[disambiguation]

as someone with no prior opinion or interest in speed-running, this is a surprisingly good watch!

some thoughts

- given the timeline of progress, i guess we might see THE perfect run in the next 12 months or so

- how did they figure out the theoretical best time in the first place?

- is machine learning being applied to this problem yet? might be able to discover some more time saves

[tantalor]

> how did they figure out the theoretical best time in the first place

A rarely used tool is brute-force searching for ideal inputs by making a computer play the game, trying all possible inputs. In theory, this process could find the ideal set of inputs for any game, but since the space of all possible inputs grows exponentially with the length of the sequence, this is only viable for optimizing very small portions of the speed run. Instead, a heuristic algorithm can be used. Although such an approach does not guarantee a perfect solution, it can prove very effective for solving simple puzzle games.

https://en.wikipedia.org/wiki/Tool-assisted_speedrun#Method

[readams]

They don't know the theoretical best time but they know what is possible using tool-assisted speedruns (TAS) using the current set of known tricks. So it's possible there are still tricks to discover that will improve the speed floor, but it's been a long time since anything new was found.

[ajsnigrutin]

> - how did they figure out the theoretical best time in the first place?

Depends on the game...

With mario, if you know the max acceleration and speed of mario, and the length of the level, you can calculate the theoretical minimum needed to come to the other end, especially if no glitches are allowed. If you have to backtrack a bit, you can calculate the optimal path to pass through. Glitches (eg. passing through walls) make this harder.

With games like Zelda, breath of the wild, it's pretty much impossible to calculate exactly, except for taking the current records, finding flaws, errors and/or adding new optimizations, and subtracting the "lost time" from the record.... until a new, better strategy comes out.

[fuzzer37]

> how did they figure out the theoretical best time in the first place?

Generally by trial and error and with a TAS, which is a "Tool Assisted Speedrun". Some people determined what all of the most optimal segments were, then programmed a computer to push the buttons at exactly the right time (down to frame perfect). So theoretically the current best time could be a local maximum, but with how long the game has been out for that seems unlikely.

This also led to a kind of funny thing where some runs were considered "TAS Only", only to be later achieved by a human in a real run.

[to11mtm]

I've been seeing this on some Castlevania SOTN tricks. I'm not sure whether this was done before or after TAS became in vogue, but while Some specific tricks are 'frame perfect', in that game you can use the map function to check your frame and slow it down. Yeah, it's still slower than a TAS, but if the time savings is substantial it's worth it.