Uploaded May 2020 | Updated September 2026, 3 hours ago
Was super inspired by Tooshi's recent stream/video, so figured I'd use my TAS pathing tools to help me find a better route. Pretty tight run overall, there's only about 4 tiles of free movement left. Have a solver going to see if I can find a better route but not expecting to do much better.
Tooshi's Egg vid: youtube.com/watch?v=InZRZLby5mo
Was super inspired by Tooshi's recent stream/video, so figured I'd use my TAS pathing tools to help me find a better route. Pretty tight run overall, there's only about 4 tiles of free movement left. Have a solver going to see if I can find a better route but not expecting to do much better.
Tooshi's Egg vid: youtube.com/watch?v=InZRZLby5mo
![Seanies Skeleton Caverns Challenge Run [TAS]](https://i.ytimg.com/vi/iOkQsfaEzHk/mqdefault.jpg "Seanies Skeleton Caverns Challenge Run [TAS]
I unfortunately ran out of time for this challenge so tried to optimize where I could/ended up having to leave 3.5 minutes on the clock but Im happy with the result. This was a good excuse to flesh out my TAS tools for SC/mineshafts (still needs a lot of work...), mail, some additional overlays, and more process of making a TAS things. Its also my first real test of the SDV v1.5.6 HF#3 TAS build (not a single desync while testing!)
Basic Challenge:
- Given the starter file from Seanie, you get 1 day to prepare to go as deep as possible in the Skull Caverns
- Skull Caverns day can not exceed 21 minutes in length.
Route Description/thought process:
- Drop shafts are fixed by game seed/day/floor so I can precompute the # of floors dropped for each level of the mines
- A floor has a 20% chance of having a drop shaft (can predict upon reaching a floor but cant forward look)
- When a floor is generated, a random number generator is seeded by the current real time in seconds (which for this run is calculated as frame offset from the file seed). This means you can wait to the next second to try and roll a new floor (why I end up pausing a lot).
- Frame perfect laddering can be done at 80 frames with transitions
- Factoring in a tool use (10 frames) and ~2.5 tiles of movement (~30 frames), an average floor is close to 120 frames
- Taking a drop shaft is 120 frames with transitions (dropping 3-15 floors), which means for N floors I can drop, I should be willing to wait 2*(N-1) seconds on a prior floor to ensure that the drop shaft is generated
- I computed the minimum transition path to reach floor 1000 (sometimes this will want to avoid 1 drop shaft to pick up a later larger one, but primarily its chaining drop shafts)
- This isnt an optimal strategy (its greedy), and its not super practical in an RTA run because the clock is running and sometimes you just wont get a shaft in the wait window. In a pure depth run, you could just wait on rolling the next floor which would minimize spent time vs depth.
- Instead I mostly use this path as a loose guide that I try to shadow, in particular getting into sync when there are large blocks of good drop shafts.
- I was able to make floor 1000 in 17:35, which I find pretty freakin cool because pure laddering would only get to floor ~945 in the 21 minute window due to how long transitions take!
Chapters:
0:00 Day 1
4:40 Day 2
Challenge Doc Link:
https://docs.google.com/document/d/1vGDPhW199F9kEuUpa_45C1IuO28zcx9R-C40fEOthLU
Shoutout to Seanie Dew, this was a fantastic challenge idea for the community!
YT: https://www.youtube.com/c/SeanieDew
Twitch: https://www.twitch.tv/seaniedew")
![Little Rocket Lab Chapter 1 Speedrun [20:48]](https://i.ytimg.com/vi/lXm3avD5tjA/mqdefault.jpg "Little Rocket Lab Chapter 1 Speedrun [20:48]
Fun little factory game, the full game I think will be pretty long but just getting started with routing by focusing on running through chapter 1 as fast as possible.
Still a lot of work to be done to clean up pretty much everything (movement is fine, menus are whatever, placement is BAD, and I keep forgetting to fill furnaces first), but the general route is starting to come together.
Basic idea: get the 5 core resources automated asap, sleep to build up stockpile, convert original layout into heavy duty motor factory. Use a standalone assembler2 to handcraft stators/heavy duty motors to speedup the end game")
![Completing the Vault in Under 8 Minutes [TAS]](https://i.ytimg.com/vi/mQPdF1ZyzH0/mqdefault.jpg "Completing the Vault in Under 8 Minutes [TAS]
PHOTOSENSITIVE WARNING: from 4:10 to 4:56 in the vid there is a ton of menu flashing, I call it out a couple times but just in case you read the desc and arent listening. Ive also modded the game slightly so that menus dont have a dimmed background to remove flashing in mail checks.
Chapters
0:32 Run Start
1:06 Day 5 Start
4:10 CONSTANT FLASHING ZONE
4:56 End of FLASHING
8:04 Run End
This is a tool assisted speedrun of the Vault Bundles in 7:32.281! This was a heck of a lot of fun to route and run, using the menu to manipulate global RNG is a totally new tech to explore, and understanding how RNG is updated each frame will make some things like fishing significantly easier in the future. Also first time trying to do commentary, some of the things like the menu manip I dont think I do a great job of explaining given the short amount of time mid-run so:
Every frame RNG is going to advance due to normal game state changes, the biggest nuisances at least in this run were Evelyn walking in circles in her house and that top green slime who kinda maybe wants to mate with the right slime, theyre unsure (actually). Interacting with the menu adds additional random calls on a frame (702 per flip of the npc social page, 1 for hovering over a new BigCraftable like the signs and 2 for hovering over a new Object like a fence post). To be able to successfully manip, I need to set up the RNG state such that I know how many calls there will be during the tool animation AND that advancing the CURRENT RNG by that many calls will trigger a doll to spawn on hoe strike.
Because some random calls lead to branching behavior theres just way too much to track manually, so I wrote a simulator that tries to forward predict game state/RNG, and then a recursive solver that looks to create a sequence of menu inputs that not only causes the RNG to be correct to spawn the doll but also limit the number of frames between tool swings. Given energy needs I dont think you could actually translate to 60k/m real time but its pretty close! It ends up being a super overpowered money strat, the only thing that could maybe rival it early is an insane fishing chest, but I havent been able to find one that compares.
SEED: 1914883252
Cart in this run saves about 3 minutes, the alternative strat I tried for cartless was going to Pierre for the green bean seed/maniping a crab in the mines/summer forage which a rough run clocked in at 10:3x. Most of that time loss is waiting for Pierre to open, the pet cutscene, and sleeping 20 extra days (which at 5s per day adds up!).
Blades Stardew University on Step Manip: https://www.twitch.tv/videos/1108799188?collection=jY8YtGjzZhbwvA
Both of these resources were super helpful in the planning stages of this TAS.
Blades Map Predictor: https://www.nexusmods.com/stardewvalley/mods/6614
StardewSeedScripts: https://github.com/Bla-De/StardewSeedScripts")
