FALLSTRUKTUR - A Speedrun World Record Post
A few weeks ago I saw someone mention FALLSTRUKTUR - apologies but I can't specifically recall who it was, I follow a lot of people who talk about games - a free platforming game about moving downwards.
The mechanics are basic - from a first-person perspective you can run around and jump, with a hefty amount of inertia. Your goal is to get to the bottom of the STRUKTUR, and your main obstacle is fall-damage - if you hit the ground too hard, you instantly die.
On its face, it appears quite Foddian. Like Getting Over It, you make calculated moves and deal with awkward platforming mechanics to navigate dangerous terrain. If you make a mistake, you lose progress. It's hard to play, but mastery is rewarding. It's short, but feels longer when you're playing through the first time.
I don't think it fully matches that genre though - in FALLSTRUKTUR, any fall is deadly and the end of your run. Personally, a hallmark of Foddian games is the sense that a fall might only lose a little bit of progress but in FALLSTRUKTUR a fuck-up is terminal. That lends it a very different tone imo - mastery is almost mandatory.
When you realise you've fucked up in a Foddian game, you look for safety, pray for a landing spot. In FALLSTRUKTUR, you have to catch the fuck-up very early, and often the only solution is to faceplant a nearby wall to scrub some speed before you hit terminal velocity - a tricky skill that requires a kind of mastery. This is often impossible, so the game becomes about quickly recovering the progress you lost, a second kind of mastery. Slow-and-steady doesn't yield much in this game.
I started playing FALLSTRUKTUR on Friday. On Monday I got the world record speedrun.
It's designed to be good
I've been speedrunning to various degrees of seriousness for about 20 years now, and something I always bang on about is that I don't generally like games that are designed around speedrunning - I much prefer games that are designed around general mastery, with speed being one of the ways to express that.
My rough timeline of speedgames looks likes this:
Spiderman 2 (GCN) -> Mirror's Edge -> SM64 -> Super Meat Boy -> Portal -> A MASSIVE GULF OF DISINTEREST -> FALLSTRUKTUR
Crucially, none of these games are speedrun-dominant - playing them fast is often a part of the design, but it's layered on top of a game that's fun at any speed. Movement tech is purely optional, just for the sickos.
FALLSTRUKTUR feels a lot like a stripped down Mirror's Edge, and Mirror's Edge is the game that truly turned my brain into the momentum-conservation-pattern-recognition machine it has been for the last couple of decades.
In the first few weeks after Mirror's Edge launched, I maintained a stranglehold of level speedrun time records on the PS3 leaderboards, unable to find strats online or watch replays, simply seeing someone beat my time by 10s and thinking "shit, there must be something I missed" and going back to craft new theories, new routes, new tech.
Speedrunning is a secondary layer
When I started playing FALLSTRUKTUR, it had a similar vibe. I beat it in half an hour or so, played a couple of workshop levels and started to get a feel for the movement - I posted a video of my PB in a workshop level that really caught my attention, then went back to the main game with the knowledge I'd gathered.
Once I figured out a basic route through the game and executed my first solid run, I got a couple of achievements for beating 2 of the 3 developer's personal best times. Uh oh. There was a third time I hadn't beat.
I must beat it.
10 minutes and a handful of PBs later I beat the 3rd developer time of 1:06.13 , and even better my route could easily be completed in under a minute. Once I achieved a 59.xx it was crunch time - am I going to run this seriously, or just play for fun?
I opened up the speedrun.com page and... 40.65s??? That's absurd.
I didn't think my route was perfect, but surely I can't be losing 19s on execution alone?
I made the decision at this point that I wouldn't be looking at any records until I felt I had truly maxed out my own routing abilities. I was sure to miss some stuff, but the fun of this experience wasn't in being the best, but in doing my best. If that meant I would be limited to outside the top 20, then so be it.
But now speedrunning is all that matters to me
I don't have any recordings of my runs from this time, which I regret - it's amazing how much you forget about old routes once a new one clicks.
The first run I have a recording of is a 52.22, 11.5s shy of the WR but 7s faster than the run that inspired me to get better. I'd figured out that the fastest meta-route was probably to take a loop around the side of the map and into a pipe, rather than through the centre, and this cut a load of time and also gave a really natural path downwards. Looking at this run now is genuinely cringeworthy - I was putting up with so many slow mistakes, and my routing was incredibly cautious.
I take a lot of "safety bounces" - where I'll take a slow drop to a hard landing to allow a large bounce and plenty of time to maneuver around obstacles. I even make a bunch of jumps because I haven't tested the fall-damage of a section yet. It's incredibly unrefined, uncalculated.
I very rapidly improved from here - 2.5 hours later I had a 48.40 using the same fundamental route, but some more confident spacings on jumps and drops.
Later that day I came back and put in a fairly long session, around 4 hours, just getting to grips with the map and trying out variations on different sections. It's a hard game to practice, as death means a full restart, and at this time I hadn't figured out a way around this.
Over the course of the evening, I found and refined a new route for the last 20s of the run - from the green pipe to the ending. Dropping straight down from the pipe with an edgebounce felt slower at first, but this section is 80% about going down, so cutting out as much horizontal travel as possible while maximising fall-speeds paid real dividends.
This route also had a bunch of areas for micro-optimisations, small bounces here and there that could save 0.2 to improve a mediocre run, massively boosting my completion rate. This increased completion rate meant I was getting a lot more practice of the whole game, and could risk more to learn the limits of the fall damage.
It was also around here that I realised I could make "custom levels" that were just copies of the main level, and move the player start position to the section I wanted to practice. This was game-changing, and allowed me to figure out a 90%+ consistent end strategy, further improving my completion rate.
It culminated in a run that kind of ruined me - a 42.25 that felt basically perfect - every small optimisation landed beautifully, and no nasty surprises in the ending.
I played for an hour or so, and I just couldn't compete with it - run after run of high-42s and low 43s, until eventually I scraped off a couple of tenths with a 42.00 - this run was equally annoying, because it felt slow everywhere, except it clearly wasn't - it was faster than the "good" run. Then I couldn't improve.
Another hour and no improvements.
This is when I started to do some analysis.
Microoptimisation
I'll save the details, but I found that certain strategies I'd started using were equal or slower than my old strategies, and worse that they set me up with worse momentum for future movement.
The key example of this is the pipe-to-pipe jump around 20s in - I had been using a complicated strat that landed on the edge of the 1st pipe, jumped over, bounce on the opposite corner and ricocheted off the 2nd pipe. This was much more difficult than simply landing in the 1st pipe quickly and jumping to the 2nd for the ricochet, while not being any faster to that last ricochet. Timing being equal, I went back to the more reliable strategy.
It turns out that the quality of the ricochet depended hugely on the angle of attack - jumping from the pipe gave a higher angle of attack, making the edge a much smaller target as I'd be coming at it from above. With the complex pipe bounce strategy, I was travelling much more horizontally so there was much more margin for error in getting the kind of glancing blow that results in a fast drop.
I had accidentally traded consistency of speed for consistency of survival, and had hit the limits of what that survivability could bring.
There were a handful of other discoveries like this that each brought a bit of speed to the run - "more consistently saving 0.3s" was a pretty standard analysis outcome. It wasn't a guarantee, but it was more likely to be fast.
In the end, it wasn't the timesave from the pipe bounce that gave me what I needed, but the modified angle of attack - with a stronger downward trajectory I could take a very weird line off the sculpture below and scramble down some rocks to the same target polygon - essentially trading 2 hard bounces for 1 hard bounce with a slide and a more direct line.
Getting teh urn
At this point I'd done a couple of hours of strat hunting, I'd made some guestimates of how much time my 42.00 PB lost to optimal execution (around 0.75s), and I'd improved my route by what looked like roughly 1s. My PB was behind WR by 1.32s. I had what I needed, surely.
Just need to execute a flawless run.
I figured I'd creep up on it - it had taken me 6 PBs to go from a 44.98 to a 42.00, so surely it would follow a similar progression to take 50% again off that?
My first PB with the new strats came about an hour later - it saved 0.68s, a 41.32. It had a good start, but a terrible pipe ricochet - all the time saved on this run came from the new pipe-to-polygon route. I know this for a fact, I have the spreadsheet to prove it.
| Of course I have the spreadsheet.
At this point I was in flow, and could tell I would get more runs. Half an hour later, I got a kind of shocking new PB, shaving 0.17s off for a 41.15. What was shocking was that it had a terrible start falling half a second behind PB, but with a better pipe bounce and a better line through the scramble it was able to steal the win at the end.
I didn't really have time to process this before another good attempt happened.
5 minutes later, I had another good start. At least as fast as the 41.30.
The pipe bounce was fine, but it threw me off to the left - I adjusted best I could, landing right on the edge between polygons and got thrown hard in the direction I wanted to go.
I landed perfectly on the edge of the next sculpture, skimming with another low trajectory into the rock scramble.
Low bounces, decent scramble, fast line out of the rocks.
Clean bounce.
My mind has gone blank.
Into the ending slide. Just need to trust my hands to do the ending strat.
Three. Clean. Bounces.
40.59.
Just like that.
Post-run Research
Thankfully no dramas - my recording was good, submitted it and it was verified by the previous WR holder.
I wasn't going to pass up the chance to steal strats now that I'd gotten WR with a near-perfect run, so I opened up KeyGo's 40.65 and sure, they use a slightly faster start than mine, maybe 0.2 but... man, they do the whole second half differently.
It seems to be the standard route, similar in path to my one from before I found the dropdown-bounce-scramble at the pipe but muuuuuch faster than mine - I don't think I'd ever have found this.
What it really allows is a very natural transition to a very different end through the mushrooms - sliding down 2 layers above mine and making an insanely fast bounce down off the obstacles to the ground. It looked very hard, but when I timed it against my ending it was clearly faster - potentially by over a second.
A second faster than 40.59 is... well, that's the sub 40. You can't stop hunting that close to a sub, not when you know a way to go faster.
So back to the lab I went - I really struggled to learn this ending. I'm not sure if it's because I play on controller so can't vector my air movement as effectively, or if there's a nuance I'm not noticing, but I went from a 90%+ consistent ending to less than 20%.
After an hour of tinkering with different approaches and backups, I went back into the grind and... struggled. Hard. How hard?
After a messy, slow, but successful run with the new ending I had a 40.30, followed by 7 failed runs in the next 2 hours. That's 7 runs that were on sub-40 pace that died in the last few seconds of the run. Far from the most devastating streaks in speedrunning history, but painful nonetheless.
Based on the pace of my best near misses, a 39.5x was about at the limit of my time distribution - a lot of them were on pace for 39.9x if I nailed the ending.
Then it just kind of happened - a run that didn't die.
The worst part of this run is that I safed the end. The slide down the last polygon loses 0.5s vs going fast, and the section from the first mushroom bounce down into the slide to the very last bounce loses 0.58s to my 40.30, before getting the sweetest skim on the last mushroom I've managed to achieve.
It barely slows down - I don't get it tbh.
What next for FALLSTRUKTUR
I'd love to see that time loss at the end and be motivated to improve but I'm honestly not.
I don't relish the idea of pushing that hard to match a run that good, only to have to fully YOLO the ending and hope for good bounces. Not without much more reliable muscle memory for the ending.
I honestly found running this game a bit intense.
I find it very hard to maintain a relaxed posture when doing hundreds of attempts of instantly-resettable precision parkour and my neck and shoulders haven't been my biggest fans lately.
What I'm far more motivated to do is find better strats.
I know I can improve my starts by about 0.2 in the first bounce, and my absolute fastest recorded start to the bottom of the Village (timed at the bounce after the wooden platform) was 11.78s, 0.2 faster than my record. So that's probably 0.4s on a fully optimal start.
I've tinkered with some other strategies to reduce airtime that can net 0.1-0.2s each reliably, but they're tricky. I want to talk a bit more about them in a separate post as it's a bit technical.
As it stands, a really impressive run by a dedicated runner could easily be a 38, but I'd like to lower that theoretical limit further, even if I won't be the person to execute it.
The Twist Ending
So I wrote this post, figured I'd want to add a bunch of links and clips to it for illustration and so chose to do that tomorrow. In the meantime I went strathunting and I, uh...
I found a 0.7s timesave.
Then I tested it out in a run and got a 39.44 that was very mediocre...
So I got back on the grind and squeezed out a 38.98, barely clinching the sub. I'm much less enthused about this run because it's basically just the last run but with the new trick - I tried a micro-optimisation early and it cost me time 0.1s, and had a slightly slow rock scramble, but I made most of it back in the ending.
I'm very happy to be the person who got the 38, but I definitely won't be the one who gets a 37. Really want to find the strats that make it possible though.
Conclusions
What I want for people to come away from this with is that speedrunning rules, mastery rules, and I think the purest joy comes from enjoying a game so much that you just want to keep playing it more and more. That's the experience I had with FALLSTRUKTUR, and I want to encourage people to find that in whatever games they love.
I think most people would have honestly been discouraged seeing their proud PB was 50% slower than the world record, and would either have looked up the "answers" or given up on the spot. I kind of hate the general culture of speedrunning as "content" that places "abnormally gifted" people out of reach because they can devote dozens of hours to a craft and get paid in the process. You just gotta play the game.
To be honest, I would have speedrun this game whether a leaderboard existed or not, but I probably wouldn't have pushed it this far without a target. I was encouraged by a friend to push for WR because we thought it'd be funny for the washed SM64 guy to come out of retirement and drop a record bomb on this little community, but honestly once I was getting close to the line I was just feeling passion for the hobby. I want people to feel that whether they're shooting for WRs or gaining and applying any level of mastery in whatever way interests them.
Getting good at things feels good.
Even if someone else is better at it.