Ultimecia's Castle

Story mode. Written for a first playthrough and kept free of spoilers — it will not tell you what happens here, only what the game is asking of you.

The last place. You arrive and the game takes away almost everything you can do.

Eight seals, and you choose the order

Items, magic, summons, drawing, your command abilities, limit breaks, reviving the fallen, and saving. All sealed.

There are eight creatures in the castle. Beat one and you pick any one seal to break.

That is worth saying clearly, because almost nothing does: no boss is attached to a particular seal. You are not hunting for the one that gives you magic back. Beat whichever you can reach and choose whatever you need most.

Most people take Draw early, and there is a good reason for that below.

You are not trapped

Walk out of the front door and the seal lifts completely. Everything comes back while you are outside.

This is widely misunderstood — a lot of players think entering the castle is a one-way door. It is not. There are portals leading back out to what is left of the world map, and while towns are gone, the chocobo forests, the ruins, the optional deep-sea dungeon and the card club on the airship are all still there.

So if you arrive underprepared, leave and prepare.

Why Draw first

Seven Guardian Forces can be drawn from seven of the eight bosses in this castle, and for every one of them this is the last chance in the game.

Beat a boss without drawing from it and that summon is gone for this playthrough. There is no recovery. The one exception is the pair at the front door, which holds no summon at all — so if you get caught out once, that is the fight you can afford to have rushed.

One of them is the strongest Guardian Force in the game, held by the dragon at the top of the clock tower. If you skipped the optional undersea dungeon earlier, this is your only remaining chance at it anywhere.

So: break Draw early, and draw from every single boss before you finish it.

You do not have to fight all eight

One source says so plainly and every walkthrough ignores it: the minions can be run from.

You only need as many seals as your plan actually requires. If your party is well built you can reach the final door having broken very few of them.

Things worth knowing before you go in

• Walk, do not run, across the bridge with the key on it, or it falls into the room below. You can go and get it, but it is an annoying trip.
• The four chests in the vault are empty. The puzzle exists only to summon something. Do not spend time hunting for treasure in them.
• The party splits in two, and the lift puzzle is decided by how heavy each group is. Guides mostly describe it as a boy/girl split, which works by accident because the three men happen to be the three heaviest. Nothing can go permanently wrong — go back to a green circle and reshuffle.
• Give the second party something that stops random encounters. It has to cross the castle on its own.
• There are thirteen draw points and eight of them never refill, so lead with your best magic user when you take those.
• Mug is worthless in here. Only one fight in the castle can be stolen from at all, and it gives an Elixir. Every boss drops its item on the kill instead, always the same item, so there is nothing to reload for.

The optional fight

There is a bell in the castle. Ringing it summons something with over a million health that absorbs all eight elements — so strip every elemental attack junction before you try, or you will be healing it.

It is worth 250 ability points, which is more than anything else in the game by a wide margin.

It is also, honestly, a bragging-rights fight. Nothing is locked behind it — the item it drops can be got in at least three other places.

Game Notes

The best final dungeon idea in the series, and I will defend that.

Most games end by handing you everything. Your strongest gear, your best summons, a party at full power, and a corridor to walk down. The last hour is a victory lap with the difficulty curve bent to let you enjoy being strong.

This one takes it all away and makes you buy it back — and crucially, lets you choose the order. That single decision turns the endgame into a real question. Do you want your items back, or your limit breaks? Can you fight three more bosses without being able to revive anybody? Is it worth going without saving for another hour to get magic sooner?

Every one of those is a genuine trade, and the answer depends on how you built your party across the previous forty hours. A junction-heavy party barely needs magic back. A party that leaned on summons is crippled until GF returns.

It is the only part of FFVIII that asks you to account for the way you have been playing. And the reason it works is the same reason everything else in this game works or doesn’t: it never explains itself. There is no tutorial telling you the choice is free and unordered. Most people work it out on the second or third boss.

Completionist mode. Full spoilers, in play order, with every permanent loss called out before you reach it. 16 entries below are marked where our sources disagree with each other — we have not picked a winner for you.

Spoilers from here.

The headline: one source, seven fabricated immunities

One widely-copied database publishes an Earth immunity on seven separate creatures in this dungeon — the cellar boss, the terrace boss, the gallery boss, the dragon at the top, two forms of the final boss, and the pair of summoned creatures in the last fight.

Not one of them has it. Every one reads Earth at exactly 100% in the game’s own elemental table.

This is worth more attention than a single wrong number, because of what makes it dangerous: the same source’s other entries for those enemies are largely accurate. It quantifies the cellar boss’s Fire and Ice weaknesses at 400% and it is exactly right. It gives the gallery boss 200% from Wind and it is exactly right. It is a good source with one repeated error — and a consistent error propagates further than a random one, because it looks like a pattern somebody verified.

Correction: the published “maximum HP” figures are not maximums

Sources publish a maximum for each boss in this castle and cap them at oddly specific levels — 51, 55, 56, 57, 58.

Those are not maximums. They are the value at whatever level cap the guide assumed.

The vault boss is the cleanest demonstration. Three sources publish 60,000 at level 100; one publishes 38,500 at level 57. 38,500 lands on its curve at exactly level 57. It is not a disagreement at all — it is two guides assuming different caps for the same enemy.

And this is a distinct mechanism from the level scaling this project has been documenting. Ordinary level scaling is two observers at different party levels writing down different numbers. This is guides publishing a ceiling that does not exist, and then a later reader treating one of them as wrong.

It also means the dragon at the top, universally published as an 89,600 HP fight, is a 140,000 HP fight if you arrive at level 100.

Correction: three bosses do not scale at all

The pair at the entrance are 10,000 HP at every level from 1 to 100. The one behind the prison key is 30,000 at every level. Flat.

That is why their figures are the best-agreed in the castle — there is no curve for anyone to sample differently. Which, incidentally, is the cleanest available proof that the disagreements everywhere else in this guide are caused by scaling rather than by carelessness: remove the scaling and the disagreements vanish.

One source publishes the entrance pair at 12,000+; five say 10,000 flat. The five are right.

The eight seals

Seven sources enumerate eight and agree on the list: Item, Magic, GF, Draw, command abilities, limit breaks, revival and saving. One guide describes only five and never finishes the list; two others list all eight without stating a number.

The part almost nothing states plainly: no seal is tied to a particular boss. Each victory grants one free choice from whatever is still locked. Seven sources describe it this way, and what varies between guides is only their recommended order. Any page implying a fixed mapping is wrong.

Two sources alone note that leaving the castle lifts the seal entirely. The other eleven omit it or imply the seal is absolute. You can walk out, prepare, and come back.

Portals back out are disputed: two sources say four and two say three. Only one names all four, and one of those four appears in no other source. Towns are gone on this disc; the chocobo forests, the ruins, the optional deep-sea dungeon, the card club aboard the airship, refining and stat-building all remain.

You do not have to fight all eight

One wiki states plainly that the minions can be fled. Every walkthrough presents all eight as mandatory. You need only as many seals as your plan requires, and community runs exist that clear the castle having broken very few.

The boss table

Every figure below is read from the game’s files. AP is 30 across the board except where noted, and every fight in this castle pays zero experience — only one source publishes an EXP figure for these bosses at all.

Three corrections from that table

The armory boss’s AP. Two sources publish 42, one publishes 40, one publishes 30. It is 30 — the lone minority source is right and the three-to-one majority is wrong. That is the third time in two batches a minority of one has won. Its escorts and the gallery escorts all pay zero, against one source giving the escorts 30 each.

The terrace boss and Ice. Four sources say it nullifies or resists Ice. One publishes it as weak to Ice — the exact inverse. It is fully immune, and nothing else is modified at all. Following the outlier means bringing the one element that does literally nothing.

The entrance pair are two records, not two phases. Three sources treat them separately and three merge them, and the tables settle it more sharply than “different drops” does:

One can only be robbed and one can only be killed. Two enemies sharing a health bar do not have opposite loot routes, so the three sources that separate them are right.

Leave Mug at home, and take the drops

Nobody describes this dungeon’s loot as a shape, and it has a very clear one.

Exactly one boss record in the castle can be mugged. Tri-Point, Krysta, Trauma, Gargantua, Red Giant, Catoblepas, Tiamat, Sphinxara and Omega Weapon all read a mug rate of zero with an empty table. The only steals available in the whole dungeon are an Elixir off the first entrance record and two Meteor Stones off Droma. The command slot is better spent on something else.

What the bosses give instead is one ability item apiece, identical in all four rarity slots at every level — no roll, no reason to reload:

Across all 142 enemy records in the game, a zero rate byte and an empty table coincide without a single exception. That is how we know an empty table is the design here rather than a gap in our reading of it.

The seven Guardian Forces

Six sources list all seven and agree: this is the last chance at every one of them. Beat the boss without drawing and it is gone for the playthrough. One source singles out one of the seven as “permanently missed” rather than “last chance” and does not explain the distinction.

Unseal Draw early. Two sources recommend it for exactly this reason and it is the strongest routing advice in the dungeon.

And the mapping is readable now, which no source publishes as a table:

Every one sits in the draw list at all three level bands, so none of them has a level requirement. And the entrance pair is the answer to a question no source asks out loud: “seven Guardian Forces from eight bosses” means the fight at the door is the one you can finish without losing anything.

Eden, from the dragon at the top, with no dissent across eight sources — its list is Flare and Eden, two entries deep at every band. One wiki states its only two sources in the entire game are this dragon and the optional deep-sea superboss. That checks out exactly: Eden appears in precisely two draw lists in the whole enemy table and nowhere else. So if you skipped that dungeon, this is the single highest-stakes action in the castle.

The final sequence

Four sources say four battles; one says three. One source describes an extra stage with two summoned creatures, which is plausibly where the disagreement comes from.

Three of the published forms match records exactly at level 1 — 4,600, 13,050 and 16,505. The other two published forms are stored under names our text table has not decoded, so we can confirm three of the sequence and not all of it.

What the records add:

Filling a documented hole

No published source gives an AP or experience figure for any final-boss form. The records settle it: every form pays zero of both — and none of them has a mug or a drop table either. Nothing whatever comes off the last fight: no experience, no ability points, no steal, no drop.

That is not an oversight in the design. It is the last fight, and there is nothing after it to spend anything on.

Apocalypse

The only source of this spell in the game, drawable from the lower half of the final form, which appears without warning partway through the fight.

Three sources tell you to kill that half quickly to deny it the spell. Two say to draw the spell first. Both are correct for different goals and no source says so — killing it fast is the survival play, drawing first is the completionist one.

One limit worth recording so that nobody later reads it as a contradiction. This spell appears in no enemy’s draw list anywhere in the game, and the final form’s own list holds Flare and Holy. That does not put five agreeing sources in doubt. It means the draw is not stored where draw lists are stored — the same limit that keeps draw points unreadable — so the records can neither confirm it nor deny it, and a future reader should not “correct” this page from them.

Omega

111,105 HP at level 1, 202,500 at level 10, 1,161,000 at level 100. The level-10 figure matches one guide’s published anchor exactly.

250 ability points — by far the largest single award in the game — and zero experience.

It absorbs all eight elements. Every source agrees, and the consequence is the one piece of preparation that matters: strip every elemental attack junction before ringing the bell, or you are healing it.

Two sources give the timer as sixty seconds and two as about forty-five. Three descriptions of the bell’s location are published and they are not compatible.

One wiki alone flags a release difference: that this enemy is always level 100 on the original console release but variable on the PC and remastered versions. That matters for exactly the release this guide is written against — a level-1 figure of 111,105 is meaningless on a build where the enemy is always level 100, and meaningful on one where it is not. The record holds a normal curve, which is consistent with the claim, though the record alone cannot tell us what the encounter does with it.

What beating it gives you: an item and the ability points. No source claims any unique Guardian Force, spell or ability is gated behind it.

The item is Three Stars, in all four rarity slots at every level — the same item the Obel Lake quest on area 25 is built around, and the same item an item database lists three other sources for. It cannot be mugged: mug rate zero, empty table. Its draw list is Flare, Holy, Meteor and Ultima at every band, all four of which the far islands hand out for nothing.

So there is no item reason and no magic reason to ring the bell. One wiki adds an entry to the in-game help section. It is a bragging-rights fight.

The castle itself

Thirteen draw points, and the published counts run from four to thirteen. One guide lists only four and names one of them as a spell no other source reports. One draw-point FAQ puts a different spell in the gallery than four other sources do. One source places a point in a completely different wing from the five that agree on it.

Two database-style sources agree exactly on which eight never replenish (Flare, Aura, Meteor, Meltdown, Ultima, Full-Life, Holy, Triple) and which five do (Slow, Cura, Curaga, Dispel, Stop). No walkthrough states replenishment for any of them. Four of the eight one-shot points need the hidden-item ability.

Five key items. Sources give three different bridges for the balanced key and three different screens for the final chest; three say the fountain key only appears once the sluice is shut and two describe it as simply found. Walk, do not run, across the bridge — every source that mentions the drop confirms the key is retrievable from the room below.

The four chests in the vault contain nothing. Six sources agree the puzzle’s only payoff is summoning the boss, and three incompatible solutions are published for the order. The only real chest is the ability item at the end of the drained waterway.

No cards anywhere. No opponent, no prize, no pickup. The castle bosses have card records with refinement outputs, but those cards are not obtainable from the fights.

The party split

Two parties of three. The lift puzzle is the dungeon’s most consequential disagreement:

The gender shorthand produces the right answer most of the time because the three male characters happen to be the three heaviest, which is almost certainly how it spread. Only one source has the actual numbers.

Nothing can soft-lock — return to a switch point and reshuffle. Two sources recommend equipping the reserve party with the encounter-suppressing ability, since it crosses the castle alone.

For the final battle you do not choose your party — three sources agree the game picks at random. A character left unconscious too long is removed for the rest of the fight and replaced from the reserves, and two sources note this can be used deliberately to cycle out people you did not want. No source explains what happens if everyone is removed.

One source reports two crash conditions — a particular character in a leader position making the party disappear, and another locking the game up when the chandelier falls. No other source mentions either, and these are the only crash claims in the entire body of research for this guide.

Game Notes

Twenty-eight areas, and the last correction is the same correction as the first.

The very first thing this project established about FFVIII was that its enemies scale, that the scaling is invisible from inside a playthrough, and that this single fact generates contradictory records from people who all observed correctly.

Here at the end it produces one final variation I had not anticipated. Guides publish a maximum HP for each boss — and those maximums are not maximums. They are the value at whatever level cap the writer assumed, and different writers assumed 51, 55, 56, 57 and 58. So five sources agree the dragon at the top has 89,600 HP, and that is its value at level 58, and at level 100 it has 140,000.

Nobody measured wrong. Everybody described a ceiling that does not exist.

And then, sitting right next to it in the same dungeon, the control group: three bosses that do not scale at all. Flat HP, level 1 to level 100. Their figures are the most reliably agreed in the castle, across every source, with no dissent.

Remove the scaling and the disagreements disappear. That is as close to a proof as this project is going to get that the community’s problem was never carelessness — it was a mechanism nobody could see, doing its work quietly under twenty-five years of careful observation.

The other thing I want to record before this vertical closes: one source has the same fabricated Earth immunity on seven creatures in this one dungeon, and it is otherwise one of the better sources we used. It got the 400% Fire and Ice weakness exactly right. It got the 200% Wind exactly right. It has one repeated error, and because the error is consistent it reads like something somebody checked.

That is the failure mode I would most want a reader to take away from all of this. A source that is wrong at random gets caught. A source that is wrong the same way every time gets copied.