I feel for you, this can be really frustrating. I can't tell you why it segfaulted without a backtrace, however.

As for STK Enterprise, it's not where my run end in the majority of cases (I reset if I don't get a good overall time at various points ; so I reset more often at the beginning when a basket ball hit me in scotland/OMC or if I don't get a good shifting sands ; then the next big reset point is Hacienda).

Having a record time after 3/4/5 races crushed by some mistakes or bad luck is sadly pretty common.

The times I use for my analysis are those displayed on the loading screens.

For the "enter overworld" data point, it's the loading screen just before, well, you enter the overworld. For the times after a challenge (except for the final run time at fort magma), it's the last loading screen before reentering the overworld. If the chest unlock cutscene is displayed, it's the time on the loading screen after the cutscene. If there is no chest cutscene (can happen sometimes when you skip really fast), it's the one after the end race screen.

For example, in my 30m26s run whose video I posted earlier in this thread, the enter overworld time is 1s213 ; the time after Nessie's Pond is 47s503.

It doesn't matter the order you did the segment in, I'll sort it out, I just need to know the total time after each segment with the segment's name.

(Give me also the 30m18s times, then ?)

Upload your 30m11s352 time, even if you want to do another one. It will be nice for my stats tracking.

My run managed to do better than your new improvements.

The video link :

I've set the diffusion to sunday afternoon.

Indeed, it would have fit well into that other one. We don't have much management options, though, no "move post", only edit/delete.

I made a compilation showing the 7 runs I had which were on pace to get the Any% WR and which got destroyed by the dumb randomness of FTL.

New world record ! I've taken some good seconds of my previous time, even if there are some frustrating bits.

YouTube is giving me a "first diffusion" option, which allows to set a moment at which the video will be diffused for the first time, with an available chat. That sounds fun to discuss live how it went rather than just looking at the final time and my graph data and have all the emotion killed before watching the run.

Let me know if you think it's an interesting idea and which time would be suitable for this.

In the meantime, I've done the FTL fail compilation, and I'm going to publish it soon.

(Edit : markdown quotes don't work here ?)

Thank you guys for your answers. I wasn't really seeking advice on what to do myself, rather, I was curious as to what other prefer and why.

It seems that both strategies have their adepts, from "reset at first significant error" to "only reset when it's completely over" to some middle-ground.

@ShikenNuggets and @theripper999 are clearly in the "reset only if there is no chance to succeed" camp, and it's true that a run which began bad can snatch a PB (or even WR). I actually have an example of this, which date from when the time was not very optimized (beside new skips). I was trying to practice a bit before starting serious attempts, and was around 15s behind during the first half. After 15 (of 19) segments, I was 25s behind. But I still ended up snatching the WR by 3s, because my previous time had a poor finish while the new one was solid.

This is not my personal preference. When you come from behind and just catch up at the end, you'll most often improve your time only by a little bit. Unlike Serguei Bubka, we don't get a cash prize for each new PB, and if you know that with your skills, you can do significantly better, you'll still have to try more and more.

@lolman-0027 [quote]My situation may be different from yours too, I run the ILs, where small differences really matter.[/quote] This is not so much a IL thing but a short run thing (see SMB1). We have ILs too in my game, thankfully without luck, and when the run last only 90s, and 0,5s can make the difference (we have the ms time enabled and the difference was sometimes less than 0,1s between two competing time), you're forced to reset a lot. But when you're forced to do close to no mistake to beat your record, there is no real difference between "resetting early" and "resetting late".

@TalicZealot [quote]Only if there is no easy way to practice late game on its own. I'm a big advocate of section practice instead of just doing tons of attempts.[/quote] In my game's case, each segment is easy to practice on its own. And actually, for the hard parts which come in the later part of the run, me and the others who have done some serious attempts have done section practice (and ILs, while slightly different, also act as a section practice).

Nonetheless, there is a difference.

Some segments are "easy" compared to others and not a big practice focus. However, this decreases consistency, and while big mistake aren't a serious concern there, some slight mistakes may slip in.

The other difference is pressure.

When practicing some of the "make or break" parts where a slight error can easily cause a loss of 10-20s, you're quite calm.

When you reach one such part while having done a very good time in the first half of the run, your heart is beating fast and it's harder to keep fully focused. I remember watching some mario game player who had a heart rate monitor displayed along his run, and in the final critical section (which was the hardest for him and where all the previous good parts of the run could be wasted), his heart began to beat much faster. I experienced the same when going for the record in my game, my heart began to beat twice as fast when I got past some tricky parts with a good time but had to not fail in the following tricky parts.

Resetting more often, the pressure is greater because an attempt getting half into the run will occur much more rarely, and when it does, the time is better so there is more to lose.

[quote]If the game is very linear the cutoff would be time-based and you would reset if you are too far behind to catch up and PB by the end. Other games have hard reset points like tight skips. In a lot of cases it would be a mix of both. Also depending on how optimized or not the run is it can be easy to underestimate how much time you can save.[/quote]

Good points.

In my case, I went with a sort of time-based cutoff, though rather than "too far behind to catch up", it was more "not ahead enough to account for screw-ups" (there also were opportunities to get ahead in the first quarter of the run without having to do really too much tries). Though I was going for sub-30 more than for purely a new PB, so it sort of was equivalent to "not too far behind to catch up". But if on pace for PB but not sub-30 after 75% of the run, I didn't reset, obviously.

@HowlingSnail [quote]I depends how optimal your current PB is, meaning how much time you can afford to waste.[/quote]

Absolutely. When it's unoptimized enough compared to your abilities, you can just try a few times and beat it without having to try to get a very good beginning.

In general, I'd be curious if we could develop a sort of dynamic computation of PB chances.

The software would have a list of times on various segments, would generate some probabilities for times based on this, and could take the current time after the last segment to compute. This would be imperfect, because if you begin using a new tech the segments time will be outdated, it wouldn't have a huge sample size, and it would need to value more recent attempts. But having some rough idea, when halfway into the run, of if you've a 3% or a 50% chance to beat your PB would be quite nice.

In a game I run, me and another guy are trying to get a time below 30 minutes, while we have PBs (and WR/former WR) in the 30m20s.

Now, analysis on our best segments show that this is very doable (an ideal run with no mistake and perfect luck would be at around 29m), but it's hard to avoid any significant blunder, and there will always be some amount of unluck (there are small random events across the run, most won't take out more than 2-3s, though in some specific spots it can result in a bigger run-killing loss).

And so this got me thinking : when is the best time to reset a run ?

I should clarify what I mean by resetting "early" and "late".

Resetting "late" means only resetting when there is no significant hope of besting the WR, i.e. enough time has been lost that it would require practically no mistake/bad luck until the end to barely succeed.

Resetting "early" means trying to get a significant advantage in the beginning of the run, resetting if on par/slower than WR pace at some point (or even faster, but not by much), even if there is still room for some bad luck and mistakes in the final part to still snatch a best time.

For some games and categories (especially those with no luck at all involved), if the WR/PB is optimized enough, there is no real difference between resetting early or late, as there is no room for error.

But in others, like mine currently, it's possible to lose 10s in a segment but still have a good shot for the record, so this question can really come up.

I can see some pro and cons to both strategies :

• Resetting late means more practice on the later parts of the run. The sooner you reset, the more often you'll do again and again the same parts at the beginning while doing only rarely those in the middle and in the end.
• Resetting late gives more opportunity to get a record with a strong finish even if the beginning was not that great.
• Resetting early means that less time is wasted in runs which are unlikely to succeed.
• Resetting early increases the chance to have a really good time in the middle of the run, and so leaves room for error, and increases the chance that when a run is finally done to its completion, the resulting time is lower.

So, what are you personal preferences ? How does it change whether you're 5% in, 25% in, 50% in or 90% in the run ? What do you think is the most efficient strategy to reduce grinding ?

30m26s...

Adding it to my graph, the best segments are improved by 9s.

Also, I got an interesting metric about "how good a run would have been without that failed segment". Comparing across the runs, the question is "taking a random segment, what is the likelihood of this run's time to beat the time of another run taken at random ?". A run which is worse in all segments across the compared runs would get 0%, one which is best across all segments would get 100%.

The results mostly match the times, but there are some odd one :

• 32m24s : 25,56%
• 32m11s : 38,35%
• 32m08s : 39,1%
• 31m47s : 51,88%
• 31m20s : 58,65%
• 30m29s : 56,39%
• 30m26s : 68,42%
• 30m23s : 61,65%

My 31m20s gets hit each time a new run is added, as it doesn't have the new skips, but it's still ahead of my 30m29s time on that metric. The fact is that, outside the new skips, it was of better quality.

The other weird outlier is my 30m26s, which is way ahead on that metric. That's because most of the time loss in that run is restricted to 2 segments, and one of these has a new skip so the big failure ends up as better than several non-skip times. Otherwise, as shows the graph, it was very consistent.

I just did a 30m53s with : -10+ seconds lost in Hacienda (a kart got a zipper and blocked my way to do the skip on the 1st lap) -10 seconds lost by failing the skip once in Black Hill Mansion -10 seconds lost in XR591 by failing the wall skip (dont ask me HOW I missed it in the run while when doing the track casually, I do it reliably)

• 25 seconds lost in the Old Mine. I made a small mistake at the beginning. I tried to throw something behind, missed my turn, and collided with an obstacle. And then, I was constantly trolled. Cake, bowling ball, moving mine kart, bowling ball, parachute... It was awful
• Missed 1st try gate skip -Several seconds of stupid time loss in many tracks, with basket balls, parachutes, and small fails of my own (like falling down in green valley and being rescued - 3-4s lost)

It was at ~16m20s after Northern Resort, 10s ahead of the current record.

Now, there will always be random bad events in a run.

But there was really a lot which went wrong, many avoidable with greater caution, and the time still isn't that bad, because there is still a margin to improve on other tracks compared to the current records. At some point, the stars will align.

There is no big novelty. You do some things differently (abysses instead of cornfield crossing, light kart on lighthouse/minigolf/zen garden, some track order shuffling), but nothing which fundamentally changes the time on a given track.

But your latest time uses the new tricks and is overall more consistent that my 30m29s. That's what it will take for a sub-30 : consistency, avoiding any major error.

Here is your time added to my graphs :

The sum of best segments don't change much at this point. I've not updated the previous WRs to have the time at entering overworld, but I figured out it would be better to check the time there so that the section time's is not influenced bat what track you do first.

New Any% world record.

Around 0,3s faster than the sum of best segments of the 5 previous WR, all thanks to the new skips.

30m29s555. Outside of Northern resort and Gran Paradisio, this new time is only 2s faster than the previous one - and I won a few seconds on Shifting Sands, too. However, as the previous record was of high quality with few fails, the same mostly remains true for the new one,

Sub-30 minutes is definitely possible with dedication and luck.

I've coded in the "fast but hard" way to unlock the SuperTux difficulty I wrote about earlier.

The standard way requires getting 250 points (basically, gold on half the challenges, silver on the other half) ; the "fast but hard" way requires beating the SuperTux times while playing in expert or less in 5 different challenges.

For example, currently, the required time in Expert in Cornfield Crossing is 2m45s ; while in SuperTux it is 2m20s. If you play in Expert but do less than 2m20s, you'll have one of the 5 required challenges to unlock the SuperTux difficulty.

All this means that SuperTux%, while of course longer than any%, won't be a grind requiring to do twice all challenges. At the same time, beating the SuperTux time on first try on the 5 fastest challenges you can have access to requires more skill.

With the latest improvements, especially the 30+ seconds gained in Gran Paradisio, the total has slipped under the 26 minutes mark (1560s).

Are there still some undiscovered skips which could allow to gain even more time ? At this point, it seems unlikely, but it may be unwise to consider it entirely impossible.

For some games, the current setup with one full-game lb and one IL lb feels quite limiting.

In the game I moderate ; we have to track multiple versions. They aren't different enough version to version to warrant being considered as different games, but over several versions the differences get quite substantial (and comparing with an old enough version, it gets to the level of a different game in the same game-serie). Only some of these versions are run (thankfully), while the other are misc. ; but even then the IL lb is being limited by having to share levels across all versions, despite some levels not being available in older/newer versions. This will get worse when the next version is released.

We also have a few different type of ILs, but can't specify different rules for each one in the edit game interface.

Not doing a new topic just to post this pic, even if it's about the all challenges instead of the Any% ;

New WR !

31m20s553 :)

I'll upload the video tomorrow, but in short I got a similar time to my last failed run after lighthouse, and got first try gateskip and a decent Fort Magma !

EDIT : video uploaded

The start is much stronger than the previous WR, and the overall performance is very solid, with the only serious mistakes in Snow Peak and Minigolf. They are what separates this new WR from a sub-31 run.

And for comparison's sake :

Another run which has come close, even closer, to beating my current WR. 31m51s670 ; doing the last challenge as intermediate because the game thought it was what I want (I suspect it's a mouse position issue) and relaunching the challenge wouldn't have been worth it.

Once again, the FTL was awful (it was down to pure luck, though I won it), but it's black hill mansion (I was hit by a kart in the 2nd lap, mismanaged and failed the skip, then didn't get back properly on the road at the end of the 2nd lap, so the 3rd lap skip didn't work either) and the failed gate skips (I've switched to a new method, which is faster when done properly (or at equal number of fails), but I lack some reliability yet) which killed the run.

I'm cautiously optimistic that with a few more tries, I'll finally get it to work, especially with some training in the two obvious problematic parts which kill the runs making it past hacienda : blackhill mansion and the gate skip (I had much better reliability when doing my sub-32 attempts, ironically).

I have also beaten my fastest time after Cocoa Temple and Shifting Sands.

Once again, I won't upload the full video because doing so would take 4 hours, but I'll add that new failed run to my previous graph :

I expected Fouks to post an any% run this month, but there is no sign of it happening.

Rather than waiting a few months and having to relearn 0.9.3 tricks, I prefer trying to get "definitive" records while my technique is here.

So far, I did two promising runs which managed to get past the main obstacle in the run, the triple-skip in Hacienda (I decided the skip in Northern Resort to be way too unreliable). In both, I was more than 20s ahead of the current WR after Candela City.

Both ended in a tragic fail, the first killed by the FTL.

The second was killed by a mix of errors and by the game trolling me and throwing me off-balance.

Using my old graph comparing WRs, here is how the new runs (purple and light blue) went :

In that second run :

• I somehow failed the last skip in black hill mansion despite my cautiousness.
• The game decided that I want to play in intermediate difficulty in the gran paradiso challenge. As, usually, you select as fast as possible, I couldn't change it before being in the challenge. So I exited, went back, selected too fast once again, exited, and finally started the challenge in expert
• Because the Old Mine challenge has only 3 AI opponents even in expert, I mistakenly thought I was again in intermediate (thanks to that Gran Paradiso issue...), and lost a few seconds exiting and going back in.
• I somehow missed the gate skip TWICE, each time flying over the wall.
• I lost 3-4 seconds with an accident going towards the final challenge, after the gate skip

Considering everything that went wrong, it's a small miracle that it was still a sub-32 time.

At the same time, I'm certain that with some more tries going at sub 9m50s after Hacienda, one of them will be free of most of these errors... Making sub-31 very possible.

I figured out the easiest target was my Northern Resort time, as even if the skip is difficult there, it was one of the time-trials with the most possible improvement remaining - so beating it would be enough, instead of scrapping a few hundreds of ms on several tracks.

My new time is a 2,64s improvement over my previous record, so this is it, under 1600s. :)

With Fouks recently giving me some serious competition, I had to improve many times which were until then "good enough" to be on top.

With that, what looked like a pipe-dream only a few weeks ago seems really doable now :

I only need to take off 2s somewhere...