AceZephyr

FF7: PSX Vagyrisk Claw Steal Manipulation

Mar 7th, 2026 (edited)
184
0
Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 10.12 KB | None | 0 0
  1. FINAL FANTASY VII
  2.  
  3. Vagyrisk Claw Steal and Battle RNG Manipulation Explanation: (https://youtu.be/Wx7ezMCNKTU)
  4.  
  5. When performing a steal against an enemy with multiple possible steal slots, the game checks in the order of the steal slots on each slot individually whether its item is stolen or not. As an example, Behemoth Horn enemies have steal slots for a Phoenix Down and a War Gong in that order. When performing a steal, a check will first be made against the Phoenix Down slot. If it succeeds, the Phoenix Down is stolen. If it fails, only then is a check made against the War Going slot, and if that succeeds, a War Gong is stolen. If an item has already been stolen from this enemy, no checks are made and "Nothing to steal" is displayed, so at most one item may be stolen from an enemy during its lifetime.
  6.  
  7. This means that the chance to steal the War Gong is reduced because such a steal check requires the previous Phoenix Down steal check to have failed, and if it had succeeded, the War Gong would not be stolen and it would not be possible to attempt another steal, since this enemy already had an item stolen from it. As such, steal slots past the first are ultimately harder to steal than their steal slot rarity would otherwise suggest, since their steal chance is dependent on failing all previous steal checks.
  8.  
  9. This is especially true with the Bagrisk enemy which has three steal slots: a Soft at a common rarity, another Soft at a common rarity, and a Vagyrisk Claw at a rare rarity. Stealing the Claw requires failing both common steal checks and passing the rare check, which is why it has been considered not RTA viable to consistently obtain a Vagyrisk Claw item more quickly than using an early Fort Condor game in the 100% speedrun.
  10.  
  11. How easy or hard a steal is is dependent on the difference in levels between the thief and the target, as well as the steal slot rarity. A good explanation of the mechanics can be found here: https://youtu.be/5U0ETH2F2Mg
  12.  
  13. Slerran had also made a video on the Vagyrisk Claw steal specifically: https://youtu.be/dPo7fzyOGAk
  14.  
  15. It suggests optimal levels for grinding steals with as few enemy encounters or as few steal attempts as possible at levels 5 or 11. However, it assumes that Battle RNG is uniform, which in this case is not actually a reasonable assumption to make.
  16.  
  17. The rest of this explanation relies on a good understanding of the mechanics of Battle RNG: https://youtu.be/2Jka1NKzn8o
  18.  
  19. An individual steal check uses a single 8-bit RNG call to obtain a random number to determine whether it succeeds or not. The current index is never changed between subsequent checks for different steal slots, and no other RNG calls are made between two subsequent steal slot checks. This means that the result of a steal is determined entirely by the thief-and-target level difference (or if they're wearing the Sneak Glove), steal rarity, and only the single RNG index state at the beginning of the steal. Because the subsequent steal checks are dependent on the same RNG index as the first check,
  20.  
  21. As an example, If a steal is made against a Bagrisk and the current index is 5 when the steal logic begins, the first steal check will call RNG and advance it to 6. If that check fails, the RNG index value the second steal check uses will necessarily be the next value, in this case 6. If that fails, the third and final check will use the next RNG index value, in this case 7.
  22.  
  23. For the Vagyrisk Claw steal specifically, we can actually catalogue all possible starting RNG index value and level range combinations that can possibly result in a steal. They are as follows (level ranges are inclusive):
  24.  
  25. 5: 1 - 36
  26. 9: 29 - 50
  27. 30: 1 - 21
  28. 39: 1 - 24
  29. 56: 11 - 14
  30. 59: 17 - 50
  31. 84: 1 - 14
  32. 94: 1 - 21
  33. 110: 36 - 52
  34. 120: 11 - 14
  35. 134: 1 - 38
  36. 148: 1 - 13
  37. 151: 1 - 35
  38. 158: 1 - 18
  39. 161: 1 - 39
  40. 164: 29 - 71
  41. 182: 4 - 32
  42. 198: 1 - 41
  43. 222: 1 - 44
  44. 225: 1 - 38
  45. 231: 1 - 46
  46.  
  47. For example, if the thief has level from 1 to 36, then the index-5 example as described earlier would result in a Vagyrisk Claw steal.
  48.  
  49. With this, assuming all index values at the start of a steal are equally likely, we can compute more accurate steal chances for optimal level-farming strategies. The optimal levels for fewest enemies are actually levels 1 and 2, and the optimal levels for fewest number of steals is a tie between 11, 12, and 13. The data is present in the following spreadsheet:
  50. https://docs.google.com/spreadsheets/d/1UBTjYrP-HlljqxLg3agoinHNZIm_huYYhJr8JvvLdAM/edit?usp=sharing
  51.  
  52. However, knowing that only a single index is used for steal calculations in this manner also suggests that it should be possible to manipulate the steal on PSX by manipulating Battle RNG. As explained in my video on the subject, PSX Battle RNG is seeded by the lowest 15 bits of the current frame counter, with each of the eight indexes being seeded to each of the eight contiguous substrings of bits in this value, with index number 0 seeded to the least significant 8 bits of the seed and index number 7 seeded to the most significant 8 bits.
  53.  
  54. Since we only need to manipulate a single index, we should be able to choose a time to start the battle at so the correct bits in the frame counter have a good value, which increments at about 60 FPS.
  55.  
  56. In theory this may sound good, but there are several complications that need to be addressed:
  57.  
  58. 1. The current RNG index increments on most frames as a result of passive runaway rand16 calls incrementing the current index value. This means that which index is the current one will change on effectively every frame at 15fps between the start of battle and the time at which the steal occurs. We work around this by forcing the encounter with Bagrisks to be a preemptive fight, which the encounter formation does allow, and by buffering the first input to ensure it takes a consistent amount of time to perform. This also eliminates party dexterity as a variable that would affect the amount of time between the start of battle and the Steal.
  59.  
  60. 2. The Joker value at the start of battle, which affects how runaway rand16 calls increment the current index value and how they call RNG, is not set at the start of battle and simply retains its state from the end of the last battle. This is addressed by starting from a hard reset without having gotten into a battle yet, forcing Joker to start in a consistent initial state of 0.
  61.  
  62. 3. Because the frame counter increments by one per frame, the least significant 8 bits that are used to seed the first Battle RNG index (Index 0) also change on every frame, meaning that it would be necessary to start on an exact frame (at 60fps) to choose an Index 0 value. However, the 2nd-least significant bit of the frame counter increments once every two frames, meaning there are 2-frame (1/30 second) windows of time where, on either of those frames, starting a battle would result in the same value for Index 1. The bits used to seed Index 2 change only every 4 frames (1/15 second), and so on, doubling as listed below:
  63.  
  64. Index 0: 1 frame (1/60 second)
  65. Index 1: 2 frames (1/30 second)
  66. Index 2: 4 frames (1/15 second)
  67. Index 3: 8 frames (~1/8 second)
  68. Index 4: 16 frames (~1/4 second)
  69. Index 5: 32 frames (~1/2 second)
  70. Index 6: 64 frames (~1 seconds)
  71. Index 7: 128 frames (~2 seconds)
  72.  
  73. Index 7 has the largest windows, with 128-frame (~2.13-second) windows where starting a battle on any frame in that window results in the same value for Index 7.
  74.  
  75. By default, if a buffered Steal is the first action performed in battle, the Index that is used for that steal is too low to be manipulatable. (I don't think I determined exactly which one, but it was definitely too low). We solve this by performing specific actions that are meant to waste time and call RNG consistently to ensure a subsequent steal occurs using, ideally, Index 7. With the manipulation I found, this happens to have Red buffer throwing a Tranquilizer on himself, followed by any character performing the steal. This results in an Index 7 steal, meaning the battle should have a 2-second window to start.
  76.  
  77. 4. We need to ensure we get a preemptive battle against Bagrisks. There is only one formation that has the Bagrisk enemy (and is not a Back Attack), against 3 of them (though because of different RNG calls as a result of seeding initial ATB, it would be required to manipulate an exact formation as well should there have been more than one). We can ensure we fight this as our first battle after starting the game with World Map RNG Manipulation.
  78.  
  79. In the example demonstrated in the video, we start an external timer when the PlayStation logo disappears, since this is about when the frame counter begins counting. A save file is loaded. The World Map is manipulated so the RNG will result in a preemptive fight against Bagrisks when moving up and down. We move enough so the battle will almost occur, then wait until 3:15.5 on the timer, then move down to trigger the battle. We had arranged our party so Red was first and Tranquilizers are the first item in the inventory, so holding Up+Circle will have Red throw a Tranquilizer on himself, then we have Cid or Cait Sith steal from any enemy. With this particular setup, we should get a steal on Index 7 with RNG index value 134, so a thief of any level from 1 to 38 will steal the Vagyrisk Claw every time when performed correctly.
  80.  
  81. ---
  82.  
  83. More work needs to be done to make this strategy able to be implemented in a PSX 100% speedrun, though much of it is completed with this demonstration. My current idea would be to use a hard reset during Chocobo Breeding to reset Joker in preparation for this manipulation.
  84.  
  85. It is possible to use Knights of the Round to determine the index value that a steal would have used by observing the damage rolls. Like steal chances, damage rolls and all other RNG calls during this attack use rand8 and not rand16 at all. The three damage amounts of the first attack each happen on indexes 3 apart (so 134, 137, 140 for example), and a steal would occur with the same starting index as the lowest of these (so 134 in this case). This can be used to perform research on console, since emulator differs in behavior with loading times and such, resulting in different index and RNG index values.
Add Comment
Please, Sign In to add comment