Commentary by Fx, Plexa (WR holder), and Blueser (#2)
Why Japanese:
Aside from the generally faster textboxes, JPN has multiple advantages over English:
- Cutscene length being shorter, this affects the RNG in several places. This works out in our favour.
- RNG will pause when menus are opened on the English version. This is a problem when you have to wait for RNG to advance anyway, because you're stuck doing nothing. In JPN, the RNG will not pause on open menus, which allows us to fill the waiting time with menuing instead.
- Out of Bounds movement is different (sometimes better, sometimes worse). This saves time overall in JPN.
- Venus1 (one of the late-game dungeons) lacks a particular glitch that's present in the English version, but performing the glitch on English to skip an encounter wouldn't save time over playing through the dungeon the normal way in JPN.
For general concepts I'll refer to the previous submission I made a few years ago. Most of the basic logic still applies. That out of the way, I'll summarize the major differences and how we save so much time:
First of all, we run on a more accurate emulator with mGBA as opposed to VBA-rr. This doesn't exactly save time, but it's a nice improvement nonetheless.
General execution was vastly improved compared to previous, largely thanks to TAStudio making it much easier to optimize things like textbox clearing or menuing.
Overall strategy:
We took a hard look at what is -truly- required to complete the game, taking an extremely greedy approach to initial theorycrafting and routing. This caused us to decide on completely skipping:
- all Djinn of the Jupiter element. They all cost too much time to justify collecting them for how small their boost for bossfights is, respectively.
- Gust timecost: 1min
- Breeze timecost: 1min20s
- Zephyr timecost: 2min30sec
- Smog timecost: 15sec (While Smog would technically be a viable option, when all other djinn of the same element are skipped, he becomes useless on his own, and so even the 15 second investment can't be justified. The total time lost from skipping all of these combined is in the range of 30 seconds to a minute (hard to accurately estimate, but it's such a massive gap to the time saved that it's obvious we should be skipping them)).
- Broad Sword. This comes with skipping Gust, as we no longer enter Bilibin at all, and so it would take far too long to go there just to buy the Broad Sword. We get the Elven Rapier instead, which is both stronger, faster, and lets us manipulate the GRN for free upon collection.
- Elven Shirt. This item boosts Agility of the wearer by a factor of 1.5, but this bonus is relatively useless to us. The time investment for the item comes up to around 27 seconds, of which we lose roughly 10 seconds overall due to lacking the Agility boost. Saving time overall.
- Assassin Blade. Used by the previous TAS to one-hit-KO lategame encounters. However, this strategy was found to be slower than manipulating for encounters that allow us to run away after a failed attempt instead. Even though we're so underleveled that we will always fail to run on the 1st turn, we can effectively manipulate to always run on the second turn instead, saving time over collecting the Blade.
Other things we considered but decided against:
- Skipping Scorch in favour of collecting Smog to have the same total number of djinn (this matters for Colosso setup), since the investment for Scorch is 45 seconds, but Scorch quickly pays off and in fact starts to save a lot of time over all other options (such as Ember or Squall).
- Collecting Ground before Altmiller Cave. With improved RNG manips, we can instead save time by delaying the collecting of Ground until after we complete the dungeon.
Major route differences:
- The general structure remains the same. We visit the same areas in the same order. However, since the time the previous TAS was made, a new major glitch was discovered that allows us to skip the Manticore bossfight in Lamakan Desert. This saves a reasonable amount of time.
RNG manip:
- We ramped up the extent to which we manipulate both the GRN (General-) and BRN (Battle Random Number) systems. We now manipulate 1 in several thousands, multiple times in a row, throughout the run. There are sections of gameplay where the chances of getting the result we want is 1 in millions, Certain boss battles are 1 in several millions by themselves, etc.
By taking the manipulations to this extreme, while also looking much further ahead in the run (up to 1 hour of looking ahead during the lategame) we manage to save a lot of time on things like random encounters (by skipping many of them where the previous TAS did not), or boss battles.
- Along with these encounter manips, we can also manip who takes damage (and when) more effectively than before, which we can use to properly spread our damage intake and heal it up later on when we need to drain Isaac's Psynergy to perform Retreat glitches faster.
Resets:
- We've cut out almost all Hard Reset points. We only Hard Reset when absolutely required for an Out of Bounds glitch that needs clean RAM. When we don't have to clear the RAM, we only Soft Reset instead to maintain our BRN and continue the RNG manip.
Puzzles&General Movement:
- We've improved the movement throughout the world from before, with more accurate movement comes fewer total encounters (in some cases going through entire zones or dungeons without any encounter whatsoever). Puzzles also fall under these optimizations, where we tried to take a fresh approach to each puzzle and see if we couldn't find a better way than what everyone had been doing for years. This led to discovering some minor improvements here as well.
That's about it for now. If you have specific questions feel free to ask.
Samsara: Judging.
Samsara: Accepting as an improvement to the published run. Tier remains Vault due to a lack of audience feedback.
feos: Pub.