Messages

This is probably nothing, but it's worth pointing out in case there is some crazy thing that's useful resulting from this. Video: https://www.twitch.tv/videos/157318168?t=23m40s

I have the following codes active:
02030E3C:00000000
02030E40:00000000
02030E44:7FFFFFFF
02030E48:7FFFFFFF

I know this causes weirdness on the world map, but I've never seen it cause huge weirdness like this before.

Basically, accessing certain doors OOB will black out the screen and the game will start writing values to the games memory. It basically filled up a lot of the 02000000-02000300 section as well as writing a ton of sprite data (tracked by the nonzero tile thing in my script, lower the number = more sprite data). Eventually the game will load the room I was supposed to end up in.

The values of memory written change depending on the coordinates of the door I access (not the destination of the door) and solves the puzzles in the game as you would expect with values written to those addresses.

I'm wondering why this bizarre behavior is having, and whether there's any idea as to how this may be replicated without the camera lock codes.

Hahaha this is amazing. As I mentioned the stuff I've found with this is usually useless, amazing that it just is coincidentally the sol blade.

Yeah so we know retreat chests is a thing. We know that interacting with psynergy stones and/or mimics is equivalent to interacting with chests so far as the retreat glitch is concerned. Why we get the Sol blade in Airs Rock, however, is unknown. Other instances where I've gotten glitchy text I've had the following items given to me
- Long Sword
- Silver Blade
- ?
- "It looks like something is buried there maybe I can dig it out" (and scoop does nothing)

Nothing as gamebreaking as the Sol Blade

https://www.youtube.com/watch?v=AIdt53_mqXQ

Why on earth is the item pulled by the retreat glitch the Sol Blade?

Bonus: you get different garbage text by talking to the stone from different directions.

Ah yup I get it now. Curious that some dungeons have map code files split and others not.

The retreat glitch still works, just the every door sends you to the entrance of the dungeon (where you would have been sent if retreat was successful).

If I'm reading the editor correctly, I see the following exits; door 1 -> world map, door 2-> door 2 of 0xFE, door 3->door 3 of 0xFE, door 4 -> door 4 of 0xFC

This doesn't indicate that there's anything funny going on. Although I'm comparing it to another room which has a "madra warp" (felix gets sent to a default position on the world map) to see what is different. Along these lines, it's probably worth clarify exactly what is going when these madra warps happen.

One example is the 0xB door in room 0xCF (Tundaria Tower Interior). Door 0xB in map 0xCF gets sent to door 2 room 0xCF. If you retreat warp through the door you get sent to madra. Looking up Room 0xB relative to map 0xCE (which is where retreat would send you) would make the game read the blank code 0xFFFFFFFF

http://imgur.com/YNQW8RG

Does this mean that the game trying to send you to door 0xFF in area 0xFFF and this sends Felix to the default position on the world map? Or does the game try to look up Door 0xB in Map 0xCE, find nothing, then default you to the default position?

I noticed that the Jupiter Lighthouse stuff is missing any of the 0xFFFFFFFF separators, hence these questions.

EDIT: the first room of Jupiter lighthouse for convenience https://gyazo.com/bc4cb999812c43654b92e1cd6273409f

Hey everyone. Jupiter lighthouse is known as the most unbroken dungeon in TLA. Everything we've thought of has failed. I took another run at the Lighthouse today and made a curious observation concerning the retreat glitch.

Normally the retreat glitch works like this. After retreat has failed the game thinks you are in the room that you would have been sent to if retreat was successful (the 'retreat room'). Example - entering Airs rock through the second entrance makes the retreat point Area 55 Door 10, if you enter another Area of the dungeon (say Area 58) and use the retreat glitch then the Area is reset to 55, while you're still physically in Area 58. In this glitched state walking through doors loads the corresponding door in the retreat room. So if you're in Area 58 and walk through Door 3, instead of being sent where this door sends you instead you will be sent to where Area 55 Door 3 will send you.

Hopefully that's not too confusing.

In Jupiter Lighthouse this doesn't appear to happen. Regardless of which door you access, you always get sent back to Area 251 Door 0. This is the same location you'd be if you'd entered the dungeon from the overworld. I'm not sure if it is Door 0, but it certainly seems like it is.

I'm just wondering if there's anything special code wise surrounding this lighthouse that might explain this. I know that encountering Alex resets the last visited Sanctum to Contigo, but beyond that I don't know of anything. 

I called it a slide because it's literally a slide :D https://clips.twitch.tv/tl_plexa/BeautifulFinchKappaWealth

Weird.  I'm not sure why it's letting you slide into the wall, but I do know for sure that movement in Golden Sun is calculated in 3 Dimensions so as soon as you're moved onto the wall tile you're pushed up on the Z-Axis (which visually overlaps the Y-Axis, I play with it some in this video) to the height of the wall.

Yeah, I think its the same as this: https://www.twitch.tv/tl_plexa/v/50609265
Although in that instance the game decides to correct my position for some reason.

@Fox, I know collision is variable (which honestly is one of the most annoying things in this game since it prevents precise movement) but how exactly does it work?

For this instance the slide is the unusual behavior which I've only seen at Lemuria/Tolbi Fountain and with buttons e.g. like in the video above. I haven't seen a cutscene interrupt it until now. I'm going to play around with it a bit and see what else I can find.

https://clips.twitch.tv/tl_plexa/CourageousTigerHassanChop

https://www.twitch.tv/tl_plexa/v/115272846

Mostly this is just for curiosity. My theory for why this is happening is that this is the same as using a switch to interrupt movement and weird sliding happening (you might recall it from Lunpa Cave). Instead of the switch interrupting movement it is the sand rising cutscene and if you're in the right position and/or in the right part of your movement animation you will slide. If you happen to slide into a collision seam or something like that in the right pixel/subpixel then you can go through and clip like I did in the clips there.

If this is the case it might be possible to use elsewhere, but I'm not sure.

Would have been about RNG stuff, we knew nothing back then! xD

I just went through and checked everything by having my script print intermediate variables and whatnot; I can't see any issue with what's going on. Maybe I just misread errors earlier on... can't seem to replicate the behavior anymore =/

My script is returning a lower rate of flee success than what the game is - afaik it's only happening after the reunion. Any idea what is wrong?

Code Select Expand

if memory.readword(0x02000420) == 0x1FE then                 -- if in battle then do the following

if el4 ~= 0 then                                 --Hacky way of getting the enemy level; doesn't work for 5+
ela = (el1+el2+el3+el4)/4
elseif el3 ~=0 then
ela = (el1+el2+el3)/3
elseif el2 ~=0 then
ela = (el1+el2)/2
else
ela = el1
end

pc1 = memory.readbyte(0x02000458)
pc2 = memory.readbyte(0x02000459)
pc3 = memory.readbyte(0x0200045A)
pc4 = memory.readbyte(0x0200045B)
if party == 0x70 then
ml = (memory.readbyte(0x02000520+0x14C*pc1+0xF)+memory.readbyte(0x02000520+0x14C*pc2+0xF)+memory.readbyte(0x02000520+0x14C*pc3+0xF))/3
else
ml = (memory.readbyte(0x02000520+0x14C*pc1+0xF)+memory.readbyte(0x02000520+0x14C*pc2+0xF)+memory.readbyte(0x02000520+0x14C*pc3+0xF)+memory.readbyte(0x02000520+0x14C*pc4+0xF))/4
end

LevelAve = ml-ela

fleeFail = memory.readbyte(0x02030092)

function flee(S) -- Flee Success Calculation
g = S
g = RNB(g)*10000
fl = 5000 + (2000*fleeFail) + (LevelAve * 500)
g = bit.rshift(g, 16)
if fl >= g then
return true
else
return false
end
end

Hey just checking.... but when fleeing after the reunion does it consider the party level to be the average of all 8 members (assuming they're alive) or just the front four?

Yeah it makes sense to me that SM has all 3 turns determined in one go. Because of that, with certain configurations the guardian ball will always defend.

In any case, 7BRN is very easy to get to and gives good odds. If it proves to be semi-inconsistent we can always manip the grn on the title screen for improved accuracy.

And just like that, 5/7 guards starting on BRN7 and 3/7 guards starting on BRN11. This is certainly what is going on.

Very curious behavior!

EDIT: the real probability (assuming the order in which the actions are assigned is random) should be 80%, so my sample calculations were a good 10% out

That looks like a promising theory and would explain the behavior I'm seeing. Shuffling the order in which moves are determined would mean the guardian ball is using a variety of different BRN seeds instead of one seed every time. BRN11 would have about 2-3 good seeds and 5-4 bad seeds, while BRN7 would have about 5 good seeds and 2 bad ones.

Star Magician actually doesn't advance the RNG by one with BRN 0, so your calculations were valid actually you're right. This lines up experimentally with what I observe where Gball always guard auras on 0BRN.

EDIT: If you could check BRNs 8-14 and 12-18 and see how many seeds will make Gball use guard aura? If the ratio of guard aura/defend lines up experimentally this is probably the reason.

EDIT2: I did 8-14 myself and got 4 defends, 3 guards giving me a theoretical ~57% chance of defend. But since Star Magician eats 3 RNs at once, and 80% of the time will consume at least one of the guard seeds I imagine the real probability is nudged higher.

Code Select Expand

BRN7 = 3EAD 62 FB - 62 -> 3
BRN8 = AF5A AD 71 - AD -> 5
BRN9 = 20DA 77 56 - 77 -> 3
BRN10= AFE5 33 D7 - 33 -> 1
BRN11= 69AC C4 C4 - C4 -> 6
BRN12= 961B AF AD - AF -> 5
BRN13= 261E B2 E2 - B2 -> 5
BRN14= 525F 36 73 - 36 -> 1


EDIT3: just saw your post below xD somehow I missed a seed

Found a second seed which gives defend, and I'm thinking my first hypothesis is correct - i.e. that the BRN determines the chance for Guard Aura to work based off the GRN. So far I've been working with BRN11, but turns out BRN7 works as well. I did some manual sampling (51 samples) and BRN7 had a 70% defend rate while BRN11 had a 35% defend rate.

This is very unusual.

Edit: If you want to manually test this, save before star magician [battery file enclosed] enter the fight and do anything with your party. Guardian ball should use guard aura and immediately soft reset when you see this text come up. This will have advanced the BRN by 7. Load back into the star magician fight and enter the battle; you're now fighting star magician at BRN7. You can test the defend/guard aura rate with this.

Alternatively, use my battery file, load gabomba catacombs and enter the mud fight - do atk/def/diamond dust/blitz and soft reset. Now load into the Star Magician fight - you'll be at BRN11. You can test defend/guard aura rate here and compare it to BRN7.

RNG results are trial and error, + I have RN counters in my routing scripts. I've attached an image which illustrates the problem I'm having. In the first image (BRN 11, 0 GRN (0 is relative due to how my script counts)) the Guardian Ball will use Guard aura, while in the second image (BRN 11, 295 GRN (relative to previous image, i.e. this state is 295 GRNs further along than the first one)) the Guardian Ball defends. 

Yeah all I'm trying to say is that guardian ball only has "one action" if it does not defend (guard aura) and that decision is made after all the pc commands are given (not really surprising in any way). The more I test this the more it seems like either i) BRN determines a % chance for guard aura, and then uses the GRN to determine success or ii) it's use is solely determined by GRN. This is based off of the fact that I can't manipulate a defend with 0BRN (trying >10 seeds), but with 11BRN attack cancelling is giving about 50% success.

I pull all my information from: http://goldensunwiki.net/wiki/Ball_enemy_line#Guardian_Ball which says that its an enemy ability and that is has increased priority