These are documented bugs and glitches in the original Pokémon Trading Card Game.
Fixes are written in the diff format.
this is some code
-delete red - lines
+add green + linesDisclaimer regarding the fixes: since the project is still in the process of being disassembled, applying code modifications that result in a different number of bytes in the instructions will lead to lots of pointer invalidation, which will potentially lead to crashes.
- AI wrongfully adds score twice for attaching energy to Arena card
- Cards in AI decks that are not supposed to be placed as Prize cards are ignored
- AI score modifiers for retreating are never used
- AI handles Basic Pokémon cards in hand wrong when scoring the use of Professor Oak
- Rick never plays Energy Search
- Rick uses wrong Pokédex AI subroutine
- Chris never uses Revive on Kangaskhan
- AI Pokemon Trader may result in unintended effects
- AI never uses Energy Trans in order to retreat Arena card
- Sam's practice deck does wrong card ID check
- AI does not account for Mysterious Fossil or Clefairy Doll when using Shift Pkmn Power
When the AI is scoring each Play Area Pokémon card to attach an Energy card, it first checks whether it's the Arena card and, if so, checks whether the attack can KO the Defending Pokémon. If it's true, then 20 is added to the score. Then it does the same Arena card check to increase the score further. The intention is probably to score the card with 20 if it can KO the opponent regardless of Play Area position, and additionally if it's the Arena card it should score 10 more points.
Fix: Edit DetermineAIScoreOfAttackEnergyRequirement in src/engine/duel/ai/energy.asm:
DetermineAIScoreOfAttackEnergyRequirement: ; 16695 (5:6695)
...
-; if the attack KOs player and this is the active card, add to AI score.
+; if the attack KOs player add to AI score.
- ldh a, [hTempPlayAreaLocation_ff9d]
- or a
- jr nz, .check_evolution
ld a, [wSelectedAttack]
call EstimateDamage_VersusDefendingCard
ld a, DUELVARS_ARENA_CARD_HP
call GetNonTurnDuelistVariable
ld hl, wDamage
sub [hl]
jr z, .atk_kos_defending
jr nc, .check_evolution
.atk_kos_defending
ld a, 20
call AddToAIScore
-; this is possibly a bug.
-; this is an identical check as above to test whether this card is active.
-; in case it is active, the score gets added 10 more points,
-; in addition to the 20 points already added above.
-; what was probably intended was to add 20 points
-; plus 10 in case it is the Arena card.
+; add 10 more in case it's the Arena card
ldh a, [hTempPlayAreaLocation_ff9d]
or a
jr nz, .check_evolution
ld a, 10
call AddToAIScore
...Each deck AI lists some card IDs that are not supposed to be placed as Prize cards in the beginning of the duel. If the deck configuration after the initial shuffling results in any of these cards being placed as Prize cards, the game is supposed to reshuffle the deck. An example of such a list, for the Go GO Rain Dance deck, is:
.list_prize ; 14fe6 (5:4fe6)
db GAMBLER
db ENERGY_RETRIEVAL
db SUPER_ENERGY_RETRIEVAL
db BLASTOISE
db $00
However, the routine to iterate these lists and look for these cards is buggy, which results in the game ignoring it completely.
Fix: Edit SetUpBossStartingHandAndDeck in src/engine/duel/ai/boss_deck_set_up.asm:
SetUpBossStartingHandAndDeck: ; 172af (5:72af)
...
-; expectation: return carry if card ID corresponding
+; return carry if card ID corresponding
; to the input deck index is listed in wAICardListAvoidPrize;
-; reality: always returns no carry because when checking terminating
-; byte in wAICardListAvoidPrize ($00), it wrongfully uses 'cp a' instead of 'or a',
-; so it always ends up returning in the first item in list.
; input:
; - a = deck index of card to check
.CheckIfIDIsInList ; 17366 (5:7366)
ld b, a
ld a, [wAICardListAvoidPrize + 1]
or a
ret z ; null
push hl
ld h, a
ld a, [wAICardListAvoidPrize]
ld l, a
ld a, b
call GetCardIDFromDeckIndex
.loop_id_list
ld a, [hli]
- cp a ; bug, should be 'or a'
+ or a
jr z, .false
cp e
jr nz, .loop_id_list
; true
pop hl
scf
ret
.false
pop hl
or a
retEach deck AI lists some Pokémon card IDs that have an associated score for retreating. That way, the game can fine-tune the likelihood that the AI duelist will retreat to a given Pokémon from the bench. For example, the Legendary Dragonite deck has the following list of retreat score modifiers:
.list_retreat ; 14d99 (5:4d99)
ai_retreat CHARMANDER, -1
ai_retreat MAGIKARP, -5
db $00
However, the game never actually stores the pointer to these lists (a notable exception being the Legendary Moltres deck), so the AI cannot access these score modifiers.
Fix: Edit all applicable decks in src/engine/duel/ai/decks/, uncommenting the following line:
.store_list_pointers
store_list_pointer wAICardListAvoidPrize, .list_prize
store_list_pointer wAICardListArenaPriority, .list_arena
store_list_pointer wAICardListBenchPriority, .list_bench
store_list_pointer wAICardListPlayFromHandPriority, .list_bench
- ; missing store_list_pointer wAICardListRetreatBonus, .list_retreat
+ store_list_pointer wAICardListRetreatBonus, .list_retreat
store_list_pointer wAICardListEnergyBonus, .list_energy
retWhen the AI is checking whether to play Professor Oak or not, it does a hand check to see if there are any Basic/Evolved Pokémon cards. One of these checks is supposed to add to the score if there are Basic Pokémon in hand, but as it is coded, it will never execute the score addition.
Fix: Edit AIDecide_ProfessorOak in src/engine/duel/ai/trainer_cards.asm:
AIDecide_ProfessorOak: ; 20cc1 (8:4cc1)
...
-; this part seems buggy
-; the AI loops through all the cards in hand and checks
-; if any of them is not a Pokemon card and has Basic stage.
-; it seems like the intention was that if there was
-; any Basic Pokemon still in hand, the AI would add to the score.
.check_hand
call CreateHandCardList
ld hl, wDuelTempList
.loop_hand
ld a, [hli]
cp $ff
jr z, .check_evolution
call LoadCardDataToBuffer1_FromDeckIndex
ld a, [wLoadedCard1Type]
cp TYPE_ENERGY
- jr c, .loop_hand ; bug, should be jr nc
+ jr nc, .loop_hand
...The AI's decision to play Energy Search has two special cases: one for the Heated Battle deck and the other for the Wonders of Science deck. The former calls a subroutine to check only for Fire and Lightning energy cards in the deck, and the latter only checks for... Fire and Lightning energy cards. In a deck filled only with Grass and Psychic types. Needless is to say, poor Rick never finds any of those energy cards in the deck, so the Energy Search card is left forever unplayed. There's an unreferenced subroutine that looks for Grass energy that is supposed to be used instead.
Fix: Edit AIDecide_EnergySearch in src/engine/duel/ai/trainer_cards.asm:
AIDecide_EnergySearch: ; 211aa (8:51aa)
...
-; this subroutine has a bug.
-; it was supposed to use the .CheckUsefulGrassEnergy subroutine
-; but uses .CheckUsefulFireOrLightningEnergy instead.
.wonders_of_science
ld a, CARD_LOCATION_DECK
call FindBasicEnergyCardsInLocation
jr c, .no_carry
- call .CheckUsefulFireOrLightningEnergy
+ call .CheckUsefulGrassEnergy
jr c, .no_carry
scf
ret
...Seems Rick can't catch a break. When deciding which cards to prioritize in the Pokédex card effect, the AI checks if it's playing the Wonders of Science deck, then completely disregards the result and jumps unconditionally. Thus, Rick uses the generic algorithm for sorting the deck cards.
Fix: Edit AIDecide_Pokedex in src/engine/duel/ai/trainer_cards.asm:
AIDecide_Pokedex: ; 212dc (8:52dc)
...
.pick_cards
-; the following comparison is disregarded
-; the Wonders of Science deck was probably intended
-; to use PickPokedexCards_Unreferenced instead
ld a, [wOpponentDeckID]
cp WONDERS_OF_SCIENCE_DECK_ID
- jp PickPokedexCards ; bug, should be jp nz
+ jp nz, PickPokedexCards
+ ; fallthrough
; picks order of the cards in deck from the effects of Pokedex.
; prioritizes Pokemon cards, then Trainer cards, then energy cards.
; stores the resulting order in wce1a.
-PickPokedexCards_Unreferenced: ; 212ff (8:52ff)
-; unreferenced
xor a
ld [wAIPokedexCounter], a ; reset counter
...Because of an error in the AI logic, Chris never considers using Revive on a Kangaskhan card in the Discard Pile, even though it is listed as one of the cards for the AI to check.
Fix: Edit AIDecide_Revive in src/engine/duel/ai/trainer_cards.asm:
AIDecide_Revive: ; 218a9 (8:58a9)
...
-; these checks have a bug.
-; it works fine for Hitmonchan and Hitmonlee,
-; but in case it's a Tauros card, the routine will fallthrough
-; into the Kangaskhan check. since it will never be equal to Kangaskhan,
-; it will fallthrough into the set carry branch.
-; in case it's a Kangaskhan card, the check will fail in the Tauros check
-; and jump back into the loop. so just by accident the Tauros check works,
-; but Kangaskhan will never be correctly checked because of this.
cp HITMONCHAN
jr z, .set_carry
cp HITMONLEE
jr z, .set_carry
cp TAUROS
- jr nz, .loop_discard_pile ; bug, these two lines should be swapped
+ jr z, .set_carry
cp KANGASKHAN
- jr z, .set_carry ; bug, these two lines should be swapped
+ jr nz, .loop_discard_pile
...A missing line in AI logic might result in strange behavior when executing the effect of Pokémon Trader for Power Generator deck.
Fix: Edit AIDecide_PokemonTrader_PowerGenerator in src/engine/duel/ai/trainer_cards.asm:
AIDecide_PokemonTrader_PowerGenerator: ; 2200b (8:600b)
...
call LookForCardIDInDeck_GivenCardIDInHand
jr c, .find_duplicates
- ; bug, missing jr .no_carry
+ jr .no_carry
-; since this last check falls through regardless of result,
-; register a might hold an invalid deck index,
-; which might lead to hilarious results like Brandon
-; trading a Pikachu with a Grass Energy from the deck.
-; however, since it's deep in a tower of conditionals,
-; reaching here is extremely unlikely.
; a card in deck was found to look for,
; check if there are duplicates in hand to trade with.
...There is a mistake in the AI retreat logic, in src/engine/duel/ai/decks/general.asm:
.used_switch
; if AI used switch, unset its AI flag
ld a, [wPreviousAIFlags]
and ~AI_FLAG_USED_SWITCH ; clear Switch flag
ld [wPreviousAIFlags], a
; bug, this doesn't make sense being here, since at this point
; Switch Trainer card was already used to retreat the Pokemon.
; what the routine will do is just transfer Energy cards to
; the Arena Pokemon for the purpose of retreating, and
; then not actually retreat, resulting in unusual behaviour.
; this would only work placed right after the AI checks whether
; they have Switch card in hand to use and doesn't have one.
; (and probably that was the original intention.)
ld a, AI_ENERGY_TRANS_RETREAT ; retreat
farcall HandleAIEnergyTrans
ret
Fix: TODO
There is a mistake in the AI logic for deciding which Pokémon for Sam to switch to, in src/engine/duel/ai/decks/sams_practice.asm:
; this is a bug, it's attempting to compare a card ID with a deck index.
; the intention was to change the card to switch to depending on whether
; the first Machop was KO'd at this point in the Duel or not.
; because of the buggy comparison, this will always jump the
; 'inc a' instruction and switch to PLAY_AREA_BENCH_1.
; in a normal Practice Duel following Dr. Mason's instructions,
; this will always lead to the AI correctly switching Raticate with Machop,
; but in case of a "Free" Duel where the first Machop is not KO'd,
; the intention was to switch to PLAY_AREA_BENCH_2 instead.
; but due to 'inc a' always being skipped, it will switch to Raticate.
ld a, DUELVARS_ARENA_CARD
call GetTurnDuelistVariable
cp MACHOP ; wrong
ld a, PLAY_AREA_BENCH_1
jr nz, .retreat
inc a ; PLAY_AREA_BENCH_2
Fix: Edit AIDecide_PokemonTrader_PowerGenerator in src/engine/duel/ai/trainer_cards.asm:
AIPerformScriptedTurn: ; 1483a (5:483a)
...
-; this is a bug, it's attempting to compare a card ID with a deck index.
-; the intention was to change the card to switch to depending on whether
-; the first Machop was KO'd at this point in the Duel or not.
-; because of the buggy comparison, this will always jump the
-; 'inc a' instruction and switch to PLAY_AREA_BENCH_1.
-; in a normal Practice Duel following Dr. Mason's instructions,
-; this will always lead to the AI correctly switching Raticate with Machop,
-; but in case of a "Free" Duel where the first Machop is not KO'd,
-; the intention was to switch to PLAY_AREA_BENCH_2 instead.
-; but due to 'inc a' always being skipped, it will switch to Raticate.
ld a, DUELVARS_ARENA_CARD
call GetTurnDuelistVariable
+ call GetCardIDFromDeckIndex
+ ld a, e
- cp MACHOP ; wrong
+ cp MACHOP
ld a, PLAY_AREA_BENCH_1
jr nz, .retreat
inc a ; PLAY_AREA_BENCH_2
.retreat
call AITryToRetreat
ret
...Fix: Edit HandleAIShift in src/engine/duel/ai/pkmn_powers.asm:
HandleAIShift: ; 22476 (8:6476)
...
.CheckWhetherTurnDuelistHasColor ; 224c6 (8:64c6)
ld a, [wAIDefendingPokemonWeakness]
ld b, a
ld a, DUELVARS_ARENA_CARD
call GetTurnDuelistVariable
.loop_play_area
ld a, [hli]
cp $ff
jr z, .false
push bc
call GetCardIDFromDeckIndex
call GetCardType
- ; in case this is a Mysterious Fossil or Clefairy Doll card,
- ; AI might read the type of the card incorrectly here.
- ; uncomment the following lines to account for this
- ; cp TYPE_TRAINER
- ; jr nz, .not_trainer
- ; pop bc
- ; jr .loop_play_area
-; .not_trainer
+ cp TYPE_TRAINER
+ jr nz, .not_trainer
+ pop bc
+ jr .loop_play_area
+.not_trainer
call TranslateColorToWR
pop bc
and b
jr z, .loop_play_area
; true
scf
ret
.false
or a
ret
...