User Files from FatRatKnight

It's starting to get a little crowded.

Mid-top: CPU car info. Figured out two unknowns and rearranged them to be just beneath the main speed, as they are base speed and RNG speed range. As well, I pick at the RNG and predict the next three potential velocities (white) the car-to-be will have.

Right: Listing of the RNG period of 28 numbers. If the RNG isn't at one of those 28, a second column and predicted values is shown.

Bottom-left: Timers, as before. Frame timer in both decimal and hexidecimal, and the time remaining.

Bottom-mid: Mostly the same as before. Position and road curve, speed and sub-speed, then distance and related calculations. Distance calculations now use an internal number to find a certain multiplier the game uses. The script is calibrated to report the remaining distance in five of the eight courses, since I can't find an automatic process to do so as of this time.

Bottom-right: These numbers are to be analyzed. The darkened ones are for bouncing off of cars. The lighter one is reporting which course I'm in.

Additionally, I have an extending green line and an orange line to indicate progress through the course.

Times measured using address 0029:
C1: 9393 frames (Saturn: 9827)
C2: 9305 frames (Saturn: 10559)
C3: 9237 frames (Saturn: 10087)
C4: 9653 frames (unable to sync Saturn's run in FCEUX)

I must note that Saturn apparently aimed for high score, rather than shortest time. The distinction here is that, when going for high score, you want to reach the finish line with as much speed as possible for extended distance scoring, but fastest time wants you to cross it with as little leftover distance as possible. Saturn also didn't know the full extent of what pausing can really do at the time, which can stop the in-game timer from ticking, which would also add to score. One can imagine how that sounds.

I've been learning more about this game, though. There are a few mistakes I can now see that I likely missed in earlier courses. I may go back to deal with them.

I've been keeping track of rerecord counts for individual courses. C1:1319 C2:2018 C3:1979 C4:1635

Pausing: There are timers that only "tick" every 64 frames, including car spawn timers and the time remaining. Bypass this "tick" by pausing at the right time. Useful for delaying appearance of cars. Each pause costs 4 frames, plus 2 more in the finish countdown if it saved a timer tick, but proper manipulation of other cars can potentially make up for that.

Bumping cars: There are seven ways to interact with other cars. I sometimes bump into the back of one straight on to improve its speed, as the follow-up curve is nasty and I really want a car "on rails" to bounce off of, and for it to keep up with me rather than me slowing down for it. The other six ways have you fly off to some side, but the back-left and back-right collisions slow you down and potentially slowing the other car as well, but are easier to cause at frame-precise times.

Car distance: There is a number for how far along the lane the car is, and another one for "sub-pixel" (x/64) value of that. This "sub-pixel" goes up by the difference in speed, regardless of who is faster. One can build it up at a higher speed, then slow down just below the other car's speed to reach the 64 threshold and get them one "notch" further ahead at minimal distance loss yourself.

Saturn may have made frame-perfect cornering, but manipulating traffic is where the real savings are. This requires some interesting maneuvering and a good memory watch script.

Times measured by 2-byte value at address 0x0029, when it freezes at the next course preview:
Course 1: 9393 frames (Saturn: 9827)
Course 2: 9305 frames (Saturn: 10559)

Seems each course is very well "packaged" away from other courses. Even sub-speed values are reset upon entry. Might be possible for sync stability for swapping in another movie's input for various courses. Maybe.

Displays a lot of numbers. Seems to be the typical thing my scripts do.

Top: The other cars info. Top numbers are ID and spawn timer. Just below that, speed and some sort of distance measurement, color changes when collision is possible. Then we have X offset for lane changes. Then a pile of unknown numbers. This, for each of the three possible cars.

Bottom: On the left is the frame timer, in both decimal (record keeping and notes) and hexadecimal (to help track timer events), and an up-to-date mirror of the time limit. Middle is a mess, but the top two numbers are X offset from center and curve of road, middle is an up-to-date mirror of speed and the handy sub-speed value, and bottom is the distance calculations calibrated for course 1, with the orange number indicating distance lost. I also have a green line that indicates how far along you are in the course, but that is also calibrated for course 1.

The game looked shiny. I got distracted on something else yet again.

For BizHawk. Geez, those forms work slow. I recommend pausing the script when you want to use turbo mode.

Currently predicts d4, d6, d8, d10, d12, and d20 rolls. It gives some pride that I can't detect errors with its predictions, but on the other hand, this could also mean my error detection might need work. Regardless, I think I know all the dice rolls in the future!

It spawns a single window with a list of numbers. Each column reports dice of different sides, with the top numbers being the immediate next result. Anyone here fluent in object properties and can tell me how to right-justify the labels?

Colorful numbers. Tiny cute informative icons. Joy.

I'll leave it up to you to determine what the numbers mean. I produced this script to make it slightly more interesting to watch that recent submission.

There is one commented line at the start of the while true loop. In particular, the one that looks like a commented out line of lua code, not the fancy borders. That's specifically for this submission. Uncomment that line for lighting up dungeons when watching that specific movie.

BizHawk 1.9.1

I don't know where my TASing mood is going, but currently it seems it says I'm not done with this game yet, what with all the analyzing I've done elsewhere.

It is critical that we take the time to kill off two members, as the game insists on having you run with a four-person party, and this divides up the experience necessary for extremely useful spells. Notably, since I'm just checking a route, I did not manipulate luck as much as I should early on. The route seems to work.

I fished around for emulator specific functions, redefined names for them at the top, and now it's mostly a matter of picking similar functions when porting to a different emulator. Mostly.

I expect some poor alignments to come of this, so the rest of the conversion would best be handled by someone else. Haven't bothered looking for a GBA BIOS, so I just used the static screen of a ROM-less BizHawk for display and stand-in functions for memory related stuff. I hope I got the right memory functions, as I didn't rightly check them when getting rid of the stand-ins.

Also, I did two adjustments in the script itself. First, I moved the "EXP to next level calculator" from a list of if/else to a table, and compressed the Bros Attack Counters into fewer lines, including using math.max instead of checking if the number is below zero.

My masterpiece! ... Well, data collection can now be done entirely without my direct input, I hope.

I fixed up the display by flipping it. Rather than the type running off the edge to the right, it now runs downward, with value going to the right instead. I can make much better use of vertical space this way. It still has no concept of screen space, but there's a lot more effective space with the direction I'm having it head.

I also do better than just list out a bunch of IDs for each card. I even identify the name (though I hope I picked the right set of pointers)! The first seven characters of the name is provided, but the full 17 characters is provided if you use the print out button (default: Z) in the script.

Pick a moment just four frames before the white flash. Start the script. Unpause emulation. If you did something wrong, either the little red number goes up (too early or wrong situation), or the exact same card keeps showing up (too late).

I made this script so that all the data collection for prize cards does not have to rely on my hands, though it's also easier for me to use. Also, I'd love to know whether the script is identifying the wrong names.

Refinements need to be made to the output & display, but this bot is working its magic!

Find the exact moment where the game generates a card, pick a few frames before that moment, run the script, and unpause. If you don't know precisely when the RNG is used, try four frames before the white flash.

Anyway, I still need to fix up the output, as it doesn't know the concept of limited screen space, but I have something!

The script attempts to identify whether combat has taken place, and if so, attempts to list out every object within it.

That's all it does. The strings are extracted from the game itself, and the script doesn't really do anything else. It's a proof of concept that I am able to get access to what's going on in there.

As a bonus, the script has some commented out piece of code. Remove the comment markers, then run the script. I haven't looked too hard at this generated list, but there might be some interesting things to look at. It's all in hexadecimal, and the six massive numbers would be memory pointers, not actual data in and of itself.

The goal is to display exactly what you're getting from the shop, in four colorful blocks. With 16 different sorts of packs to choose from, all combinations of (attack, magic, item, mix) and (leaf, brown, black, Moogle) are shown. It should correctly identify everything I know and hint at things I don't know.

The script displays the limits of what I know, but I hope it isn't hard to make adjustments to fill in the gaps.

Version zero. Because it definitely needs more work.

On the left side are the raw 100-sided dice rolls. Jump a few times to get an idea as to what order the numbers are in.

Along the top is the predicted card values: The two-digit number is the raw 100-sided die roll for the identifier, the one-digit number is the value, and it is yellow if the card is premium.

Somehow, I missed the part where the pack quality varies. The prediction likely works with the basic pack. Probably falls apart if you pick something with a fancier binding than the leaf one. What else don't I know? I like to analyze things, so it's no trouble at all.

Optimized the damaging part and the boss of Area 3 - Old Capital. Should be faster than my old Snes9x WIP. I did accidentally clip off the track for a brief moment near the end, but the amount of slowing that resulted thankfully did not cost me any frames, so I left it in as I would rather keep my boss progress.

... And I made some progress into Area 4 - Grass Land. Halfway through, too! And that one "fluffy snowball" that looks like I missed, flying past me to the right is actually hit by my charged shot. There is a solid explosion sound, and my score increases, that's your clues.

The part I stopped at has these growing plant towers of sorts... Difficulties:

Why, hello there, puzzle. I expect difficulties optimizing my score.

Ah, HyperZone. Yet another project I got myself distracted on. Will I ever stay on something long enough to get it done? Area 2 - Blast Furnace done.

Been redoing HyperZone on lsnes. Differences between this and my Snes9x TAS:

Fast Lap (lap 5) begins at frame 7255. Recommend starting at 7000. For reference, An unassisted record (WR I believe) by edward406 is 22"31, so at least I'm doing something right this time by having my TAS beat "a mere human," unlike my Maximum Velocity attempt.

26"30 - 23"58 - 23"59 - 23"63 - 22"06 = 1'59"16
Bad opener (WR is 25"36). The goal of this TAS wasn't a fast total, but a fast lap.

Speaking of which, if I was able to match the WR opener with Expert CPU rather than Practice Golden Fox CPU, I'd beat the unassisted WR for the five lap time, despite it not being the primary goal.

Let's see... Trick the timer to tick "01 rather than "02 for the first frame of the lap, edge out every sub-pixel I can entering the lap, abuse the rail to brake even faster, abuse the fact my accelerator is useless for maintaining boosted speed but useful for dealing with my momentum's direction, and realize the "dirt" doesn't slow me down worse than normal roadway (but somehow doesn't let me accelerate on it even boosted).

Made a start on Big Blue, trying out a mediocre opener. Expert doesn't look like it'll cut it for five lap times. Verification movie for Master it is.

Ended at level 19. I guess the legendary armor and shield aren't needed. Didn't need more than 1 HP throughout the last fight.

Compared to my last WIP, I did slay that hydra boss over a dozen seconds faster. It was just a matter of raising my Arm Strength during the fight so that the Blade of Muramasa actually does more than 2 damage. Sword of Flames was sacrificed for inventory management purposes.

85 Arm Strength backed by Blade of Muramasa only does 1 to 2 damage -- No better than just punching the thing. By the time I finish, I have 97 Arm Strength. The only better weapon to find is in that small chamber right after the boss. Don't bother magicking the boss, it's immune.

In any case, I only did minor-medium luck manipulation. Enough to delay some of its attacks. I did not deal with the 1-2 damage jitter. It has 250 HP to knock down. I should check if cheating my strength to 99 has any effect, other than muting the Arm Strength messages. Would let me know if I should keep the Blade unbroken until my arms grow a bit.

On a side note, I'm really, really glad frame perfection alone is enough to survive those arrow traps... 100+ damage doesn't help someone who only has 44 HP.

First, I had to get some extra keys, because I was one short.
Second, I figured out a glitch that lets me jump past teleporters.
Third, that three floor maze was a lot smaller than I thought.

I'm making progress once more. It appears likely I'll, at long last, give this run the finishing touches it needs.

Just a few questions I need to find answers for:

... I'm really hoping my 5 Magic Endurance doesn't prove fatal...

I'm a second slow. I blame the opening bumps. Practice mode Golden Fox CPU does wonders. GP mode Expert CPU not so much, apparently. Whether it's my technique or just that these CPU are not good bumping material is in question. I want Master unlocked!

Regardless, if anyone wants this .lsmv file that lasts only one lap, I'm making it available.

Also, those dirt patches reduce your turning rate, but only reduce speed by roughly 4 per frame. The electric rails don't affect turning rate, but reduces speed by 16 per frame. Your turning rate is still limited by your momentum, which usually can't be sped up reasonably, but the dirt patches are more than enough to let your momentum catch up with your facing.

It's a glorified memory watch script. Though at least it'll tell how much of your speed is being applied to each axis depending on which direction your momentum is headed.

What will I work on next? I have no clue. I'm rather chaotically analyzing games at the moment. ... It's something, I guess.

Oh, and whoops. Last file mixed up on-screen positions of the X and Y positions with the X and Y velocities.

VBA-RR v23.6 svn480

Despite a perfect hole-in-one run where all ten shots make it in for the contest, it would appear there is just no impressing certain NPCs. I guess, technically, not a single one of my shots stopped on the green. Man, I'm a terrible player.

Note Shot Lesson 2 is complete, but Shot Contest 2 isn't cleared, even though I scored 10/10, got the EXP, and unlocked a star hole. I intentionally delay at texts to give you a chance to read most of it.

SRAM anchored, so I can skip the long introduction. SRAM also has the POW clubs so I can play with a little more distance, but I otherwise have default level 1 stats. A funny exhibition for what TAS can do.

... Alright, fine. That recent golf submission reminded me of this game.

Emulator: VBA-RR v24m svn480

Unassisted fast lap for Pawn 1, Bianca City - Stretch Circuit: 19"28

I'm doing something wrong, as I'm 0.20 seconds slower. Either that, or the Master level Rival in Training gives enough of an advantage over Expert level Grand Prix. But I'd like to say it's my error.

I know, this is meant to be a verification movie. But it's no good if I don't hone my knowledge over the game. So make the verification good at something, or try to.

Splits: 36"33 34"11 23"64 40"15 19"48

Emulator: VBA-RR v24m svn480

http://www.youtube.com/watch?v=y7OeNApAntY

I don't know... Either I'm playing this game horribly wrong, or I'm doing it incredibly right. The game was telling me to turn back most of the time. I got a time of 36"35 for my first lap before blowing up in a tragic explosion, while still in first place in Expert.

To convince you to look, this is a hilarious 36"35 to watch (my opinion). I don't believe I was driving in any fashion a sane player would do.

Seems to work fine (though this was made in v24 and probably won't work in v23.6). I'm curious about thoughts of any experts who've been messing with multiple versions of VBA.

Go into name entry. Run script. Mess around in there. Tells you what password for any name you input as you're messing around.

I left a few notes in the code, if you need them.