It's been a well-known article of faith among the few people familiar with rocket knight a while ago that kega fusion is a really inaccurate emulator for the purposes of RKA speedrunning. This stems from a run done a few years ago. A player called chum posted an emulated time on the SDA forums w/ video. ShadowJacky's wr at the time was a high 28, and chum got, I believe, a low 29. The overwhelming consensus was that chum's run, which he did in something like 2-4 days of learning, was significantly worse than SJ's run, and should not be anywhere near SJ's time. Therefore, the conclusion went, kega fusion must be an inaccurate emulator. It was probably saving 30-45 seconds over the course of the run. /obviously/

and then nobody played rka for years.

recently a few more people have picked it up. some of that number have used kega fusion, so this issue has come up again. I was watching a stream of AND4H's and made an offhand remark that we should prob just ban kega fusion already. That turned into a discussion about how inaccurate it was, and I decided to do my homework.

homework's here.

What you should look at primarily are the two columns labelled 'diff' (columns E and I). Those values are the # in seconds of whatever section it was I was timing. Left diff column is console, right is emu. So for example, running the intro loop x2 took 114.38 seconds on console (E2) and 114 seconds on kega fusion (I2).

That's less than half a second's difference over 2 minutes. This doesn't add up to anywhere close to how much time we all were assuming kega fusion saved.

The sections I tried timing were sections of the game that required little/no player input & lagged some. Most of them were boss kills to the start of the score countdown. For a couple, I let the boss play out a bit without player interaction (letting the stage 4 boss loop once while standing in the safe spot, the opening autoscroll sequence before you can damage stage 6 boss, etc). And since kega fusion wasn't saving the seconds I assumed it would, at the end I tried timing the longest pure dps-to-an-enemy-with-no-complications section I could think of, the last phase of the gradius boss. I guess I wasn't close enough to it or something on the emu runthrough lmfao

As it stands, I'm limited by the situations I can test. Anything that really depends on player interaction is too variable to framecount like this, which cuts out most of the gameplay portions. (why don't I tas it? because neither kfusion nor console (obv) have tas tools) But as it stands, it looks like kfusion is saving very approximately half a second over two minutes of play. Over the course of a 28 minute run, that works out to 14 seconds. No idea what the margin of error is on this, but it's probably big.

still not sure why we came to the conclusions we did when viewing chum's run. this shit is obviously not saving 45 seconds. It was probably some combination of seeing "learned the game in 2 days" and sub-30 and the run being a lot less clean all come together in a way that psyched everyone out. I can fully accept that I was probably looking for confirmation bias when watching AND4H's stream the other day & saying there's no way that speed would be possible on console. I bet it fucking was possible after all. I'm gonna blame it on the lag feeling more real when you have a controller in your hand, because you're trying to input so you're interacting with it more ;_;

I would also be interested in seeing someone replicate a test like this with bizhawk. But I don't care enough to do that. Also interesting would be seeing someone who's familiar with playing against a set of splits to switch around between console and one or two emus & see if any of them feel off. But again, I'm not interested in doing that either, so... yeah.

A dozen seconds over the course of the game. Do we care to do anything about that? I think it's notable, certainly, but that's not a big enough difference for me to feel an urgent need to ban kega fusion anymore unless bizhawk is significantly closer to console. And as I don't currently know or care enough to find the answer to that question....

I would be happy to do any testing relating to console/emu split differences, though I'll probably never get the opportunity to. Doing it myself is out of the question since I have a PAL console, but on the offchance we meet at an event, I'm down to look into it.