Final Fantasy VIII

Triple Triad

Updated 2026-09-13

Full spoilers.

Triple Triad is the only system in FFVIII that can permanently destroy your own collection, and the mechanism that does it — how rules migrate between regions — is the worst-documented thing in the game. Four incompatible models of that algorithm are in print. The files cannot settle it, because the logic lives in code nobody has read and the card ids in the enemy records are still undecoded numbers.

This page counts the disagreements rather than picking a winner, and then tells you what to do given that nobody knows.

The board

3×3 grid, five cards each, alternating placement. Every card carries four values, one per edge, from 1 to 9 plus A. Place a card so its facing value beats the neighbour’s facing value and the neighbour flips to your colour. Whoever owns more cards when the ninth lands wins.

A counts as 10 for arithmetic, which matters only under Plus.

Nothing above is disputed anywhere. Everything after it is disputed somewhere.

The special rules

Open — you see the opponent’s five cards. Nine or more sources describe it identically, and three of them editorialise that it is the one rule worth deliberately spreading. It is the only special rule that helps you more than it helps the machine.

Same — if a placed card’s values equal the touching values of two or more neighbours, and at least one of those neighbours is enemy-owned, every matched enemy neighbour flips. The enemy-owned condition is stated explicitly by one source and agreed by three more. Triggers Combo.

Same Wall — the four board edges count as A for the purpose of Same.

• Inert unless Same is also active: five sources, one of which states flatly that it does nothing on its own.
• Makes both Same and Plus easier: one source, contradicting the other five.

One wiki adds a quirk nobody else records: Same Wall still spreads into and out of regions where it does nothing, so it can sit in a region as dead weight.

Plus — on placement, each neighbour’s facing value is added to the placed card’s value on that side. If two or more of those sums are equal, the enemy cards involved flip. Seven sources agree. Triggers Combo.

Combo — not a rule you can toggle. It fires automatically whenever Same, Same Wall or Plus flips anything. Each flipped card then behaves as though freshly placed and flips its own weaker neighbours, cascading. The cascade uses plain number comparison only — it does not re-check Same or Plus. Two sources stress this; none contradicts it.

Elemental — element icons scatter onto random squares at the start. A matching card gains +1 to all four values; a non-matching card, including every non-elemental card, loses 1 from all four. Roughly a third of the card pool has an element.

One wiki states — alone, unverified elsewhere — that these ±1 modifiers do not feed into Same, Plus or Same Wall, only into ordinary comparisons. If true it is the single most important footnote in the rule set, and no other source in this survey mentions it.

Sudden Death — a draw restarts immediately, and each player’s new hand is the five cards that were their colour on the board at the end of the previous round. Repeats until somebody wins.

Random — your hand is chosen for you. See below; this is the one that costs you cards.

The trade rules

One — the winner takes one card from the loser. Default everywhere, and the starting trade rule in all eight regions.

All — the winner takes the loser’s entire hand.

Direct — each player keeps whatever cards were their colour when the board filled. A loser can come away with cards and a winner can lose them. Six sources describe it this way. One Steam guide frames it as “the winner keeps only what they captured”, which wrongly implies the loser gets nothing.

Diff — the winner takes cards equal to the margin. This is the worst-sourced number in the whole card game: four formulations, no two identical, and nobody acknowledges the others.

• Equal to the margin, and the winner takes all if the margin exceeds 5 — one source.
• 6–4 takes 2, 7–3 takes 4, 8–2 or better takes all — one source, note the non-linear jump.
• The possible counts are 2, 4 or 5 — one source.
• Simply “the difference in score”, with 6–4 taking 2 — two sources.

Nobody has reconciled these. Treat Diff as “you will lose an unpredictable number of cards” and do not plan around a specific figure.

The trade-rule ladder is probably a myth

The widely repeated model is that trade rules degrade one step at a time — All to Direct to Diff to One.

Three sources assert or imply that ladder. Two explicitly deny it. One of the two calls it a long-standing myth by name; the other states that All does not become Direct, it goes straight to One.

This is not a small disagreement, because the ladder is the basis of a lot of published advice about how long a favourable trade rule will last. Two of the most technical sources in the corpus say the ladder does not exist.

The eight regions

Eight regions, agreed by six sources that list them explicitly. No source in this survey gives any other count.

Esthar: six sources list Elemental and Same Wall normally. One says Same Wall is hidden there — in effect, but not shown on the rules screen. A rule that is active and invisible is a complete explanation on its own for why players conclude this minigame is broken, and it is single-sourced.

Lunar: five sources say all seven rules run at once, Open included. One omits Open, listing the other six. The practical consequence of the five-source reading is good news: a region holding every rule cannot have anything mixed into it, so playing there can only ever abolish something.

Which town belongs to which region is spelled out fully by one wiki and partially by two others. Most sources never say, and that gap is a routine source of accidental rule carrying — you cross a region boundary without knowing you did.

The no-rules opponent: one wiki alone documents a plainclothes Garden student who appears at random by the Balamb Town waterfront, plays with no special rules at all, and whose defeat wipes every special rule from the Balamb region, Open included. It does not touch trade rules. No other source mentions him. If he exists as described he is the only eraser in the game, and losing Open to him is a real cost.

How rules move — the contested core

What every source agrees on

• Every region has its own rule set, and those sets mutate over a playthrough.
• The rules you carry are the rules of the last region you played in. Carrying replaces; it does not accumulate. Four sources define it this way.
• A region can only change if you arrive carrying at least one rule it does not have.
• Once a rule is in a region it stays until specifically abolished. Nothing decays on its own.
• When you challenge someone while carrying a foreign rule, the NPC offers to mix the two sets. Answering yes plays the union and runs the rule-change algorithm. Answering no plays the local set and changes nothing. Seven sources.
• Refusing repeatedly makes the offers stop, after which you can play that region in peace. Four sources.
• Quitting at the card-selection screen still counts as a game for rule-change purposes, which is the fast way to grind attempts. Four sources; one adds that merely reaching the rules screen is enough.

One source contradicts the entire prompt mechanic, stating that the first game in a region that does not know all your rules simply is played with both sets, with no prompt and no choice. One source against seven.

The four models

Here is the centrepiece, and it does not resolve.

Model A — three equally likely outcomes. After a mixed game, exactly one of three things happens at random regardless of whether you won, lost or quit: a rule you carried spreads in, a rule the region already had is abolished, or nothing happens. Two sources. Neither gives percentages.

Model B — weighted selection, up to three checks. The game picks a rule from the pool with Open weighted at 25% and every other rule at 12.5%. If the picked rule can spread — you carry it, the region lacks it — it spreads and the process ends. Up to three such checks run. If none spread and the third picked rule is already present in the region, there is a 50% chance that rule is abolished instead. Three sources, citing identical numbers.

Model C — unspecified randomness, Open dominant. It is random, Open spreads far more often than anything else, and you cannot choose what spreads or dies. Save first. Three sources.

Model D — no algorithm, because there is no prompt. The union simply happens. One source.

So: three sources give a concrete weighted algorithm, two give a flat three-way roll, three give no mechanism at all, and one denies the prompt exists. Model B and Model C are compatible in spirit — B’s 25% Open weighting is exactly what C describes qualitatively — but B and A are not, and nothing reconciles either with D.

None of this is settled by the shipped data. The rule-change logic is game code, not a table, and no entry on this site treats any of the four as fact.

What to do given that nobody knows

Work from the parts all four models share, because they are the parts that carry every lever you actually have.

1. Save before every mixed game. This is correct under all four models and it is the only advice that is. Five sources name casual acceptance of the mix prompt as the most common irreversible mistake in the minigame.
2. Answer no by default. A declined game changes nothing under every model but D, and D is the one-source outlier. Decline until the offers stop, then farm the region in peace.
3. Carry deliberately or not at all. To move a rule from region A to region B: play one game in A, travel to B without playing cards anywhere in between, challenge, accept. Five sources describe exactly this. The canonical worked example is playing the Timber gate guard for Galbadia’s Same-only set before walking into Dollet.
4. Do not count on forcing abolition. One source — the most technical in the corpus — points out that if you arrive carrying only rules the region already has, nothing can spread and the algorithm is pushed down the abolition branch. That follows from Model B and from nothing else. It costs you nothing to try and it is not a plan.
5. Do not standardise every region to the same set. If all eight regions hold identical rules you have nothing left to carry and you permanently lose the ability to change anything anywhere. One source states this — the least repeated and arguably most consequential strategic warning in the entire corpus.

The reset mechanics behind the published recipes

One source supplies the reason the save-scum recipes work at all: FFVIII’s RNG runs a fixed sequence, a soft reset preserves your position in that sequence and a hard reset returns you to the start. A reproducible opening — inspecting draw points or magazine stacks, then a set pattern of challenge and refuse — lands you on a known index.

Two modern caveats, from one source: the 2019 Remaster removed soft-reset manipulation entirely, so use draw points rather than magazine stacks, and never run the Hi-Speed boost while doing rule work.

Random, and why it is the rule to kill

Random replaces your chosen hand with five cards picked for you. Every source that ranks the rules ranks it worst; one names abolishing it the single top priority.

The pool it draws from is unresolved and it matters enormously.

• Your whole inventory — four sources.
• Your deck — three sources.

No source addresses the contradiction. If Random draws from your collection then every rare card you own increases the odds it deals you one, and the danger compounds as your collection improves. If it draws from your deck, it is an annoyance. The gap between those two readings is the gap between “this rule can cost you five unique cards in one match” and “this rule is irritating”.

The compounding trap, which no single source states in one place: Random removes your ability to keep rares out of your hand, so under All you can lose five unique cards in one loss, and under Direct you can bleed them out of a game you won. Dollet, Trabia and Centra all start with Random by default and it can spread into Balamb and Galbadia.

Assume the inventory reading until somebody proves otherwise. It is the assumption that costs you nothing if you are wrong.

The two published recipes for abolishing it conflict

Both target Random in Dollet, both assume save-scumming, both start by picking up Galbadia’s Same-only set in Timber and saving at the Dollet hotel.

Two sources, two different counts, and only one ties the count to the Queen’s position.

Both recipes depend on a mechanism only one source in the entire survey documents: that declining a game outright advances the card-rule RNG by one step without spreading anything. Every “decline N times, then accept” recipe in circulation rests on that single-sourced claim. That is almost certainly why the two published counts differ — they are hand-calibrated to a mechanism nobody else has described, let alone measured.

Either recipe’s stated outcome is the same: either Open spreads into Dollet or Random is abolished. If Open spreads, reload and repeat. That is the real procedure, and the decline count is noise around it.

Trade-rule spread is a separate system

Two sources document this in numbers, and their figures match each other exactly. One source flatly denies the whole thing, saying trade-rule spread cannot be manipulated and you cannot even observe which region is dominant. Two against one, with the two agreeing to the byte.

• Dominance. Each region carries a value from 0 to 10. Playing in a region raises its Dominance by 1; playing elsewhere lowers a region’s Dominance by 1. At 0, some other region becomes dominant at Dominance 1.
• Spreading. Each game anywhere carries a 25/256 chance, about 9.76%, per Dominance level that the dominant region’s trade rule is copied onto a random region. At Dominance 10 that is near certain.
• Degeneration. Each game also carries a chance that a random region’s trade rule reverts to One. That chance climbs from 0/256 to 7/256, about 2.7%, per challenge, cycling and resetting once it passes roughly 250/256. When degeneration chance reaches 50% or higher, the Queen remarks that people are avoiding risky trades — an in-game tell for a hidden counter.
• Queen adoption. Playing anyone in the Queen’s current region carries a 90/256 chance, about 35%, that the whole region adopts her personal trade rule. Challenging her directly has roughly a 220/256 chance, about 86%, of shifting her personal rule one step afterwards. The simpler framing from the same source: her personal rule is just the last trade rule she played you under, so you reroll it by re-challenging her.
• One source adds that while the Queen sits in a region, that region’s trade rules stop degenerating toward One.

Disc 4 lock: at the point of no return the Queen’s region and personal trade rule freeze permanently and she sits at the crash site. Three sources. Whatever she is holding when you cross that line is what she holds forever.

The Queen of Cards chain

She starts in Balamb Town — six sources, one specifying the train station. She relocates whenever a rare card changes hands with her. Her possible homes are Balamb, Dollet, Deling City, Fishermans Horizon, Winhill, Esthar, Shumi Village and the Lunar gate.

The gating condition everyone agrees on: she only mentions her painter father, and the chain only advances, when she is in or heading to Dollet. Five sources. Two of them recommend bouncing her between Balamb and Dollet exclusively, one adding that this also minimises foreign rule spread.

Movement odds are badly sourced. One guide gives a directed table of possible destinations without weights; one puts Balamb to Dollet at 33%; one gives no odds and says to reload whenever she announces anywhere but Dollet; and one claims — alone and unverified — that taking a rare card back from her in the same exchange makes her stay put for a while.

Save before every match with her. Three sources. An unintended relocation costs you the reload you did not make.

The chain itself

Five sources give the same five-step sequence.

Two sources dissent from the consensus five. One adds a sixth step — losing Phoenix back to her after obtaining it — which the other five treat as the terminal reward. One compresses the chain to MiniMog, then Chicobo and Alexander on the same visit, then Doomtrain, omitting the Sacred step entirely; that version is incompatible with all five consensus sources.

One source is the only one to assign disc windows to individual steps: 1, 1–2, 2, 3, 3.

Whether recovering each card is mandatory splits two against two. Two sources frame “lose the card, then win it back from her son before proceeding” as a required sub-step; two treat recovery as optional. Her son lives in the painter’s house in Dollet and has a chance of playing any rare card you previously lost to her — four sources.

The trade rule to use against her is direct conflict. Two sources say challenge repeatedly until it reads Direct, quitting and re-challenging otherwise. One says set it to All. Neither side acknowledges the other. Direct is the safer of the two under any reading of the trade rules, because it is the only one where losing does not automatically hand over a card you did not choose.

Deadline. One source says the quest must be finished before Disc 4, with the practical window being after the Ragnarok in late Disc 3 but before the point of no return. One says it can be started anywhere from early Disc 1 to Disc 3.

Two traps, one source each: the Timber pub man plays Doomtrain very rarely and a dozen-plus games is normal, and if you miss Alexander at the Lunar Base, Piet is later found at an unmarked crash site south of Esthar. A third source notes the MiniMog jogger disappears late and can be made to reappear by challenging his sister.

The CC Group

Entry: win Triple Triad games inside Balamb Garden until CC Jack reveals himself. Fifteen wins, per three sources; a fourth gives no number. One source alone specifies that games in the Cafeteria, Library and 2F Classroom do not count toward the fifteen — which, if true, is exactly the detail that makes a player conclude the trigger is broken.

The order is disputed.

• Jack, Joker, Club, Diamond, Spade, Heart, King — three sources.
• Jack, Club, Diamond, Joker, Spade, Heart, King — one source.
• One source additionally marks Joker as optional rather than a required link.

Who holds Gilgamesh: two sources say the King. One says Dr. Kadowaki. The two sides partly reconcile — both King sources describe the King being unlocked by playing Kadowaki and then resting in the dorm repeatedly, so the third source has likely recorded the unlock step as the holder.

Two more single-source prerequisites: Spade requires having visited B1F first; Heart appears after leaving Fishermans Horizon per one source and once Garden is mobile per two others. One guide describes Spade as fielding “Level 8 boss cards”, which is internally inconsistent — Levels 8 and 9 are Guardian Force cards, not boss cards.

The Disc 4 payoff, which is the real reason to do this

With the quest complete, the whole CC Group boards the Ragnarok and can be replayed indefinitely, playing rare cards you are missing. Four sources.

The left-hand Diamond twin is the special one. She ignores the normal shuffle and will play any rare card not currently in your inventory — missed, lost or refined away. Four sources. One adds she serves them lowest-level-first; another specifies Levels 8 through 10 including modded cards.

One source alone excepts PuPu from what she can produce, against three that state no exception.

Each member on the Ragnarok also belongs to a region for rule purposes — Jack to Balamb, Club to Dollet, Joker to Centra, Diamond to Trabia, Spade to Esthar, Heart to Fishermans Horizon, King to Lunar. Two sources. That makes the Ragnarok roster a rule-carrying hazard in a place you would not expect one.

Completable on Discs 2 and 3, not Disc 4. This is the deadline that makes every other card decision safe or unsafe.

Card levels and rarity

110 cards, 11 per level, across 10 levels. Four sources agree on the total.

Three sources each for the common and rare halves. Level is not a strict power ordering — a high-level card is generally stronger, but the card tables have plenty of exceptions.

Three recovery routes for a lost rare, and they are not interchangeable: a rare lost to an ordinary NPC is won back by replaying that NPC; a rare lost to the Queen is won back from her son in Dollet; a rare lost at any point before Disc 4 is recoverable from the CC Group only if that quest is finished.

The hard missables, from the most complete published list: Shiva (Zone on the White SeeD Ship, requires the Girl Next Door magazine), Alexander (Piet, Lunar Base) and Laguna (Ellone, Lunar Base) — all three with Disc 3 deadlines — plus Selphie, from a classmate at the Trabia Garden fountain. One source adds PuPu as the rarest of all, requiring the Ragnarok, Enc-None, four UFO sightings and five Elixirs, and warns that fighting it instead of handing over the Elixirs forfeits the card permanently.

Every one of those is recoverable from the CC Group on Disc 4 — and only then.

What the game’s files actually settle

Very little of this page comes from the data, and it is worth being exact about which little.

Settled. Each enemy record carries three card bytes at offset 248, and they read as card, mod card, rare card — not as low, medium and high level bands. The proof is the always-obtains-card flag: seven enemies carry it, and every one of them has bytes 1 and 2 set to 255, meaning empty. Under a level-band reading those seven would give no card at all above level 20, which is impossible for bosses whose cards every player owns. Under drop/mod/rare they simply have a drop card and nothing else, which is what you would expect.

The practical consequence: carding an enemy is not level-banded the way its draw, mug and drop tables are. 101 of the game’s 142 enemy records change what they draw between level bands; the card bytes are a different structure entirely and do not.

Also settled from the same table: 81 of the 142 enemy records have no card at all, 48 have a drop card, an identical mod card and a rarer third, and 10 have only the first byte set — the boss shape.

Not settled, and nothing on this site pretends otherwise: the card ids are still numbers. The id-to-name table was resolved for magic, Guardian Forces and items, and that resolution explicitly does not extend to card ids or devour ids. So nothing about which enemy yields which card can be cited as settled from the datamine. Every card-holder claim on this page and on the area pages is published research, not file evidence.

Card Mod

Card Mod is a Quezacotl ability, 80 AP. It converts a card into items. 110 refine formulas exist, one per card. The wider refinement chains — what those items then become — are a separate subject and are not covered here.

Level 10, character cards. One card in, three items out.

Squall → Three Stars · Ward → Gaea’s Ring · Edea → Royal Crown · Quistis → Samantha Soul · Zell → Hyper Wrist · Rinoa → Magic Armlet · Selphie → Elem Guard · Irvine → Rocket Engine · Kiros → Accelerator · Seifer → Diamond Armor · Laguna → 100 Hero

Level 9, Guardian Force cards. One card in.

Bahamut → 100 Megalixir · Diablos → 100 Black Hole · Odin → 100 Dead Spirit · Pandemona → 100 Windmill · Cerberus → 100 Lightweight · Alexander → 3 Moon Curtain · Carbuncle → 3 Glow Curtain · Leviathan → 3 Doc’s Code · Phoenix → 3 Phoenix Spirit · Doomtrain → 3 Status Guard · Eden → 3 Monk’s Code

Level 8, Guardian Force cards. One card in.

Gilgamesh → 10 Holy War · Chubby Chocobo → 100 LuvLuvG · Angelo → 100 Elixir · MiniMog → 100 Pet House · Chicobo → 100 Gysahl Green · Quezacotl → 100 Dynamo Stone · Shiva → 100 North Wind · Sacred → 100 Dino Bone · Minotaur → 10 Adamantine · Ifrit → 3 Elem Atk · Siren → 3 Status Atk

The one rule that governs all of it

Do not refine a rare card until the CC Group quest is finished. Three sources call this the clearest irreversible mistake in the game, and the logic is not disputed by anyone: refine a Level 8, 9 or 10 card, reach Disc 4 without the CC Group aboard the Ragnarok, and that card is gone with no recovery path in existence.

Whether anything is unrecoverable even with the quest done splits three to one. Three sources say no — the left Diamond twin restores every refined, lost or never-collected rare, and one of them states categorically that it identifies no card as unrecoverable. One source says PuPu is the exception. One further source flags five cards as technically recoverable but agonising, because its author’s opponents rarely played them: Chubby Chocobo, Doomtrain, Irvine, Kiros and Phoenix.

The checklist, ordered by how often each mistake is cited

Each of these is treated above. Together they are the page in eleven lines.

1. Accepting the mix prompt without saving first — five sources.
2. Letting Random spread — four or more.
3. Refining a rare card before the CC Group quest is done — three.
4. Missing the Disc 3 card cutoffs: Shiva, Alexander, Laguna, Selphie — two.
5. Crossing into Disc 4 with the Queen somewhere you did not intend — three.
6. Not saving before a Queen match — three.
7. Playing under All or Direct with a rare collection while Random is live — three, none of which states it in one place.
8. Standardising every region to the same rule set — one, and the warning nobody repeats.
9. Assuming a soft reset rerolls the RNG — one for the mechanism, one for the Remaster.
10. Fighting PuPu instead of handing over five Elixirs — one.
11. Trusting the All → Direct → Diff → One ladder — two sources call it false, three still print it.

Game Notes

Everywhere else on this site, a disagreement between guides ends with a table settling it. This page is the one where it does not, and I want to be clear about why rather than apologise for it.

The rule-change algorithm produces exactly one observable per attempt: after a match, a rule either appeared, vanished, or did not. No counter, no log, no confirmation message, nothing on any screen that tells you a roll happened. A player running a hundred trials by hand gets a hundred single bits and no way to know whether a run of nothing meant “the check failed three times” or “the check never ran”. Every one of the four models fits that data. They are not competing readings of a number the way an elemental table is — they are competing readings of a silence.

So the four models are not four mistakes. They are four honest reconstructions of an unobservable process, and the reason three sources cite identical percentages is almost certainly that they share an ancestor, not that three teams measured the same thing.

What strikes me is that the practical advice converges anyway. Every source, whichever model it believes, arrives at the same instruction: save before you accept. Under Model A that is right because a third of outcomes are bad. Under Model B it is right because three weighted checks are not steerable. Under Model C it is right because you were told it is random. The models disagree completely about the machine and agree completely about what to do at the controller, which is a strong hint that the modelling was never the load-bearing part.

The design deserves a word too. FFVIII built a collection minigame whose state is global, shared across eight regions, mutated by a hidden roll after matches you may have played for unrelated reasons, with permanent one-of-a-kind items inside it and no undo. That is a system that is hostile to being documented — not by accident, but because the pieces that would let you document it, a visible counter or a confirmation line, were all left out. Twenty-five years of guide writers did not fail at this. They were handed a process with no output and asked to describe it.

And the files, which have settled most of what this project set out to settle, do not help here. They settled the shape of the card bytes and they will not tell us which card any number means. That is the honest end of the page: we know the three card slots are drop, mod and rare rather than level bands, and we know nothing at all about who holds what — from the data. Every card location you have read anywhere, including here, is somebody’s playthrough.