The Deckbuilder Boom After Slay the Spire

The most important thing in Slay the Spire’s shop is not a card. It’s the little service at the bottom of the screen where a man with a beard will, for gold, remove a card from your deck permanently.

Everything else in that shop makes your deck bigger. That one option makes it better, and the distinction is the entire genre. Six years of imitators have copied the cards, the relics, the branching map and the three-act structure, and a startling number of them have shipped without understanding that the scissors were the load-bearing wall.

Dominion did it on a table first

Donald X. Vaccarino’s Dominion came out in 2008 from Rio Grande Games and took the Spiel des Jahres the following year, and it invented the thing. Before Dominion, a card game meant a deck you built before you sat down — Magic’s sixty-card list, agonised over for a week, then shuffled and played. Dominion put the construction inside the game. Everyone starts with the same shabby ten cards, seven Coppers and three Estates, and the play is buying better ones from a shared market into a deck you’ll shuffle through again in three turns.

And Dominion, on day one, contained the insight. The Chapel costs two coins and lets you trash up to four cards from your hand, and it is one of the most notorious cards in the game, because the correct opening in a huge number of Dominion setups is to buy a Chapel and then set fire to the deck you were given. Your Coppers are the problem. They’re what you draw instead of the good thing.

That’s a subtraction game wearing a purchase interface, and it was solved in 2008 by a man designing on a kitchen table.

What Mega Crit actually invented

Slay the Spire went into early access in November 2017 and out of it in January 2019, from Mega Crit — Anthony Giovannetti and Casey Yano, two people. They’ve been consistently open that the direct ancestor is Dream Quest, Peter Whalen’s 2014 iOS game, which is single-developer, hand-drawn in a way that looks like a dare, and which put the deckbuilder and the roguelike run in a blender before anyone else did.

What Mega Crit added is the map.

Look at an act of Slay the Spire before you take a step. The whole thing is laid out: monster nodes, elite nodes, question marks, campfires, a shop, the boss at the top. You pick a route. Elites drop relics and might kill you. Campfires let you rest or upgrade a card, one or the other, never both. The question marks are a gamble. So before you’ve fought anything you’ve already made the game’s central decision — how much risk you want to buy, and what you want to be paid for it — and you made it with full information about the shape of the act and no information about your luck.

That’s a pre-commitment device, and it’s the piece almost nobody copies properly. It converts a roguelite run from a sequence of surprises into a plan you’re watching survive contact.

The rest of the design is arithmetic with sharp edges. Three energy a turn, draw five, discard the rest. The Ironclad opens with five Strikes, four Defends and a Bash, and every one of those Strikes is a card that isn’t the card you want by act two. So the run is a slow argument between addition and subtraction: the game showers you with rewards after every fight, and taking them all is how you lose.

Why everybody could suddenly make one

The boom has an unglamorous cause. Deckbuilders are the cheapest genre on the shelf to build.

No animation budget worth the name. No physics. No engine work — Slay the Spire shipped on LibGDX, a Java framework you’d use for a jam game. Turn-based, so no netcode, no frame timing, no input latency to agonise over. The content is text on rectangles. A two-person team can ship a complete, deep, 200-hour game with a codebase you could read in a fortnight, and Steam will sell it to a hundred thousand people at fifteen quid.

That’s the same economic argument that made roguelites eat the market in the first place, pushed one step further. The generator replaced the level designer; the card replaced the generator. What you’re paying for is one designer’s judgement about numbers, and judgement is the cheapest thing in this industry to buy and the hardest to find.

The legibility thing

Here’s why the deck won, and it’s a point about the player’s head rather than the developer’s budget.

In an action roguelite your build is a soup. Twenty-eight modifiers stacked on each other, most of them invisible, interacting through code you can’t inspect. You’re strong and you don’t know why. You die and you learn nothing, because the causal chain from “I picked that up in room four” to “I got flattened in room nineteen” is unreadable.

A deck is a legible state. You can open it. You can count it. Twenty-four cards, and you know that the Whirlwind you paid three energy for shows up once every five turns, and you can do the sum. When a run collapses in a deckbuilder, the corpse is diagnostic — the deck is right there, and the reason is usually sitting in it, and you can see that you took eleven cards and removed two.

That’s the thing the genre gave people, and it’s why it recruited an audience that had bounced off everything else in the roguelike lineage. Losing became instructive rather than mysterious. You can be taught by a deckbuilder in a way an action roguelite structurally cannot manage.

The boom’s failure mode

Six years of these things and the pattern in the weak ones is consistent: they balance cards and forget to balance the deck.

Give a designer a card and they’ll ask whether it’s too strong. That’s the easy question and the wrong one. The real question is what happens to the rest of the deck when this card is in it, because a deckbuilder is a game about a probability distribution and every card you add moves the distribution. A card that’s individually reasonable and dilutive is a trap, and the genre’s best moments are all traps of exactly that shape — the flashy rare you take because it’s flashy, which quietly costs you the consistency that was keeping you alive.

Games that get this build the friction in. Monster Train (Shiny Shoe, 2020) stacks a vertical battlefield on top so position competes with the deck for your attention. Inscryption (Daniel Mullins, 2021) does something stranger: it keeps breaking its own frame, so the deck you’re optimising is repeatedly revealed to be the wrong object to be optimising. Both of them found a way to stop the player from just accumulating.

The weak ones hand you rewards and let you take them, and the run tips over into a hundred cards of nothing, and the player can see it happening and has no scissors.

The twenty rungs

One more thing Mega Crit got right, and it’s the same insight that keeps the better roguelites honest: the game gets harder as you get better.

Beat the Spire once and it unlocks Ascension 1. Beat that and you get Ascension 2. There are twenty rungs, and each one adds a specific rule — elites hit harder, you start damaged, the shop charges more, bosses get an extra beat. Ascension 20 is a different game from Ascension 0 played by a different person, and the only thing you carry up the ladder is your own competence.

Compare that to the manor-upgrade model, where the reward for persistence is a larger health bar. Mega Crit’s reward for persistence is a smaller margin. That is the roguelike contract, smuggled back into a roguelite through the achievements screen, and it’s why the game has a decade of life in it while the imitators are on sale.

The four characters do similar work by refusing to be the same distribution. The Ironclad wants a small deck of big hits. The Silent wants a large deck that draws itself. The Defect wants an orb engine and the Watcher, added in 2020, wants stances and will kill you for standing in the wrong one. Four characters, four different arguments about what a deck is for — genuine content from the same codebase, produced by moving numbers.

Where it went

The interesting descendants have drifted off the deck entirely, which is the proper sign of a healthy idea.

Balatro (February 2024) keeps a standard 52-card deck as its substrate and moves the build into the Jokers — five slots, a hundred and fifty rules, and a score that’s a multiplication rather than a sum. It’s a deckbuilder where the deck is almost furniture, and it’s eaten this year because LocalThunk understood the legibility point better than most of the field: everything that’s happening to you is on the table, in objects, in an order you chose.

Luck be a Landlord did it with a slot machine. Backpack Battles did it with a rucksack. Fights in Tight Spaces did it with a John Wick corridor. The chassis travels, because what’s underneath it is old and portable: a bag of stuff you edit, an engine that draws from the bag, and a shop that will sell you subtraction if you’re clever enough to want it.

That’s the genre’s actual canon in one sentence, and Vaccarino wrote it on a card that cost two coins.