effectively maxed. This will have little regard for how you've counted up the number of same-strat untieds each year, and more interest in whether times are actually achieveable.
-
really cool post. thanks for sharing!
-
Circling back. Whiteted thinks my comment about basing predictions off statistics is unreasonable. Care to offer limit predictions?
The more data points we have (times submitted), the more we can predict about that set of data while only looking at samples. That predictability is rooted in the fact that a natural, consistent source is producing the data points -- human* beings (with common DNA), working toward a singular objective (lowering times). When nature is involved, certain statistical functions become extremely effective at predicting things. This is a key premise in AI... let the data drive predictions. Obviously this works better at scale, so if we're talking about predicting the next year or two or five, even, knowledge of the game is probably a better approach for prediction. So perhaps we agree, mostly.
Lol how is that going to work
A dumb/simple way to actually test this out would be plotting the history of a given WR in Excel on a chart and generating a trendline (and plugging numbers into the formula it creates). Had Bosshardt used this techniques in (2005?), his predictions would almost certainly look better. We have the WR database now, so we can actually try this...
* Ace may be meta-human.
-
I don't see how it would be inherently better in this case to use a broad scientific approach when it has a number of issues to begin with:
-People will eventually stop going for times if they're too fucked and take too long to realistically get, or if they're outright impossible. We have a constantly developing understanding of what might be possible and what isn't, which affects our goals a fair bit. That's why someone like Boss, who played every level pretty thoroughly and was at the upper level in terms of skill and game knowledge at the time, had a pretty shrewd idea of what might be possible. Understanding speedrunning at that level and then making educated predictions is a lot more valid IMO than trying to big brain it.
-The only players of consequence in this model would be the ones that might get an untied, greatly limiting the data set. The number of people who can realistically get an untied has to be in the dozens, and that number isn't guaranteed to stay the same over the span nor will their time contributions, and their goal would have to be strictly to play for untieds.
-Statistical models are great for predicting things like the scale of weather events over a period of time, or how tall a certain species of tree might grow. These types of things are simplistic compared to sports or the stock market, for example, which can also take to data driven analytics. However the competitive ecosystems in the latter two examples assume all parties are following the same goals and putting in their full effort, neither of which is true of speedrunning. We decide what to do in our hobby based on what's fun, and base our rules largely on the community's consensus on what makes for interesting competition.
-Even if the game stays active, untieds in GE are going to become rarer as time goes on. Even now you see level specialists emerging, since the community knows grinding for untieds is only going to become more shit, and many players are attempting to plant their flag before the viable ones are taken. Strat innovations can still be found but ultimately there's so much uncertainty predicting what they could save and on what levels. (Whatever the data predicts, WhiteTed is sure to come up with some mystical strat and annihilate the model. :nesquik:)
-Though obviously certain levels like Train had strategy innovations and many levels saved enough from 2.x to make an untied or even multiple second cut possible, I don't see how any kind of statistical model would be able to predict things like that and a huge portion of Boss's shortfall came from those innovations.
-GE times have minimal "granularity" due to the in game timer rounding to the full second. Nobody is going for the lower decimal TWR unless they think an untied might be possible. If the decimals were tracked it'd probably make for a better application for a data driven approach.
If you want to make a data centered model (and it sounds like you do) then go for it, it could be interesting but I think there will undoubtedly be artifacts predicted that are downright impossible. Such a thing would have value in the context of a discussion but to say it's a better approach than what Boss came up with is pretty fallacious.
-
Circling back. Whiteted thinks my comment about basing predictions off statistics is unreasonable. Care to offer limit predictions?
https://forums.the-elite.net/index.php?topic=21361.msg464059#msg464059 (https://forums.the-elite.net/index.php?topic=21361.msg464059#msg464059)
I'll pred 1:09:5X, because I think the narrowing of focus will see us through 1:10:0X.
At the end of the day it's no use drawing statistics from runs where people aren't using the fastest strats (think B2 SA, Frig A :azn: , S2 00a).
And where they are - take S2 A & Runway SA - what are you actually going to measure from pool of runs? I just don't see how you are going to make any sense of whether the untieds are possible without looking in at the strategies and asking things like "how fast can you do the r-lean?", "how many r-leans can a human tolerate before they commit suicide?", "how common are boosts?".
-The only players of consequence in this model would be the ones that might get an untied, greatly limiting the data set.
A set which you are definitely a member of rhakiath
Strat innovations can still be found but ultimately there's so much uncertainty predicting what they could save and on what levels.
For sure. I think we'll increasingly be able to say with more confidence that certain levels have optimal strategies but atm there's a good few still up for debate.
-
A big chunk of time cuts are done by players who are able to get multiple uwr (for example only 14 players with an uwr have just 1 second to their name)
So here is a list of players who cut more than 1 second from the game and have submitted a time to the rankings within the last 3 years (their "Time Cut" is overall, not just the last 3 years in case it isn't obvious to anyone)
The ranks are based on the Time Cutter Rankings (https://wrs.the-elite.net/goldeneye/timecutter.php?y=all)
| Rank | Player | Time Cut | Last Time Cut | Last Time Submitted |
| 3 | Bryan B. | 2:12 | 09.07.2016 | 11.09.2018 |
| 5 | Karl Jobst | 1:17 | 31.01.2019 | 07.11.2019 |
| 6 | David Clemens | 1:14 | 01.08.2019 | 01.08.2019 |
| 7 | Rayan I. | 1:13 | 26.10.2019 | 28.12.2019 |
| 11 | Ilari Pekkala | 0:33 | 22.04.2010 | 08.07.2020 |
| 13 | Henrik Norgren | 0:27 | 14.04.2020 | 05.07.2020 |
| 14 | Marc Rützou | 0:20 | 01.07.2019 | 01.07.2019 |
| 15 | Ryan W. | 0:19 | 11.06.2015 | 20.07.2019 |
| 26 | Luke Szklarz | 0:05 | 10.09.2017 | 25.04.2020 |
| 28 | Daniel Coelho | 0:04 | 09.02.2020 | 17.04.2020 |
| 28 | Gus Riolo | 0:04 | 21.02.2020 | 12.07.2020 |
| 28 | Ryan Gibbs | 0:04 All on Runway 00 | 10.07.2018 | 20.09.2018 |
| 32 | Luke Pettit | 0:03 All on Frigate | 01.03.2017 | 23.03.2020 |
| 37 | Carl-Magnus Wall | 0:02 All on Surface | 27.01.2020 | 28.06.2020 |
| 37 | Dan Parker | 0:02 | 17.05.2019 | 17.06.2020 |
| 37 | Fernando Almeida | 0:02 All on Surface 2 00 | 05.12.2019 | 05.03.2020 |
| 37 | Hayden King | 0:02 All on Egypt | 20.08.2018 | 03.05.2019 |
| 37 | Joris Quevedo | 0:02 All on Jungle | 04.07.2020 | 05.07.2020 |
| 37 | Philip Bezgoubov | 0:02 | 18.10.2017 | 30.04.2020 |
Take from this what you want, but we really need those top guys who have cut more than 5 seconds to actively play if we want to continue the trend of 10+ seconds cut every year (which didn't happened only once in 2009)