We're down to our top three here. And it is a really strong performance from all at this point. Today, to survive, each contestant will need 7 different achievements at 10.6 ratio, which is quite impressive, particularly since the past few days have already needed 7 at 10+ as well. Really, the difference between 10.0 and 10.6 is inconsequential at this point. As a comparison, I have a total of 8 achievements with a 10.0 or higher ratio in my entire gaming career. These guys all basically did that yesterday. And the day before, and they're going to do it again today, because I do not think we're done here yet.
We have 49 days in the books, and the trendline has had me guessing "55-60" for a while. But we are now finally getting to the point where a sub-55 day contest is plausible, because we're at the top-3. Not only is that a psychological break-point, that's also a prize break point. Any one of the contestants could say at any time now, "Screw it, I'm in the money, I am not killing myself for that little bit of more prize and prestige." But as long as they want to keep going, as always, this will continue for a few more days at least.Here's another interesting little fact. This is the first Leap Frog that hasn't had a 0-elimination day*. If we're going to go ~55, we're going to see a number of zero elimination days. Zero elimination days are at the end usually anyway, and they have been an indicator that the contest difficulty hadn't ramped up enough from day to day.
This time, there is no doubt that the difficulty is ramping up. So if we get no zero elimination days, this contest is over tomorrow or the day after. That would be 51 days at the longest. Still in my initial wheelhouse of prediction
, but not what the mathematical models (as flawed as they are) predicted. And that's simply because even though exponential decay is a strong match, it is not necessarily a good predictor.NOTE
: I screwed this up pretty badly. Thanks to JimbotUK
for catching my error. I'm redoing the graphs and everything as I type, so... things will change. The analysis largely remains the same, but some of my statements are off. I've made a bunch of corrections, but it's not as clean as I would have liked anymore.
So, time for some graphs. This remains my favorite.
Projected Leap Frog contest length (days) based on exponential decay formula for croaks that day.
I've shown previous versions of this graph a few times. It has a few interesting peaks. Day 1 is so low because so many people croaked on the first day, as I called out in my first blog on this year's Leap Frog. Then for the first two weeks the croak date rises because as a percentage, fewer people are croaking as it goes on. The difficulty just didn't ramp enough early on to really shake anything more than the casuals out. Then at around Day 14 we get our first maxima. The exponential decay formula at this point projects an 85.9 day contest, which is just insane. But this is before the contest really starts getting hot. Day 15 drops significantly as a third achievement is added to the requirement. But Day 23 was surprisingly strong as well, projecting an even higher 86.6 day contest. That fourth week shows the grit of the generic hardcore achievement-hunting community as the trendline doesn't really move, but after Day 27, the inevitable decline sets in. Five achievements per day that next week all at a high ratio shows that even the hardcore have a breaking point. Except maybe our trio of finalists.
So, why do I always think of exponential decay when these contests run?
Gee, I don't know.
Here, let me put that on a logscale.
Exponential decay graphed on a log scale should be linear.
Our linear trendline has a R^2 of 0.9663. Do you know what that means? Time for one of my favorite nerdy images of all time.
There is a 95% chance that I am 20% cooler than this.
Trendlines and confidence aren't exact matches, but I can't resist.
OK, one more fun graph before we move on to our other contest news.
I've labeled a couple of particularly interesting days to talk about.
Day 1 (373 eliminations) I've talked about before. No need to really repeat anything. Day 8 (91 eliminations) is the first day with 2-achievements required. Day 15 (61 eliminations) is the first day with 3-achievements required. But this weekly peaking did not continue. Day 22 was tame, even fewer eliminated than Day 21.
Two more significant data points on here that I have labeled are Days 23 and 24. I had an asterisk up above, this is my asterisk. There were no zero-person elimination days so far this LeapFrog. Except Day 23. Day 23 is an odd one because the contest was extended. Basically, Day 23 was two days long and every day after that is Day=Day-1. In spite of having an extra 24 hours to earn their achievements a surprisingly spiky number of contestants went out on Day 24, now new Day 23. This was likely due to a comfortableness with having gotten 4 achievements once per the requirements.
Day 29 was lower than Day 28 (27 eliminations) as well. Even with accounting for the "extra day" none of these fit the spikiness of the new achievement patter So after three weeks, it was the ratio that was the bigger determiner of who got eliminated when rather than the spikiness of the achievement jump. Obviously, each additional achievement stuck and made each day afterwards harder, but that spikey jump in difficulty just didn't matter as much as the number of contestants waned.NOTE
: I didn't update this graph after Jimbot's correction. The extra zeros don't actually impact this one.
Finally some discussion on this. Do any of these graphs point towards evidence that could be used to improve the contests for the next round? It's hard to say. The contest team has a tough job of trying to make a contest appealing to the casual fan while also being interesting and competitive enough for the hardcore fan. I can identify a few things that might make the competitive side more interesting. E.g. starting at 3 achievements at 1.0 ratio might be a more interesting contest for the competitive people. Since the early achievement jumps were largely knocking out casuals, starting higher earlier gives more delineation between different levels in the competitive group. For the casuals, though, the biggest thing remains how to handle the 373 on Day 1. Is there more that can be done to increase participation among those who sign up to participate? Does removing a badge for those who fail to at least earn a single achievement help this? Are the registrations periods too long and let people forget that they signed-up? Are there insufficient reminders going out before the contest starts, or during/on the first day?
I don't think there are easy answers, but having all of the data on hand can help make those decisions when they are finally made be more informed.
The UHH is back for a third round, same as Leap Frog. It's harder to do graphs for this one because the big data isn't as accessible. There are no eliminations right now, so I can't graph eliminations, and when they come, they're perfectly scheduled in bracket form. So my analysis will be slightly different.
I love the UHH format. It gives me extra incentive to play my old games that I've left to languish and it introduces me to new community members, usually less vocal ones, who have similar game collections as me. Therefore, it's a reasonable assumption that they are, like me, pretty cool people. I've definitely met new friends in the previous UHH efforts, so I am always excited to see who I get paired up with next.
For Round 1 of the Qualifiers, we had a bit of a SNAFU, though. The matchups weren't done according to the rules; they were including games we had excluded legally. Therefore the matchups were re-run on the quick. Because of this I see a lot of lists that have multiple achievements from the same games. Let's analyze this.
The first step of matchups is determining best matches. This is an incredibly difficult comparison. Let me check myself vs. a random competitor. Compare games that are available for each of us, compare positions and achievements in each of those games and then compress that down into a "match rating". Now I have a rating integer for one possible matchup. Repeat that for each of the other 1670 potential matchups until I have a rating integer for each. Now I'm done... time to do the 1671 other people in the contest until everyone has an integer for potential matchups. And this has to be done fairly fresh, because the game collection is constantly changing. Even if we can't take games out by exclusion, we can still complete games. We can start new ones. And even the games that aren't removed or added are subject to change. We'll earn achievements every minute of every day here. This is a complex process with 1.4 million moving parts. As a result, it gets done early so that the hard part of the math is ready to go.
So the best matchups were already assigned. But the games permitted got screwed up somehow. So my opponent and I were the best matchups based on games we maybe can't play. But are we based on the games we actually can? No way to know. My matchup is still pretty good, but because Rich was dealing with a time crunch and needed to make some potentially sub-optimal matches work better, I suspect that the screws were loosened on the achievement picking algorithm from the matchups to allow for less variety in the name of preserving the hard math already done, even if erroneously.
So, here's my Round 1 matchup.https://www.trueachievements.com/event/UHH3/matchup/8815
4 achievements from Ori and the Blind Forest
3 achievements from Don't Starve: Giant Edition
2 achievements each from BattleBlock Theater
, MASSIVE CHALICE
and Castle of Illusion Starring Mickey Mouse
That's 13 achievements in games with multiple achievements, whereas in all of my previous UHH lists (7 of them) I've had a total of zero achievements from duplicate games on that list. What does that mean for me?
Well, it makes my list a lot more attainable, actually. I was able to knock out four last night by just playing one game for a bit. It took longer than I expected, but it wasn't particularly hard. That lack of "hard" I think will result in higher scores overall this period than we'll see next period and going forward when the algorithms are fixed. If there are 20 different games... it's sometimes hard to even play 20 different games in the week. Certainly, I want to take advantage this week and earn as many of the 'easier' points as I can because I'll need those points to qualify for the KO stage, if I can make it that far. I can't count on any free points because my lists tend to be pretty tough.
Round 1 has been a barn burner so far. We're barely 24 hours in and already we have a 20-achievement earner. Congratulations once again to JamP0und32
. It seems such a small thing, but the fact that he does it with basically every single list and in a record amount of time is a testament to the dedication.
Stats-wise, we're seeing a lot of GwG titles. This is, of course, to be expected. As a 'free' game to everyone with Gold, which is basically everyone in the contest, there are a lot of those achievements floating around in our pools. And they're also some of the games most likely to be unstarted, which means that certain lists have early game achievements from those games more than from other games. But for all that, it's not a GwG game at the top of the list. That belongs to Rare Replay
, though it's friend in the "achievement overload" department, Halo: The Master Chief Collection
, is on the list too. Not currently present is Killer Instinct
, but I'd expect that to show up later this week, or later in the contest at least.
With four weeks of qualifiers, this definitely feels more like a community engagement event than a competition right now. But once the qualifiers are done, I'm sure the knives will come out.