Shadow of the Colossus: The Boss Rush as Elegy

A boss rush is normally a bonus mode. You unlock it after the campaign, it strips out the corridors and the trash mobs, and you fight the good bits back to back until you die. It is understood to be dessert — the concentrated form of a game whose main meal was something else.

Shadow of the Colossus is a boss rush shipped as the whole product. Sixteen fights, no other enemies, no towns, nobody to talk to, no loot, no levels. Fumito Ueda’s team took the position that the corridors between the good bits were the part you could remove, removed them, and then did something stranger: they left the corridors in as geography. You still ride for four minutes to reach a colossus. There’s just nothing in the four minutes.

Eighteen years on, that decision is still the most misread thing about the game.

The emptiness is the load-bearing wall

Every review at the time noted the Forbidden Land was barren, and a fair number treated it as a limitation — a big pretty world Team Ico ran out of time to fill. It’s the opposite. The emptiness is doing the single hardest job in the design, which is making sixteen fights feel like sixteen events rather than sixteen encounters.

Run the counterfactual. Put wolves in the valleys. Put a village with a blacksmith. Put three collectibles on the ridge. Now the ride to Colossus 7 is a stretch of gameplay with its own texture and its own reward curve, and the colossus at the end is a spike in a line that was already busy. Your nervous system arrives at the fight already engaged, already fed, and the giant is one more thing happening.

Now take it all out. The ride is nothing. Wind, hooves, an enormous sky, the sword’s light. Fifteen minutes into a session your baseline has dropped through the floor — and then a two-hundred-foot creature stands up out of the ground. The contrast is manufactured. Boredom, precisely dosed, is what makes awe possible, and Ueda is one of the few designers with the nerve to spend a player’s attention that way.

This is the exact opposite bet from the one the modern open world makes. The tower, the icon, the constant drip described in why every open world puts a tower on the map, all exist to guarantee you’re never at baseline. Which means they can never do what Shadow of the Colossus does at the top of every hour. You cannot buy awe on credit. You have to save up.

The grip meter is the game

Strip the fiction and here’s the machine: a stamina bar, a climb, and a surface that is trying to shake you off.

That’s it. Every colossus is a variation on get on, hold on, get to the sigil, stab it before the bar empties. The bow softens things, the sword’s light finds the weak points, some fights need environmental setup to make the mounting possible at all. But the moment-to-moment tension in all sixteen is the same one: your grip is draining, the thing under you is bucking, and you have to decide whether to spend the last of the bar climbing higher or hunker down and let it calm.

It’s a beautiful economy because it’s legible in the body. You aren’t reading a number. You’re feeling a diminishing resource while being thrown around, and the decision — push on or wait — is the same decision a real person would make hanging off something. Ueda’s subtraction method from Ico is running here too: no health economy worth managing, no ability tree, no build. One resource, one verb, sixteen readings of it.

The failure of most boss-rush games is that they vary the enemy and hold the player constant. Team Ico held the player’s toolkit almost entirely constant across the whole game and let the colossi teach you what your unchanged tools can do. By Colossus 12 you have learned nothing new mechanically and yet you’re materially better, because your reading has improved. That’s a much rarer kind of progression than a skill tree, and it’s why the game’s descendants tend to be the ones that trust it — Armored Core VI’s boss-rush spine works the same way, dressing an escalating series of set-piece duels in a game that pretends to be about mechs.

The horse is an argument

Agro deserves her own paragraph because she’s the most divisive control decision in the game and the most deliberate. You do not drive Agro. You suggest things to her. Point her at a gap and she’ll refuse it; pull hard on the reins mid-gallop and she’ll carry her own momentum through the turn like an animal with mass and an opinion.

Players who bounce off this call it unresponsive, and by the standard of a vehicle it is. She’s a poor vehicle. The point is that she isn’t one. Every other horse in games at the time was a fast-travel skin with a gallop button, and Ueda’s team built a second body with its own inertia and its own reluctance — the same design instinct that produced Yorda, applied to transport. You spend hours negotiating with her, which is the only reason the game can later spend her.

It also does quiet mechanical work. Because Agro won’t simply obey, riding requires attention, which means the long empty crossings aren’t quite dead time. You’re doing something. It’s just something with no reward attached, which is a category the medium almost never uses.

Where it fights itself

The PS2 could not run this. That’s a fact of the record and everyone who played it in 2005 knows it: the framerate sags whenever the geometry gets ambitious, which is to say during exactly the fights that are supposed to be overwhelming. There’s a masochistic argument that the strain is atmospheric. It isn’t. It’s a technical ceiling, and Bluepoint’s 2018 rebuild is a better game for lifting it.

The camera fights you too. Locking on to something that fills the screen is a problem with no clean answer, and the answer here is a camera that swings wildly and occasionally hides the handhold you’re reaching for. Some deaths are yours. Some are the frame’s.

And a few colossi are duds. The number varies by taste, though almost everyone’s list includes at least one where the puzzle is opaque enough that you’re riding in circles trying to work out what the designer wants — the flying one in particular, which is either a highlight or a twenty-minute exercise in swimming, depending on how quickly you spot the trick. Sixteen is an ambitious count. Thirteen would have been an unimprovable game.

The record also says there were meant to be more — designs cut during development, some of which have been recovered by people picking through the discs. Knowing that, the pruning reads as an editorial act rather than a shortfall. Somebody in that team was ruthless.

Where to play it

The 2018 PlayStation remake is the version to start with: same design, rebuilt to run, and Bluepoint left the mechanics alone in the ways that matter. The 2011 high-definition release is the middle option and it’s fine. The PS2 original has the grain, the murk and the sag, and I’d only send a purist to it.

Play it in big sittings, ride rather than warp, and resist looking anything up. The map does almost nothing on purpose — see the map screen as an admission of failure — and the sword’s light is a better navigation system than any minimap that has shipped since.

Spoilers below

Wander has brought a dead girl, Mono, to the Shrine of Worship and struck a bargain with Dormin, a voice in the rafters, to bring her back. The price is the colossi. Kill all sixteen and she lives.

The design’s real cruelty is structural. Each colossus you fell sends a black tendril into you and Wander gets visibly worse — paler, veined, staggering. The game never tells you this is bad. It just shows you, over ten or fifteen hours, a man degrading in exact proportion to your success. Every fight you win is a fight you should have lost. And the colossi themselves are, with a handful of exceptions, not attacking you. They’re asleep, or grazing, or wandering. You wake them up and climb them and put a sword in their heads and they thrash and bleed and go quiet, and the music that plays afterwards is a lament rather than a fanfare. Ueda scored your victories as funerals from the first one.

That’s the elegy in the title. The boss rush is a mourning structure — sixteen repetitions of the same act, each one costing you more, with the emotional needle moving from thrill to duty to something close to shame by the back half. If the game had a bestiary of trash mobs, that arc would be impossible, because you’d have a baseline of guiltless violence to measure against. Sixteen kills and nothing else means every act of violence in the game is one you chose and the game made you look at.

Then Dormin reassembles inside Wander, Lord Emon’s men run a sword through him, the seal goes up, and the whole thing washes out into a basin with a horned baby crawling out of it — which is either an Ico prequel or a shrug, and the ambiguity has kept the argument alive for eighteen years.

Mono wakes. That’s the joke. The bargain was honoured. You get exactly what you asked for, and Ueda spends the entire runtime making sure that by the time it arrives you’ve stopped wanting it.