Spelunky is a video game. It’s pretty fun, and pretty difficult. I played through Spelunky classic and it took hours upon hours to even get through the game. It’s also procedurally generated - which means each time you’re crawling through a new random dungeon.

Spelunky was released in 2009/2012, and people are still trying to break it. Here is a high score tracker - note it’s been updated 9 times in Oct/Nov 2018.

And these high scores are not just “I managed to do things 0.01 seconds faster through quick reflexes” - these are “I discovered that, if I fire the shotgun a bunch, it rearranges what layer gold spawns on in the screen, which means I can use a glitch to collect each gold piece twice.” It’s like if Usain Bolt spent years practicing weird things on the track, and then discovered if he walked on his hands he could finish the 100m dash in 5 seconds. More details in this fascinating only-semi-dense article.

(For an example of one of the tomes in the Library of Spelunxandria, maybe try The Spelunker’s Guide to Render Dupe. More general info in the Wiki.)

Why do people do this? How do people do this so well?

A few reasons seem obvious to me (caveat: all just hand waving here):

Spelunky has intentional depth

Some credit has to go to Derek Yu, maker of the game, who has inserted some really-fiendishly-difficult challenges/easter eggs, like the Eggplant Run. This encourages exploration.

Spelunky has accidental depth

Some of the “glitches” in this game are the result of programming shortcuts. For example, the Ghost is supposed to be unkillable, but in practice it just has 9999 HP. Also, nothing’s supposed to hurt the ghost, but maybe accidentally, lava deals it 99 damage. Therefore, you can kill the ghost by making it go through lava 101 times.

Spelunky offers immediate, deterministic, reproducible feedback

If I had to pick one, this is the biggest reason. In a biology experiment, you have to wait a week for your cells to grow - so if there’s some complex thing that happens only when 100 variables are all just so, you’ll take a million weeks to get there. But on a video game, you can very quickly test your hypotheses.

There are low stakes to experimenting in Spelunky

Worst case, you just wasted some time.

Spelunky can be easily internet-collaborated

Like, if 0.0001% of any population will love Thing X, that’s 1 person in a city of a million people. But, it’s 7000 people in the world! If they can collaborate, great. It’s not easy for them to collaborate on, say, ending homelessness. But it’s really easy to collaborate on something that’s on computers and offers immediate, deterministic, reproducible feedback.

Most problems are not like this

I guess that’s all. My brain keeps going to “… this is amazing! can we build this kind of arcane tower in other domains?” but that way lies Gamification. Basically: hard things are hard.

Appendix: “who are the Pannenkoeks of other video games” - Pannenkoek being a legendary Mario 64 researcher.