The most aggressive thingPersona 5 Royal does to you happens on an ordinary Tuesday in June. School finishes. The game returns control. You have one afternoon and one evening, and in front of you sit a part-time job that would raise your Charm, a friend who has been waiting three in-game weeks to advance a relationship, a book that would raise your Knowledge before an exam, and a dungeon with a deadline in eleven days. You can do one afternoon thing. Some of those options will eat the evening as well.

Nothing threatens you. No enemy is on screen. This is the most tense the game ever gets, and Atlus knew that when they built it.

The system, stated plainly

Persona 5 came out in Japan in September 2016 and in the West in April 2017, directed by Katsura Hashino, with Shigenori Soejima’s character design and Shoji Meguro’s soundtrack doing an enormous amount of the heavy lifting.Royal is the expanded edition — Japan in October 2019, the West in March 2020 on PS4, and eventually everywhere in October 2022 when Atlus finally put it on PC, Xbox, Switch and PS5.

The structure is a school year. Each day gives you a small number of discretionary slots. You spend them on Confidants — twenty-odd relationships, each ranked one to ten, each granting mechanical benefits as it climbs — or on the five social stats, Knowledge, Guts, Proficiency, Kindness and Charm, which gate the Confidants. Meanwhile the plot delivers Palaces: cognitive dungeons with a hard calendar deadline. Fail to finish one by its date and the game ends. Actually ends.

So every single day is an allocation problem with an audit at the end of it. That is the machine. The phantom-thief business, the jazz, the extraordinary menus — all of it is upholstery on a spreadsheet, and the spreadsheet is where the feeling comes from.

Why the scarcity works

Scarcity is easy to design and hard to makehurt. Most games with a time limit produce anxiety, which is a cheap emotion.Persona 5 produces regret, which is expensive.

The mechanism is that the game makes the thing you gave up visible. You skipped Yusuke, who texted, and whose Confidant is at rank four, and whose rank five would give you a combat ability you can name and want. The cost of every choice is a person with a face and a text message. That is the whole trick, and it is why the strategy-game version of this system, where you allocate abstract workers to abstract buildings, does not feel like anything.

The second mechanism is that the game refuses to let you win the allocation. A completionist run of Royal is documented to sit well past a hundred hours, and even a perfect one is a scramble, because the Confidant ranks are gated behind social stats that are themselves gated behind days. You are always behind. The design’s honest position is that a year is too short to be good at everything, which is a considerably more mature statement than the phantom-thief plot manages in its entire runtime.

The real ancestor of this is not the RPG lineage at all. It isTokimeki Memorial, Konami’s 1994 dating simulator, which established the loop of stat-raising against a school calendar with affection as the scoring function.Persona 3 imported that structure into Atlus’s demon-fusion RPG in 2006 and discovered that the calendar made the dungeonmatter, because now the dungeon had an opportunity cost. EverythingPersona 5 does is the third refinement of that fusion. Fifteen years earlier, on the home computers I grew up with, the closest equivalent wasElite’s fuel economy — a resource that turned exploration into a decision — and the shared idea is old: the interesting number is the one you cannot have more of.

Where Royal fights itself

Here is the argument I want to make about the definitive edition, and it is not a comfortable one.

Royal’s improvements are almost allloosening. Palaces now contain Will Seeds, which reward exploration with SP-restoring accessories; SP was the original’s scarcest combat resource and the reason a Palace took multiple visits. The grappling hook opens shortcuts. Safe rooms let you leave and return without losing progress. Morgana’s early curfew — the notorious business of a cat sending you to bed — is relaxed. Ranked SP items are purchasable. Showtime attacks give you a free burst of damage on a random timer.

Each of these is a quality-of-life win. Collectively they mean that a Palace which used to consume three or four calendar days now goes down in one, and those recovered days go straight back into your Confidant budget. Atlus made the dungeons kinder and thereby made the calendar softer, and the calendar was the antagonist.

I do not think this ruinsRoyal. I think it is a real cost that reviews at the time under-reported because the additions were so obviously generous. The 2016 game’s cruelty was doing structural work: when SP ran out, the Palace ended, and you went home having spent a day and achieved nothing, andthat was the day you learned what the game was about.Royal rarely gives you that day. It replaces it with a third semester — new content from November onwards, a new Confidant in Takuto Maruki, a new party member in Kasumi Yoshizawa — that is, and I say this having sat with it for years, the best-written material Atlus has ever shipped and structurally the wrong place to put it.

Because the third semester arrives after the calendar has finished threatening you. It is a coda. It is superb, and it is playing in a mode the entire preceding hundred hours had been arguing against: a stretch of time where the pressure is narrative rather than arithmetical. The best thing inRoyal is the part ofRoyal that stops beingPersona 5.

The upholstery, briefly

I have been dismissive about the surface and should correct that, because the surface is why anyone tolerates the spreadsheet.

The interface is the most confident work in the medium. Every menu is animated, angled, red-and-black, scored, and no two transitions in the game are the same shape. That is functional work: a game asking you to spend a hundred hours in menus has to make the menus a place you enjoy standing, and Atlus solved that by treating the pause screen as a piece of graphic design rather than a list. Meguro’s soundtrack does the same job in the other channel: the battle theme is the reason a random encounter you have fought two hundred times still snaps you awake.

And the writing, when it is good, is very good indeed. Futaba’s Palace, in particular, does something the genre almost never manages: it takes a character’s psychology, renders it as architecture, and then makes navigating the architecture the act of understanding her.

When it is bad it is very bad. The plot’s politics are adolescent, the Phantom Thieves’ celebrity arc goes nowhere it has not been dragged, and there are stretches of the middle third where the game will hold you in a cutscene for forty minutes to say something it said at the start.

The verdict

Persona 5 Royal is a hundred-hour game about the arithmetic of a finite year, wearing the best clothes in the industry. Its greatness is entirely structural — the fact that a Tuesday in June is harder than any boss — and its expanded edition is a slightly compromised version of that greatness in exchange for content nobody could reasonably decline.

PlayRoyal, since it is the only version anyone can now buy, and play it on whatever is nearest; the 2022 ports run fine and the PC one is the obvious pick. Take the third semester. Do not look up an optimal Confidant route, because a schedule someone else calculated removes the entire game.

For a much smaller, much sharper treatment of the same idea — time as a currency you are always short of, with none of the upholstery — readCitizen Sleeper, which does in eight hours what this does in a hundred and is honest about which of those is a virtue. And for the other end of the JRPG’s structural ambition,Nier: Automata makes the shape of the playthrough itself the argument.

Spoilers below

Maruki is the reasonRoyal exists, and he is a better antagonist than Shido by a distance that is almost embarrassing.

The third semester’s premise is that the counsellor who has been helping you all year has acquired the power to remove suffering, and has used it, and the world you wake up in is one where everybody’s worst thing simply did not happen. The dead parent is alive. The ruined career is intact. The friend’s trauma is gone, and so is the person the trauma made.

What makes this land is that it is acalendar argument. Maruki’s reality is a place with no opportunity cost — every choice is available, nothing is foreclosed, no Tuesday in June ever charges you anything. He is offering the player exactly what the player has spent a hundred hours resenting the absence of. The game has trained you to want this. The whole system has been a machine for generating the specific hunger Maruki proposes to satisfy.

And the refusal — the choice to reject a painless world and go back to the one where things cost — is the only moment in the entirePersona series where the theme and the mechanics say the same sentence at the same time. Everything else inPersona 5, the rebellion, the masks, the chains, is a metaphor stapled to a combat system. This is not a metaphor. This is the game asking whether you understood what the schedule was for.

The answer arrives in February, in a fistfight in the sky, which is the most Atlus thing imaginable and does not diminish it in the slightest.

Somewhere around the second act ofPathfinder: Wrath of the Righteous, the game stops asking you what kind of hero you are and starts asking what kind ofthing you are willing to become. It offers, among other options, becoming a lich. This goes considerably past an evil alignment tick-box: the game grows a new questline, a new set of abilities, a new attitude from your companions, and a substantially different ending. It does this for ten separate answers to the question.

Owlcat Games released this in September 2021, three years afterKingmaker, funded by a Kickstarter that pulled in around two million dollars. It is adapted from Paizo’s tabletop adventure path of the same name and built on Pathfinder’s first-edition ruleset, which is to say the ruleset that took Dungeons & Dragons 3.5 and asked what would happen if nobody ever said no. That question is the entire aesthetic of this game, and I want to defend it.

The sheet is the toy

Twenty-five base classes. Archetypes on top of those in numbers that require a wiki. Multiclassing with no meaningful guard rails. Feats that combine into things the designers plainly did not sit down and enumerate. If you have spent any time in thePathfinder build community you will know the folk canon: the Scaled Fist monk dip that turns your charisma into armour, the Vivisectionist alchemist who is a rogue with better chemistry, the Sword Saint whose entire job is to make one attack per round mathematically obscene.

The honest criticism of this is that it is not balanced, and the honest answer is that balance was never the promise.Wrath of the Righteous is not a game about a curated encounter budget. It is a game about a character sheet as a construction kit, where the pleasure is the pleasure of ademoscene release — someone found a hole in the rules and drove a lorry through it, and everybody else gathered round to admire the lorry.

The real ancestor here is SSI’s Gold Box run on the eight- and sixteen-bit machines.Pool of Radiance in 1988 gave you a party, a rulebook, and no interest whatsoever in whether your combination of decisions was sensible; the manual assumed you would read it and the game assumed you had. Everything that came after — theBaldur’s Gate line, and especiallyThrone of Bhaal’s High Level Abilities, where 2000-era BioWare quietly conceded that the endgame of a build game is absurdity — descends from that permission. Owlcat picked the permission back up after twenty years of the genre apologising for it.

What the mythic paths actually do

The headline is the ten mythic paths, and it is worth being precise about their mechanical shape, because “branching narrative” undersells it.

You commit around Act 2. Aeon polices causality and can retroactively unmake things. Azata is a chaotic good party that summons a dragon called Aivu who grows up over the campaign. Demon eats your enemies’ powers and your own restraint. Trickster rewrites the game’s jokes into rules — it can make a critical hit on a Perception check literally see through the plot. Lich raises your dead enemies and quietly poisons every companion relationship you have built. There is also Gold Dragon, Angel, Devil, Swarm-That-Walks and Legend, which discards mythic power entirely to become spectacularly good at ordinary things.

The design read: a mythic path is alens, applied at the midpoint, that recolours content you were going to see anyway. That is an enormously efficient piece of engineering. Owlcat did not build ten campaigns. They built one campaign with ten sets of rules about how you are allowed to interact with it, and because the paths land at Act 2 rather than at character creation, you have already met everyone before the lens goes on. Your companions therefore have opinions about the change. Regill, the hellknight, approves of order arriving from any direction. Arueshalae, the succubus trying to stop being one, has a complicated time watching you become a demon.

This is the thingWrath does that no other CRPG of its era matches, including the one that sold ten times as many copies.Baldur’s Gate 3 has better faces, better cameras, better everything you can photograph.Wrath of the Righteous has more consequential branching per pound, and the branching is welded to the build rather than sitting beside it.

The crusade, and the argument about it

Then there is the army layer. In between the dungeons,Wrath hands you a strategic map, a recruitment budget, stacks of crusader units, and a set of turn-based battles that play like a thriftyHeroes of Might and Magic. It is optional in the sense that you can set it to resolve automatically. It is not optional in the sense that it sits in the middle of the game asking for your time.

Most reviews at the time called it a mistake. I think that is too quick, and the reason is thematic rather than mechanical. The adventure path is about acrusade — a mass mobilisation of ordinary people against a demonic incursion — and every CRPG convention pushes against that theme, because the CRPG convention is that six exceptional individuals resolve everything personally. The crusade layer is the game’s way of insisting that the war exists whether or not your party is in the room. When your mythic power scales up the units you can field, the two layers finally touch, and the campaign map stops being homework.

The problem is the tuning. The economy is opaque, the general system rewards a couple of obviously correct picks, and the layer arrives before you have any reason to care about it. Owlcat’s own answer — the Enhanced Edition update in 2022, and the option to skip it — is a studio admitting the theory outran the execution. I would rather a studio reach for something structurally interesting and land it two-thirds of the way than ship the safe version.

Where it fights itself

The launch was rough. This is a matter of public record and Owlcat spent the following year patching it hard: scripting faults, encounter bugs, save issues, the usual toll of a game with this many interacting systems shipped by a mid-sized studio. It is in a substantially better state now, with the Enhanced Edition and five DLC releases behind it, and the console versions arrived in 2022 with the fixes baked in.

The deeper structural fault is the difficulty.Wrath is a game whose default settings assume you have read the rulebook, and whose enemies from Act 3 onwards start layering resistances, spell immunities and attack routines that will simply erase an unoptimised party. The customisable difficulty sliders are the most important feature in the game and the least advertised: you can tune enemy stats, damage taken, and the swinginess of the maths independently, and doing so is the correct response rather than an admission of anything. The record puts a full run somewhere well past a hundred hours, and a hundred hours is a real thing to ask of someone’s life. Any game asking it should let the asker set the terms.

And it is long past the point of shame in the fourth act. The Midnight Isles content, the drift into demon-realm sameness, the sheer volume of trash encounters between the good ideas — this is a game that would be better at eighty per cent of its length and does not believe that for a second.

The verdict

Wrath of the Righteous is the most generous CRPG of the last decade, and generosity is a real virtue with real costs. It gives you more rules than you can hold, more branches than you can see in one run, and more army admin than you asked for, and it does so out of a conviction that the player is an adult who can decline the parts they dislike. That conviction is rarer than good writing and considerably rarer than good faces.

Play it on PC where the mods and the build community live. Turn the difficulty sliders to something honest before Act 3 rather than after it. Pick a mythic path that scares you slightly, because the game is at its best when the answer to “what are you becoming” is one you are not entirely comfortable with.

For the opposite approach — an RPG where the character sheet has almost no combat function at all and is instead a set of arguments you have with yourself — seeDisco Elysium. The two games are as far apart as the genre stretches, and they are both right.

Spoilers below

The Trickster path is the one I want to single out, because it is a joke that turns into a thesis.

Trickster’s mythic abilities are structured as punchlines: improving your Perception until you notice things the plot did not intend you to notice, improving Knowledge until enemies are humiliated by trivia, turning the rules themselves into a bit. It reads as the comedy option for about ten hours. Then the game starts quietly showing you what a person who treats a demonic invasion as material actually looks like from the outside, and the companions start noticing, and the ending has a cost the jokes were papering over the entire time.

The Lich path does the inverse and does it more cruelly. It is the strongest path mechanically and the loneliest narratively — the game gives you power and then removes, one by one, the people who liked you. Several companions leave. One in particular can be kept only by doing something to them, and the game does not soften what that is.

The Aeon path’s best moment is the retroactive one: the ability to declare that a thing which happened did not, applied to a specific historical injustice the campaign has already presented as settled fact. It is the single most CRPG thing in the game — a rules interaction that is also a moral position — and it exists because Owlcat took Paizo’s mythic rules literally instead of politely.

That is the case for maximalism. A restrained version of this game would have had one ending, four classes and no lorry-sized holes in its rules, and nobody would still be arguing about it five years later.

Before you accept a mission inDeep Rock Galactic, the game tells you how long it will take. This has nothing to do with the vague estimated-campaign-length figure on a store page. On the mission select terminal there is a row of little icons, one to three, for length, and another row for complexity, and they mean what they say. A one-dot mission is a quarter of an hour. A three-dot is closer to half. You can look at the wall, look at the clock, and make an informed decision about whether you have time for this before the washing machine finishes.

I have been trying for years to explain why that small piece of interface design makes me trust Ghost Ship Games more than any studio operating a seasonal calendar, and the honest answer is that everything else about the game follows from it.

The four verbs

Ghost Ship, a Copenhagen studio, putDeep Rock Galactic into early access in February 2018 and shipped 1.0 in May 2020. The premise is that you are one of four space dwarves employed by an aggressively cheerful mining corporation to go into procedurally generated caves on an alien planet, extract something, and get out before the bugs finish eating you.

The design decision that makes it work is that each of the four classes owns a way of moving through the world, rather than owning a role in a combat triangle. The Driller carries a pair of drills that eat tunnels through solid rock, and he is the reason a vertical shaft becomes a ramp. The Engineer plants platforms out of a gun and strings ziplines, and he is the reason an unreachable mineral vein is now a floor. The Gunner fires ziplines across chasms and drops a bubble shield that pauses the world for six seconds. The Scout has a grappling hook and a flare gun, and he is the reason anyone can see anything at all.

Every cave is fully destructible. Put those two facts next to each other and the whole thing clicks: the map is a problem, and each of you holds a different tool for deforming the problem. When a mission goes well it is because four people independently reshaped the same rock into something navigable without ever discussing it. When it goes badly it is because the Driller has tunnelled somewhere private, the Scout is two hundred metres up a wall, and the Gunner and the Engineer are having a nice quiet time being eaten.

The real ancestor of this is not the co-op shooter lineage at all. It isBoulder Dash — Peter Liepa and Chris Gray’s 1984 C64 game, one of the first things I ever loaded off tape that made me think of terrain as a material rather than a backdrop.Boulder Dash understood that digging is a verb with consequences, that the tunnel you cut is a tunnel that exists afterwards, and that the tension in a mining game comes from the geometry you yourself created.Lemmings took the same idea onto the Amiga seven years later and made it about other people’s stupidity.Deep Rock Galactic is the version where the terrain you ruined is a shared social space, and the stupidity is yours.

Why the loop holds

The mission structure is where the craft is. A Mining Expedition asks for a quota of Morkite; Point Extraction wants Aquarq crystals hauled to a central pad; Salvage has you defending stationary uplinks on a timer; Escort Duty walks a drilling machine called Doretta through the cave and dares you to keep her alive. They are all, structurally, the same shape — go in, do a task under pressure, call the drop pod, run for it — and yet they do not blur, because the task changes what the cavemeans.

In a Mining Expedition the cave is a larder and you wander it greedily. In Salvage the cave is a defensive perimeter you have five minutes to understand. In Escort Duty the cave is a corridor being carved in front of you by something that does not care about your opinion. Same rocks, same bugs, same four dwarves. Completely different reading of the space. That is the trick that lets a game with a handful of mission types stay legible across hundreds of hours, and it is a far more efficient use of design effort than shipping thirty modes.

Then there is the extraction. The drop pod lands, a timer starts, and now every mineral in your pack is a bet against your ability to sprint. This is the single best-tuned moment in the game and it is essentially free: it costs Ghost Ship nothing to add a countdown, and it converts the last ninety seconds of every mission into a small farce. Somebody always dies. Somebody always has to be carried. The dwarf who mined the most is invariably the one who is furthest away when the door opens.

Nitra is the other quiet masterpiece. It is the mineral that buys resupply pods, at eighty per pod, and it is scattered like everything else. Which means your ammunition economy is amining problem. Run dry and the answer is to go find some rock. No shop, no loadout screen mid-mission, no crate. The resource that keeps you shooting is the resource you are already there to dig, and the two systems fuse instead of sitting next to each other.

The part everyone else should copy

Deep Rock Galactic runs seasons. Season 01 arrived in late 2021 and they have kept coming. Here is what a season is: free, for everyone, with a progression track you can work through by playing normally. Here is what happens when it ends: the cosmetics and gear from that season roll into the general loot pool, permanently available to anyone who shows up afterwards.

Read that again, because the industry standard is the exact inverse. Nothing expires. Nothing is held hostage behind a date you missed. There is no paid tier of the pass. The studio sells optional cosmetic packs, and that is the whole monetisation. A player who buys the game in 2026 can obtain everything a 2021 player has, by playing.

This is a coherent theory of what a co-op game is for, and charity has nothing to do with it. Ghost Ship appears to have concluded that the product is the fifteen minutes in the cave, and that anything which makes those fifteen minutes feel like an obligation is damaging the product. The Deep Dives, the weekly three-stage runs on a shared seed, are the one concession to a calendar, and even those are a treat rather than a tax: a special hard thing that expires, sitting on top of a permanent library that does not.

The social furniture matters here too, and it is easy to be sniffy about. The salutes. The “Rock and Stone!” shout mapped to a button. The Abyss Bar on the space rig where you can drink beers that apply modifiers and dance badly at a jukebox. Every one of those is a small mechanism for turning four strangers into a crew before the mission starts, and they work on a population of anonymous players in a way that no amount of voice-chat etiquette guidance ever has. It is the most consistently pleasant public lobby in the genre, and that is a design achievement; the demographics did not manage it on their own.

Where it fights itself

The gear progression is slow in a way the rest of the game is not. Overclocks — the build-defining weapon modifiers — come from Deep Dives and machine events and a forge with a randomised cost, and the randomisation means the specific overclock you want may simply decline to appear for weeks. A game this generous with its content is oddly stingy about its builds.

And Hazard 5 is where the design’s honesty runs out slightly. The lower hazards are a physics comedy; Hazard 5 is a game about knowing the spawn logic, and the gap between them is a small cliff wearing the costume of a ramp. The custom difficulty settings added later paper over this, and they are the right answer, but they arrived years after the players who bounced off had already gone.

The verdict is thatDeep Rock Galactic is the best-structured co-op game of its generation, and the structure is the argument. It is on PC, PlayStation and Xbox, has been on Game Pass, and asks for a fifteen-minute commitment and nothing else. Play it with three people you like. Play it with three strangers; that works too, which is the point.

If you want to see the same studio’s ideas run through a different mill,Deep Rock Galactic: Survivor takes this world and pours it into the auto-shooter shape I wrote about inthe Vampire Survivors review, and it is a much better licensed spin-off than it had any need to be. For the other end of co-op — the one that punishes dawdling instead of scheduling around your evening —Risk of Rain 2 is the counter-argument.

Spoilers below

A game with no plot has little to spoil, though it has a shape worth knowing about.

The season storylines — the Rival Corporation’s robots, the caretaker sequence, the various things Mission Control declines to explain — are told almost entirely through voice lines, terminal barks and the occasional new enemy that turns up without introduction. This is Ghost Ship’s most underrated decision. The narrative is ambient, skippable, and never once stops a mission to deliver itself. You learn that a rival company has been mining the same rock because you found their machinery in your cave, got shot by it, and had to deal with that instead of watching it.

Compare that to what a live-service game usually does with story, which is to interrupt your fifteen minutes with a cutscene about a war you did not enlist in. Ghost Ship’s version respects the same rule as the mission timer: your evening is yours, the cave is the game, and anything the studio wants to tell you has to fit around the digging.

And then there is Karl. Nobody will tell you who Karl was. The dwarves toast him, blame him, invoke him. The joke has run for years and the studio has never explained it, which is exactly the right call — Karl is a folk practice, and folk practice dies the moment somebody writes the lore page.

There is a version of this review that is four words long and reads “it costs four quid”. Poncle’sVampire Survivors has been on Steam since December 2021, went 1.0 in October 2022, and has spent the years since being the cheapest thing on anyone’s account that they have still somehow put a hundred hours into. It won the BAFTA for Best Game in 2023, which remains the funniest sentence in recent awards history — a browser game built in Phaser by one bloke in Italy, sat in the same category as budgets with a comma in them.

The joke has been told. What has not been examined nearly enough is the actual engineering. BecauseVampire Survivors is one of the cleanest pieces of subtractive design I have seen since the sixteen-bit years, and the thing it subtracted is the thing every other game in its lineage assumed was load-bearing.

The removal

You move. That is the input. The left stick, or WASD, and nothing else. Your weapons fire on their own timers, in their own patterns, forever, whether you are paying attention or not. Monsters come in from the edges in tides. Every kill drops an experience gem; every few gems is a level; every level throws up a small menu of weapons and passive items and asks you to pick one. Thirty minutes on the clock, and at thirty minutes Death arrives and takes the run off you regardless of what you have built.

Strip a twin-stick shooter of the aiming and you would expect to be left with nothing. Instead you are left withpositioning, and it turns out positioning was carrying the whole genre the entire time. The direct ancestor here is well documented — Galante has been open that he took the shape fromMagic Survival, a Korean mobile game from 2021 — but the deeper root is the arcade lineage that runs throughRobotron: 2084, where the second stick was really just a way of expressing which pile of enemies you had decided to be nearest to.Vampire Survivors noticed that the interesting decision in that genre was always the standing — the shooting was bookkeeping — and had the nerve to delete everything else.

What this does to the moment-to-moment is peculiar and specific. You stop reading the screen as targets and start reading it as terrain. A whip that fires horizontally means you want enemies on your flanks; a King Bible orbiting your body means you want to be inside the crowd rather than backing away from it; a Garlic aura means the correct play is to walkinto the thing that is killing you. Your build silently rewrites what “safe” means, and half the skill of a run is noticing that the geometry of safety changed two levels ago and you are still moving like the old build.

The economy underneath

The level-up menu is where the design does its real work, and it is worth being precise about why it lands, because “you get a choice of upgrades” describes a hundred games that feel like nothing.

You have six weapon slots and six passive slots. Weapons cap at level eight. A capped weapon plus the correct passive item unlocks anevolution, which arrives only from a chest dropped by a boss after the ten-minute mark. That is three separate resources — slots, levels, time — all converging on a single delivery window, and the effect is a run that has a genuine dramatic structure rather than a difficulty slope. Minutes zero to ten are you deciding what the run is about. Ten to twenty are the evolutions cashing in and the screen turning to soup. Twenty to thirty are you finding out whether the soup is thick enough to survive what is coming.

The screen-turning-to-soup is the part people describe as the game playing itself, and it is where the criticism usually stops. I would argue the opposite: the soup is thereward, and it is a reward the design has to earn by making the first ten minutes genuinely precarious. EarlyVampire Survivors, before your first evolution, is a horrible tense scrabble in which one bat can end you and the gems are always eight steps too far away. The power fantasy at minute twenty-two is only legible because you remember being nearly killed by a bat at minute two. Take away the fragile opening and the whole thing collapses into a screensaver, which is precisely what happens in the dozens of imitators that let you start strong.

Galante’s professional background before this — years designing for the gambling industry — is a matter of public record and he has talked about it openly, and you can see the fingerprints without needing to be rude about it. The gem-collection radius, the near-miss, the drip of small rewards between the big ones: this is somebody who knows exactly which frequency the reward loop wants to run at. The difference is that a slot machine’s schedule exists to extract, and this one exists to keep a decision interesting. Same tooling, aimed somewhere better.

Where it fights itself

The meta-progression is the weak joint. Gold from runs buys permanent PowerUps — more might, more speed, more luck, more revives — and it does the standard roguelite thing of making early runs artificially bad so that later runs can feel earned. It works, in the sense that it kept me coming back. It also means that a substantial chunk of your first several hours is spent losing to a wall that exists because a spreadsheet says you have not paid yet.Hades, which I have written abouthere, got away with this by attaching a story beat to every failed run;Vampire Survivors attaches a number.

The Arcana system, added at 1.0, is the more interesting late addition and the more uneven one. Arcana cards are run-wide rule changes — one makes your first weapon fire on a timer independent of cooldown, another converts recovery into damage — and the good ones are genuinely build-defining in the wayBalatro’s jokers are, a comparison I have leaned on before inthat review. The problem is that a few of them are so much better than the rest that the “choice” of two Arcana at the start of a run is frequently no choice at all. When your deck contains a card that trivialises the ceiling, the ceiling stops being a place you visit.

And the content has sprawled. Poncle has kept adding — Legacy of the Moonspell in late 2022, Tides of the Foscari in 2023, anAmong Us crossover, aContra one, and eventually the official Konami-licensedOde to Castlevania in 2024, which retroactively legitimised an aesthetic the game had been cheerfully gesturing at since the itch.io days. Each pack is generous and cheap. Collectively they have turned a game whose original virtue was that you could understand all of it into one with a collection screen you scroll. The free Adventures mode was an honest attempt to re-impose a shape on that pile, and it half works.

What it is actually for

Here is the read.Vampire Survivors is a game about the pleasure of a system you built becoming legible to you all at once. The reason it survives its own tedium — and there is tedium; minutes twenty-two to thirty of a strong run are frequently just holding a direction — is that the legibility arrives as asensation rather than as information. You do not read a stat sheet and conclude the build worked. You watch the screen fill with your own consequences.

That is a real, specific thing that games can do and other media cannot, and it is why the hundred clones that copied the auto-attack and the XP gems mostly feel hollow. They copied the loop. The loop was never the point. The point was the ten-minute window of genuine fear that makes the twentieth minute mean something, and fear is expensive to design and free to leave out.

Play it on whatever you have — it is on PC, Xbox and Game Pass, Switch, PlayStation, and free on mobile with ads, and it runs on hardware that would struggle to open a browser tab. Buy the base game, ignore the DLC until the base game bores you, and give it the first hour on the understanding that the first hour is meant to be a bit miserable.

If it hooks you, go toRisk of Rain 2 next, which does something structurally related with a clock that gets angrier the longer you dawdle, and which asks for rather more of your hands.

Spoilers below

The unlock tree is the real second game, and it is where the design shows most personality.

Moongolow is the trick everyone remembers: a stage that runs a fixed fourteen minutes and then ends in a cataclysm, dropping you somewhere considerably less pleasant, and it is the only moment inVampire Survivors that has anything you could call a set piece. It works because the game has spent hours teaching you that stages are inert arenas, and then one of them turns out to have a plot.

Green Acres and The Bone Zone are jokes — endless, arbitrary, essentially test rooms left in with the lights on — and their presence tells you something honest about the project’s origins. This was a browser toy that never fully stopped being one, and Poncle has declined to sand off the parts that give it away. The secret characters unlocked by typing nonsense into the main menu are the same instinct: a game that remembers cheat codes were once a folk practice rather than a store page.

The Randomazzo, the Yellow Sign, the escalating chain of relics that each unlock the ability to find the next thing — this is the structure that keeps people at four hundred hours, and it is essentially an ARG bolted to a screensaver. Whether that is genius or a cheerful mess depends entirely on how you feel about a game that hides its best ideas behind its worst ones. I lean towards genius, on the grounds that nothing else that came out in 2022 was this confident about what it could afford to leave out.

I loaded games off cassette on a C64 for most of my childhood, which means I spent a meaningful fraction of the 1980s listening to a tape deck screech at me for four minutes to find out whether the thing on the inlay card was any good. It usually was not. That is the part everybody forgets when they say the old games were better. The old games were a lottery, played at £1.99 a ticket, and the thrill was structurally inseparable from the odds.

UFO 50, released on PC on 18 September 2024 by Derek Yu’s Mossmouth, is a box of fifty tickets. It is presented as the complete catalogue of UFO Soft, a fictional developer who made games for a fictional 1980s machine called the LX System between 1982 and 1989. None of the games are real. All of the games are real. The distinction stops mattering about twenty minutes in, which is the trick.

The fiction is the design document

The team — Yu with Eirik Suhrke, Jon Perry, Paul Hubans, Ojiro Fumoto of Downwell fame, Tyriq Plummer and others — built the constraint before they built the games. One palette. One notional machine. One studio’s imagined career arc. Then they made fifty complete games inside it and dated them.

That constraint is doing three separate jobs, and only one of them is nostalgia.

The first job is coherence. Fifty unrelated minigames would be a Wario Ware. Fifty games from one imaginary studio have ahouse style — recurring mascots, sound palettes you recognise across a decade, ideas a designer clearly tried in 1983 and got right in 1987. Barbuta, dated first, is a deliberately obtuse thing that withholds almost everything and expects you to map it on paper. It is also the worst-reviewed game in the box by consensus, and it is placed first on purpose, because a studio’s first game is supposed to be the one where they had not worked it out yet.

The second job is permission. The fiction lets these designers ship a game that isrude. Real 1980s games did not explain themselves, did not respect your time, and frequently did not tell you the rules. Modern design has spent thirty years correcting that, mostly correctly. But a modern game that withholds is read as broken, whereas a 1984 game that withholds is read as a 1984 game. The frame buys the team the right to be genuinely unhelpful, and several of the best things in the box only work because of it.

The third job is the argument, which I will come to.

The good ones are properly good

The line that gets thrown around is that UFO 50 is “twelve great games and thirty-eight demos”. That is lazy. The distribution is real — there are entries here I bounced off inside five minutes and would not defend — but the hit rate is far better than any compilation cassette I ever owned, and the ceiling is higher than the pitch implies.

Mortol is the standout structural idea: a platformer where your stock of lives is the level’s building material, because each corpse becomes a step, a bridge, a switch held down. Dying is the verb. It is a genuinely publishable idea that would carry a full-price indie release on its own, and it is sitting in a box with forty-nine others.

Party House is a deckbuilder about hosting parties where the guests you want are the guests who might ruin it — a push-your-luck engine with a social skin that has no business being this tight. Grimstone is a full Western tactics RPG, hours long, with a party and a job system. Campanella is a physics flying game about momentum and patience. Velgress is a vertical climber built on the anxiety of rising death. Vainger is a Metroid-shaped thing with modular power slots. Golfaria is golf that grew a metroidvania. Night Manor is a point-and-click horror game with real dread in it.

Any one of those, polished up and released alone with a trailer, would have picked up coverage. That is the density we are talking about.

What holds them together is that the team understood which 1980s conventions wereload-bearing and which were merely damage. The obtuseness is kept, because obtuseness is what made a 1984 game a place you inhabited for a month rather than a thing you consumed in an evening. The genuinely broken parts are quietly fixed: the collision is honest, the inputs read on the frame you pressed them, the difficulty is hard in ways you can learn from. Anyone who has actually gone back to a beloved C64 title in the last decade knows how much of the misery was technical rather than intentional. UFO 50 keeps the intent and throws out the misery, and that editorial judgement — exercised fifty separate times — is the real labour in the box.

The cherry is the real design

The meta-layer is the part I keep thinking about. Finish a game and you get a Gift. Meet a harder, game-specific condition — a score, a challenge, a deeper completion — and you get a Cherry. The Gift says you saw it. The Cherry says you understood it.

This is a superb piece of engineering because it solves the compilation’s oldest problem. Every collection I owned as a kid had the same failure state: you play each thing for ninety seconds, decide, and never return. The Gift/Cherry split gives you two distinct reasons to stay, calibrated to two distinct kinds of player, and it puts the decisioninside each game rather than in a menu. You are never being asked to like all fifty. You are being asked to find out which three are yours.

The design ancestor is the high-score table on a machine in a chip shop, where the point was that someone else had already proved the number was reachable. UFO 50 rebuilds that pressure without a leaderboard, purely through the implied competence of a fictional studio.

The argument

Here is what the fifty fake games are actually arguing, and it is a better argument than the packaging suggests.

The claim is that the 1980s constraint producedvariety as a by-product of poverty, and that the variety was the good part. Nobody knew what a game was yet. A team of three had a machine with sixty-four kilobytes and no genre conventions to obey, so what came out was strange — golf with a map, a platformer made of corpses, a party sim, a mech game, a fishing thing — because nobody had yet worked out which of those were supposed to be commercially viable. The market answered that question in the 1990s and the answer narrowed everything.

UFO 50 stages a counterfactual: what if that decade had been run by people who already knew how to design? Same hardware ceiling, same palette, same absence of tutorials — with thirty years of accumulated design literacy behind the keyboard. The result is fifty games that feel period-accurate and are quietly smarter than anything the period actually produced, and the gap between those two facts is the thesis.

The honest ancestor of this whole object is the covertape and the budget label — Mastertronic, Codemasters, the magazine cassette blu-tacked to the front of Zzap! that I fed into a Datassette before I had read the review. Those tapes were the delivery mechanism for exactly this experience: a dozen unlabelled things, most of them bad, one of them yours forever. UFO 50 is that tape with the failure rate tuned down and the ambition tuned up.

Where it fights itself

The fifty-game shape has a fifty-game cost. There is no way to sample this efficiently, and the entries are not sorted by quality — deliberately, since the fiction requires a career arc rather than a greatest-hits. You will spend hours on things you do not like to reach things you love, and the game is perfectly comfortable with that. If your gaming time comes in ninety-minute slots after the kids are down, that friction is a real charge against it.

The other cost is that the deepest games here — Grimstone especially — are asking for the commitment of a standalone release while sitting behind a menu that implies a snack. Several of the best things in the box are structurally disadvantaged by their own container.

The verdict

UFO 50 is the most generous thing released in 2024 and one of the few games I would describe as an act of scholarship. It understands the 1980s as adesign condition rather than an aesthetic, which is why it earns the pixels in a way that a thousand pixel-art indies with a CRT filter never do. Buy it on PC. Understand that you are buying an argument with fifty pieces of evidence attached, that you will hate some of the evidence, and that at least two of the fifty will end up in your permanent rotation.

If the appeal here is the density of ideas per hour, the other 2024 indie worth your time isAnimal Well, andBalatro is the piece to read on what happens when one of these small, strange systems escapes the box and eats a year of everybody’s life.

Spoilers below

The catalogue has an internal continuity, and finding it is a genuine pleasure that the marketing sensibly left alone. Entries reference one another across the imagined decade. Mortol gets a Mortol II further down the timeline that takes the corpse-as-scaffolding idea and complicates it rather than merely enlarging it, which is exactly the move a real studio makes with a surprise hit. Mascots and sprites recur. Sound motifs carry between games years apart in the fiction.

The effect of that is stranger than a straightforward Easter-egg hunt. By the time you have played twenty of these, you have opinions about UFO Soft as acompany — which of their designers you rate, when they lost the plot, which 1986 experiment obviously came from the same person who made the 1983 oddity you disliked. You are doing criticism on a body of work that does not exist. That is an absurd thing for fifty games in a Steam release to achieve, and it is the clearest evidence that the fiction was the point rather than the wrapper.

There is a moment, somewhere around twelve hours intoNier: Automata, where the credits roll, the game thanks you, and then it hands you a fresh save slot and expects you to start again. Most games treat that screen as the exit. This one treats it as a chapter break. PlatinumGames and Square Enix shipped it in Japan on 23 February 2017 and in the West a fortnight later, and the thing that made it a word-of-mouth monster was the same thing that should have killed it: it asks for your time twice, then a third time, then once more.

Asking a stranger to play your game four times is an enormous imposition. I take that seriously. A 40-hour commitment is 40 hours of somebody’s actual life, and the medium is full of designers who spend it like it’s free. Director Yoko Taro has been running this trick sinceDrakengard in 2003 and the firstNier in 2010, and both times the structure was the part people forgave rather than the part they praised.Automata is where it finally works, and the reason it works is mechanical rather than literary.

The chip system is the argument

Start with the thing nobody puts on the box. 2B’s abilities live on plug-in chips slotted into a storage grid with a fixed capacity. You spend that capacity on the things you’d expect: auto-heal, ranged attack buffs, drop-rate boosts, melee damage. You also spend it on your HUD. The health bar is a chip. The minimap is a chip. Damage numbers, the XP display, the enemy targeting overlay — all chips, all occupying slots, all removable.

Pull them out and you get more room for combat upgrades. You also lose the ability to see how close you are to dying. The game turns interface into an economy, and the exchange rate is brutal and honest: legibility costs power. I have never seen the trade stated so plainly. Every game has a HUD budget — art directors argue about it, players mod it away — andAutomata is the only one I can think of that put the argument in your hands and made you pay for the answer.

It goes one step further. One of those chips runs the operating system. Remove it and 2B dies on the spot, and the game files the death as one of its endings rather than a bug. That is a joke with a design thesis inside it. The interface is diegetic; the android is running software; the software includes the bit that draws your health bar. Once you understand that, the ending structure stops looking like an art-house imposition and starts looking like the same idea at a larger scale.

Why the second pass earns its keep

The pitch is that you play Route A as 2B, then Route B covers the same events as 9S. If that were a straight replay with a new hat, the game would deserve every complaint it gets. It isn’t a replay. 9S is a scanner model, and his combat kit is built around hacking, which drops you into a twin-stick shooter played inside the enemy’s head. Same fights, different verbs. The bullet-hell overlay that runs on top of the third-person action — machines firing lattices of white spheres while you’re mid-combo — is Platinum showing off, and it’s also the connective tissue that lets the game slide between genres without a loading screen.

More to the point, 9S can read the machines. Route B gives you access to information 2B didn’t have, so the same scene plays with a different amount of knowledge in your head. That is the whole engine. The repetition isn’t padding because yourcomprehension is the variable being upgraded, and comprehension is the only stat in this game that can’t be farmed.

Route C is new content outright, and by the time you reach it the game has stopped explaining its structure and started using it. The endings past the fifth are largely jokes — walk away from a mission, eat the wrong fish, quit at the wrong prompt — and they exist to teach you that the ending list is a systems menu rather than a narrative achievement board.

Where it fights itself

I’m not going to pretend the seams aren’t visible. The open world between set-pieces is thin, the city ruins and the desert and the amusement park are connected by a lot of running, and the sidequests range from genuinely wounding to fetch-quest filler that Platinum clearly built to a schedule. The PC port shipped in March 2017 in a state that took a fan patch — FAR, by Kaldaien — to make presentable, and Square Enix only patched the resolution and window handling years later, in July 2021. That’s four years of the definitive version of an acclaimed game being a community project. Worth remembering when a publisher tells you the platform matters.

The combat, too, is Platinum on cruise control. It’s fluid, it’s readable, it has the dodge- cancel rhythm you’d expect from theBayonetta lineage, and it has nothing like the depth ofBayonetta. Difficulty on Normal will not test you; the auto-chips will literally play the fights for you if you let them, which is a design statement in itself and also an admission that the fighting is a delivery mechanism. Play it on Hard, where one hit is catastrophic and the HUD chips suddenly feel like life support you can’t afford to unplug.

The ancestor

Everyone reaches forChrono Trigger and New Game+ here, and the shape is right — multiple endings, a replay that recontextualises — but the real ancestor is closer to the adventure games that built a full run around a single missing fact. The structure thatAutomata actually inherits is the one where the game withholds a perspective rather than a key, and you can’t buy your way past it.

For the modern version of the same idea, look atInscryption, which does its frame-breaking in a single sitting and pays for the compression with a weaker back half. OrHades, which solved narrative repetition by making the story respond to the loop instead of sitting behind it — the cleanest answer anyone has given to this problem, and a useful contrast, because Supergiant made repetition voluntary and Yoko Taro made it compulsory. And if you want the piece that tookAutomata’s melancholy and its multi-ending structure somewhere tighter and considerably nastier, that’sSignalis.

The verdict, argued

Nier: Automata is a game with mediocre traversal, decent combat, an ugly launch on PC and a script that lurches between undergraduate philosophy and genuine grief, sometimes in the same conversation. It is also the most structurally intelligent big-budget game of its decade, because it found a way to make the player’s time investment into the medium of the work rather than the price of it. The chip grid is the thesis in miniature: everything you see costs something, and the game will let you sell your own eyes for damage.

The four-route ask is real and it’s the honest thing to warn people about. Route A alone is a competent Platinum action game with a strange tone. Stopping there is the equivalent of reading the first act and filing a review. If you don’t have the hours, that’s a legitimate reason to skip it entirely; the game does not have a short version and pretending otherwise does nobody any favours. If you do have the hours, it repays them at a rate almost nothing else manages.

It’s on PS4, Xbox One, PC and Switch, and the Switch version —The End of YoRHa Edition, October 2022 — is the one that finally treats the structure as a feature to be supported rather than an obstacle. Play it on Hard. Strip the HUD when you’re feeling brave. Put it back when you aren’t.

Play next:Hades for the friendly version of the same problem, orSignalis for the version with worse dreams.

Spoilers below

The reason the structure earns its reputation is Ending E, and Ending E only works because of everything above.

After the fourth route resolves — after A2 and 9S have had their argument with a sword — the credits roll for the last time, and the game turns them into a shoot-’em-up. You fly a small ship at the names of the people who made it, and the names shoot back. It is unwinnable. The difficulty scales past any reasonable input, you die, you retry, you die, and the game asks whether you’d like help.

Say yes and rescue messages appear: real sentences written by real players who finished before you, floating in the bullet field as encouragement. They join your formation. They take hits for you. Dozens of them, wrapped around your ship like a shell, and every one of them is somebody who beat this sequence and then agreed to a bargain the game explains only at the end — to leave a message and offer help, you delete your save file. All of it. The chips, the routes, the hours.

The prompt is unambiguous about what it’s taking. And the game has spent four routes teaching you that data is what an androidis — that 2B and 9S are backed up, restored, replaced, and that the horror of YoRHa is precisely the persistence of the file. Then it asks you to give yours away so a stranger you’ll never meet can get through a credits sequence.

I know exactly what it is. It’s a magic trick with a permanence lever, engineered for maximum effect, and the emotional physics are shameless. It also works, and it works because it’s mechanical. Nobody tells you sacrifice is meaningful. The game charges you for it, in the one currency it has spent forty hours teaching you to value, and takes the payment without ceremony.

That’s the whole desk’s argument in one screen: the mechanic makes you feel it, and the mechanic is the only thing that could have.

It’s April, and the year is over. Balatro came out on 20 February, a solo project from an anonymous Canadian developer trading as LocalThunk, published by Playstack, and it passed a million copies inside a few weeks with no marketing apparatus worth the name. Whatever else 2024 does, it will be doing it in Balatro’s shadow, on people’s phones, on their lunch breaks, in the tab behind the spreadsheet.

The reason is not poker. LocalThunk has been fairly open about barely playing poker; the hand rankings are a UI he borrowed because everybody already knows them. What’s underneath is something else entirely, and it’s the oldest trick in this business done better than anyone has done it in years.

What it is

You play a run. Each run is a ladder of antes, and each ante has three blinds — Small, Big, and a Boss Blind with a rule attached that breaks something you were relying on. To beat a blind you need to reach a score. You have a standard 52-card deck, a hand of eight cards, a small number of plays and a small number of discards. You select up to five cards, the game identifies the poker hand, and it scores.

The scoring is the game. Every hand produces two numbers: chips and mult. Your score is chips × mult. A pair is 10 chips and 2 mult, so 20 points, and the Small Blind of Ante 1 wants 300. So you play a few hands, you scrape past, and you go to the shop.

The shop sells Jokers. There are a hundred and fifty of them and each is a rule. One adds chips per club. One adds mult for every discarded card. One multiplies your mult. One doubles the effect of the Joker to its left. You have five slots.

Also on sale: Tarot cards that transform individual playing cards, Planet cards that permanently level up a hand type so that every future Full House scores more, Spectral cards that do something drastic with a cost attached, and Vouchers that change the run’s rules outright. Fifteen starting decks, each rewriting the opening position. Eight escalating stakes that layer restrictions on top.

You need 300 at Ante 1. You need hundreds of thousands by Ante 8. Do the maths on what has to happen in between.

The whole design is one multiplication sign

Here’s the argument. Balatro’s central insight is that chips and mult are on opposite sides of an operator, and almost every decision in the game is you choosing which side to feed.

Chips are the safe side. They add. A Joker that gives +30 chips is +30 chips forever, dependable, boring, and it will not carry you past Ante 6.

Mult is the dangerous side. Additive mult is a solid living. Multiplicative mult — the ×3s, the ×1.5-per-condition, the ones that scale off something you have to maintain — is where the run either explodes or dies, because a ×3 is worth nothing without chips to multiply and everything with them.

So the run has a shape, and it’s the same shape every time and it never gets old: the first three antes you’re building a chip base and it feels like admin; the middle antes you’re hunting for the multiplier that will make the base mean something; and then either you find it and Ante 7 evaporates in one hand for four hundred thousand points, or you don’t and you die at Ante 6 doing perfectly respectable arithmetic.

That escalation from “300” to “300,000” across forty minutes is the drug.Vampire Survivors runs the identical curve — trivial start, absurd finish, the player’s own build outgrowing their comprehension — and does it with no decisions in it at all. Balatro makes yousign every step of the escalation. You chose the Joker. You sold the other one. You are personally responsible for the number, and the number is ridiculous, and that combination is why people cannot put it down.

The boss blinds are the balance patch

The obvious failure mode for a design like this is the degenerate strategy: find the one engine, run it every game, watch the game die. Balatro’s answer is elegant and it’s built into the ladder.

Every ante ends in a Boss Blind, and boss blinds attack your assumptions rather than your score. One debuffs an entire suit. One only lets you play one hand type. One blocks your discards. One turns your cards face down. One demands you play five cards every hand.

What that does structurally is force every build to have a second gear. A run built entirely on flushes meets the boss that debuffs a suit and has to have an answer ready three antes before it knew the question. So the shop stops being a place where you buy the best thing and becomes a place where you buy insurance you hope to waste, andthat tension — optimise now versus survive later — is a resource-allocation problem dressed up as a card game.

It’s the same instinct behindInscryption’s willingness to break its own rules, though Inscryption breaks the frame for a narrative payoff and Balatro breaks it purely to keep you honest. Slay the Spire did the ancestral version of this with its elite and boss relics; Balatro tightened the loop until an entire Spire run fits in the time it takes to boil a kettle.

Why it feels like that

Talk to anyone who’s played it and within a minute they’ll do the noise. The score punch — the way the chips tally with a rising pitch, each card flipping and firing, the Jokers going off in sequence left to right, the number climbing in audible steps and then the whole thing landing with a thump.

That’s not decoration. It’s the payoff structure made physical. The hand is already resolved the moment you press play; the game could show you the total instantly. Instead itperforms the multiplication, card by card, so you get four seconds of accelerating evidence that the thing you built works. It’s a fruit machine’s reel-stop rhythm applied to a decision you actually made, which is the respectable version of the same neurology.

The presentation carries it. Everything is chunky, CRT-warped, curled at the edges, and the whole thing looks like a card game running on a machine that should not be running a card game — a deliberate and very well-judged bit of texture from someone who clearly understands that a slightly wrong phosphor glow makes numbers feel heavier.

Where it fights itself

The luck floor is real. There are runs where the shop offers you nothing, the Jokers don’t talk to each other, and you die at Ante 5 having played correctly throughout. That’s the genre’s tax and Balatro pays it more than most, because the multiplicative engines it’s built around are binary — you have one or you don’t. Slay the Spire could grind out a win on fundamentals. Balatro often can’t.

The higher stakes expose the seams too. Gold Stake asks for a level of consistency that the deck’s variance doesn’t really support, and the result is a fair amount of run-abandoning at Ante 2 when the opening doesn’t cooperate. That’s not a difficulty setting so much as a lottery with a longer queue.

And the game had a genuinely stupid month. In March it was pulled from several storefronts after a ratings body decided that pictures of playing cards constituted gambling content and slapped an adult rating on a game with no money, no wagering and no chance to lose anything but forty minutes. Balatro simulates a slot machine’sfeel precisely, which is worth being honest about — the reel-stop dopamine is engineered and it works on people. It also has no gambling in it. The rating was wrong about the object and accidentally right about the mechanism.

The verdict

Balatro is the tightest escalation engine anyone has shipped this decade, built by one person, running on a laptop, in a genre everyone assumed was solved. Its real achievement is structural: it found a way to make a number going up feel earned, by putting a multiplication sign in the middle of it and making you responsible for both operands.

Buy it on whatever’s nearest. It runs on anything, the console versions are identical to the PC one, and the thing you’re buying is a forty-minute loop you’ll run four hundred times.

If the escalation is what got you,Vampire Survivors is the same curve with the decisions removed, andHades is the same loop with a reason to press start.

Spoilers below

The endless mode past Ante 8 is where the design’s real character shows, and it’s the part I’d argue about.

Beat Ante 8 and you can keep going, and the required score stops climbing steadily and starts going somewhere the number formatting can’t follow. The game begins displaying scores in scientific notation, then in a naming scheme that gives up on dignity entirely, and the blinds ask for figures that no honest engine can reach. The only way through is to break the game — to find the specific Joker interaction that produces a genuinely unbounded loop and let it eat itself.

That’s a confession. Endless mode admits that the multiplication engine, given no ceiling, has no interesting equilibrium; past a point the game is only playable by exploiting it, and LocalThunk simply left the door open and let people find out.

I think that’s the correct call, and it’s also why Ante 8 is where the game actually lives. The eight-ante ladder is the designed object: a curve tuned so that a good run peaks exactly as it ends, which is the hardest thing in this genre to get right and the reason Risk of Rain’s loop and Spire’s Act 4 both wobble. Balatro gets to stop while it’s still beautiful. Everything past that is the developer showing you the machinery, with the panel off, on purpose.

In 1991 I could not play Monkey Island without the cardboard wheel. You’d line up the pirate’s hair with the pirate’s chin, read off a date, type it in, and the game would let you past. Lose the wheel and you owned a coaster. Every Amiga in the country had a shoebox of this stuff: code wheels, Lenslok, and the lookup tables that made you find page 14, line 3, word 6 in a manual you’d otherwise never open.

Publishers did that to stop people copying disks. What it actually did, by accident, was make the manual part of the machine. The game was on the disk and the game was also on the kitchen table, and you couldn’t run one without the other.

Tunic is what happens when someone takes that accident seriously.

What it is

Andrew Shouldice released Tunic in March 2022 after roughly seven years of work, published by Finji. It came to Xbox and PC first and reached PlayStation and Switch in September of the same year. It has spent the time since drifting through subscription services and sales, which is how most people meet it now — a small isometric action-adventure with a fox in a green tunic, obviously wearing the first Zelda’s clothes.

That’s the trap. It’s dressed as an homage and it’s actually an argument.

The manual

Scattered through the world are pages of an instruction booklet. Not lore fragments — an actual manual, laid out like something that fell out of a 1987 box: glossy illustrated spreads, a map, diagrams, a bestiary, arrows pointing at buttons. You collect the pages and you can open the booklet at any time, and the booklet is where the game keeps everything it hasn’t told you.

The manual is almost entirely written in a language you cannot read.

There’s an invented script — Trunic — running through every page, and the first time you open it your eye slides off it and lands on the pictures. Which is the point. You start reading the manual the way a nine-year-old reads a manual for a game they don’t own yet: guessing from diagrams, inferring from arrows, building a theory of what the game must contain out of pure iconography.

And then a page teaches you something real. There’s a spread that shows you how to dodge-roll, and until you find it, you do not dodge-roll — the button worked all along, and you simply did not know the verb existed. Later there’s a page that shows you a mechanic so fundamental that finding it retroactively re-explains everything you’ve been walking past for six hours.

This is progression made out ofknowledge rather than items. The character never gets a new ability. You do.

Why it works

Metroidvanias gate you with objects, and objects are honest but inert: the game withholds the double jump, gives you the double jump, and the space of what you can do expands by exactly one predictable increment. You knew the double jump was coming. You’ve played this before.

Knowledge gating has a different shape. When Tunic hands you a page, your entire back catalogue of memories re-sorts at once. Every strange wall, every suspicious statue, every geometric thing you clocked as decoration — the page doesn’t open one door. It opens all the doors of that type, everywhere, retroactively, and the game didn’t have to build a single new room to do it.

That is enormously efficient design, and it’s also the reason the game can’t be patched into being easier. You cannot hint your way around it, because the thing being withheld isn’t in the save file.

The related trick is that Trunic is not decoration. It’s a real cipher — a consistent mapping to English phonemes, learnable, and people did learn it, sitting down with the pages and cracking the script like a philology homework. The manual is fully readable if you do the work. Shouldice built an entire functioning writing system and then made almost nobody need it, which is the single most confident act of restraint in the medium.

Chants of Sennaar went at the same problem from the front, making the decipherment the loop and giving you a notebook to be wrong in. Tunic buries it and lets you walk past. Both work. Sennaar is the better teacher; Tunic is the better ambush.

The manual as an artefact

Look at the pages themselves and you find the second layer of the joke. They’re faithful to a specific era of print — the slightly off registration, the airbrushed box-art idiom, the bilingual clutter, the way European manuals crammed six languages into a booklet none of us read. There are hand-scrawled annotations in the margins in biro, because of course there are; every used manual in every second-hand game I ever bought had somebody’s map of level three in the back.

Those annotations do heavy lifting. They’re the previous owner. Somebody was here, they figured some of this out, and they left you circled hints in a hand that isn’t the manual’s. It gives the game a social texture without a single line of multiplayer code — the same trick Souls messages pull, achieved with a pen.

None of this is nostalgia bait, and I’m allergic to nostalgia bait. The point isn’t that manuals were nice. The point is that manuals were asecond information channel the game couldn’t see, and when that channel died — when everything moved in-game, into tutorials and tooltips and quest markers — designers lost the ability to withhold. If the game must teach you everything it can do, then everything it can do is a checklist. Tunic reopened the channel and immediately used it to lie to you.

Where it fights itself

The combat is the weak link and always was. It’s a slow, stamina-gated, roll-and-poke system with a shield, and it wants to be taken seriously enough to be tuned but loose enough to be forgiving, and it settles in an unhappy middle. The bosses have real teeth and the moment-to-moment fighting doesn’t have the precision to make that teeth-baring feel fair. There’s an option to switch off damage entirely, and I don’t think it’s a defeat to use it; the game’s actual content is above the neck.

The isometric camera is the other tax. Tunic hides things behind geometry deliberately — an entire class of secret depends on a path being invisible from your fixed angle — and this is genuinely clever the first six times. After that it’s a game where you occasionally walk into walls hoping. Depth ambiguity as a puzzle mechanic has a low ceiling, and the game finds it.

And the mid-game asks you to do a lot of running. The world folds beautifully and the shortcuts open, and there’s still a stretch around the halfway mark where you’re crossing three biomes because a page told you something and the thing it told you about is a long way away.

The verdict

Tunic is a game about the pleasure of not being told, and it holds a position almost nothing else in the medium is willing to hold: that the player is capable of figuring it out and will enjoy the figuring more than the finding. It spends its first hours letting you believe it’s a small polite Zelda tribute so that the betrayal has somewhere to stand.

Play it anywhere — it’s on everything now, and it’s light enough to run on a toaster. Play it with a notebook and a pen, actually, and be ready to be embarrassed by how long it takes you to notice you’re allowed to draw things.

The pairing I’d suggest isReturn of the Obra Dinn, which runs the other great knowledge-progression system of the era, andInscryption if what you liked was being lied to about what kind of game you’d bought.

Spoilers below

The Holy Cross is the whole thesis, and it’s the best thing I’ve seen a game do with an input.

There is no new item. There’s a sequence of directional presses, and the manual has been showing you where to use it the entire time — hidden in the page borders, in the decorative frames, in the fold. The instruction was in your hands from the first page you picked up. You looked at it a hundred times. It was printed on the paper.

And so the game’s real final ability isliteracy. Once you see the border code, you go back through every page you’ve collected and read the game’s own documentation as a walkthrough, which is precisely the ritual the code wheel and the lookup table trained a generation to perform — go to the manual, find the page, read off the answer, come back. Shouldice took the most hated piece of 1980s anti-piracy friction and rebuilt it as the reward.

The two endings sharpen it. You can beat the Heir with a sword, which is the answer the game’s combat has been training, and it’s the lesser ending. Or you can gather the pages, understand what the fox has been doing to the previous heirs, and end it another way entirely — an ending available only to a player who read the paperwork. One route is reflexes. The other is attention. The game knows exactly which one it thinks is worth more, and it never once says so out loud.

My car developed a habit. Every time I opened the boot, the headlights came on. Not a problem, exactly — a tic, the kind of thing you’d mention to a mechanic and he’d shrug at. Then, deep in a zone with a storm closing, I opened the boot to stow salvage in the dark and lit myself up like a fairground for everything in a hundred metres.

That’s Pacific Drive. That’s the entire pitch, and I’ve never played anything else that does it.

What it is

Ironwood Studios released Pacific Drive on 22 February 2024 for PS5 and PC. It’s a first-person survival driving game set in the Olympic Exclusion Zone: a cordoned-off stretch of the Pacific Northwest where the science went wrong in the mid-1950s and the resulting weather has spent decades not obeying anything.

You are a driver who ends up inside the wall with a battered station wagon and a garage. The loop is a run. You pick a destination, you drive out, you scavenge resources from abandoned buildings and dead vehicles, you find and gather the anchors that power the gateway home, the storm notices you’ve done it and comes for you, and you drive very fast at a portal while everything you own falls off the car. Then you’re back in the garage, and you spend what you took to repair what you broke.

Three voices come to you over the radio — Oppy, Tobias and Francis — and they have the good manners to be characters while doing a quest log’s job. The writing is dry and specific and knows exactly how much to withhold.

The quirks system is the reason to care

Survival games have a companion problem. If you want the player to be attached to something, the standard tools are a dog, a child, or a talking device with a voice actor, and all three work by telling you to be attached. Ironwood gave you a car and made the attachment emerge from a bug tracker.

Here’s the mechanic. As you drive through the zone and take damage, your car develops quirks: persistent, irrational faults that link an input to an unrelated output. Opening a door pops the boot. Turning left switches on the radio. Braking kills the electrics. The game tells you a quirk exists; it does not tell you what it is.

To fix it, you diagnose it. There’s a board in the garage where you record what you did and what happened, and you narrow the cause down by hypothesis and test — open the door, watch what moves, log it, do it again. When you’re confident, you name the fault and the car forgets it.

Look at what that does. In every other survival game, damage is a number you top up with a resource. Here damage is abehaviour, and diagnosing it is a genuine act of attention paid to a specific object. You cannot fix your car without learning your car. And because quirks are generated per-vehicle and per-playthrough, your wagon’s list of tics is unique to you and unshareable. It is not the car in the trailer. It’s the one you broke.

That is a character. Built entirely out of unreliable state transitions, with no dialogue, no face and no arc. It’s the smartest thing anyone did with a survival loop last year.

The design detail that makes it sing is that quirks are rarely fatal. A headlight that fires when you open the boot is an inconvenience nine times out of ten, so you don’t fix it immediately; you file it, you work around it, you build a private mental model of your vehicle’s nonsense and you drive accordingly. The game is counting on that laziness. It wants you carrying a list of small forgivable faults into a place where one of them will eventually matter, and the moment it does, the consequence is legible all the way back to a decision you made hours ago in a warm garage. Neglect with a delay fuse. Very few games trust the player to be the author of their own ambush that patiently.

The damage model does the emotional work

The other half is granular. The car is panels, doors, tyres, windows, battery, engine, and each is tracked separately with its own condition. A door doesn’t have hit points that lower a global health bar; a door gets bent, then it doesn’t close, then it comes off, and now you’re driving through acid rain with a hole in your side.

The result is a language of decline you can read at a glance. You come back into the garage and the state of the vehicleis the story of the run — one wing crumpled where you clipped a pylon, rear window gone from the hail, a tyre you limped home on. Nobody narrates it. You look at it.

This is where the game earns its comparison toDredge, which runs a structurally identical bargain: go out, get greedy, the dark is coming, and every extra minute is a bet. Dredge tightens the screw with a grid-inventory Tetris and a sanity meter. Pacific Drive tightens it with the weather and the thing you know your rear axle can’t take. Both are pure risk-return engines dressed as a job.

The real ancestor is Jalopy (2018) — the Trabant road-trip game where the entire drama was a car that could not be trusted and a boot full of spare parts. Jalopy had the mechanical intimacy and no jeopardy. Ironwood added the storm.

Where it fights itself

The mid-game sags, and it sags for a reason worth naming. The tension curve depends on scarcity — on wanting a resource badly enough to take a stupid risk — and about fifteen hours in, the crafting economy tips. Once your wagon has decent panels and you’ve unlocked the better fabricator tiers, runs stop being desperate and start being errands. You go out with a shopping list. The storm becomes a schedule rather than a threat.

The game’s counter is to send you deeper, where the anomalies are nastier and the resources rarer, and it half works. But the fundamental problem is that survival crafting trees resolve, and a design whose whole engine is precarity has to keep the player poor.Citizen Sleeper understood this cleanly enough to make it the theme — the dice degrade, the clock runs, and stability is always a lie. Pacific Drive lets you actually get comfortable, and comfort is the death of it.

The anomalies are the other soft spot. They’re a wonderful bestiary in concept — the abductor, the bunnies, the things that follow — and in practice most of them resolve to “an object that damages the car if you’re near it” without a distinct answer. They behave like weather. The game would be better with fewer of them and a real verb for each.

And the driving, which had to carry everything, is merely good. The wagon has weight and the roads have surface, and it never quite reaches the tactile authority that would let the traversal itself be the reward on a bad-loot run. Compare what the handling model does for a game like Euro Truck Simulator, where the act of steering is sufficient payment for the hour: Pacific Drive needs the zone to be interesting because the road, on its own, isn’t. Every stretch of empty tarmac between anomalies is time the design has to fill with something, and the somethings run out.

The verdict

Pacific Drive is a great idea executed with real conviction and a loop that can’t quite sustain itself to the end. The quirks system alone justifies it — it is a genuine invention, the first mechanic I’ve seen that manufactures affection out of debugging, and I expect to see it stolen within two years by somebody with a bigger budget and less nerve.

Play it on PC if you have the option; the PS5 version is solid and the DualSense work is a nice touch. Play it in long sessions in the dark, and stop when the runs start feeling like a commute, because the first fifteen hours are as distinctive as anything released this year and there is no shame in leaving a game while you still like it.

Then go and takeDredge out for the same bargain in a boat, andStill Wakes the Deep when you want the same weather with none of the control.

Spoilers below

The Remnant material is where the fiction and the systems finally converge. The zone isn’t a disaster site; it’s an experiment that never stopped running, and the deeper you go the clearer it becomes that ARDA’s people didn’t lose control so much as decline to regain it.

What I keep turning over is the ending’s implication about the wagon. The whole game has been quietly suggesting that the car is a participant — the quirks that feel like preferences, the way the vehicle keeps turning up where you left it — and the late-game revelations make that literal enough to reframe every hour you spent with the diagnosis board. You weren’t debugging a machine. You were negotiating.

Which means the quirks system was never a repair mechanic. It was a conversation in the only language available, and the game had you fluent in it before it told you anybody was listening. Reveals rarely earn themselves that thoroughly, because most of them arrive as information. This one arrives as a re-reading of every hour you already spent at the diagnosis board, squinting at a wiper motor, being answered.

There is a moment about two hours into The Lost Crown where you get the dash, and the game stops being a competent Metroidvania and becomes something you have to be physically removed from. It’s the air-dash, the Shadow of the Simurgh, and it does what every dash does — crosses a gap, cancels a state, opens the old rooms — except that Ubisoft Montpellier tuned the acceleration curve on it with what I can only describe as malice. It snaps. Sargon leaves a smear of light and arrives somewhere with the momentum still in your thumbs.

I have been playing platformers since a C64 and a tape deck, which is to say I have been playing platformers for long enough to be extremely boring about how things feel. This one feels correct.

What it is and what happened to it

Ubisoft Montpellier released Prince of Persia: The Lost Crown on 15 January 2024, on PS5, PS4, Xbox Series, Xbox One, Switch and PC. It’s a 2D side-on Metroidvania. You play Sargon, the youngest of a warrior band called the Immortals, sent to Mount Qaf to retrieve the kidnapped Prince Ghassan and finding that Mount Qaf has some opinions about the passage of time.

Montpellier are the studio behind Rayman Origins and Rayman Legends, and you can feel that pedigree in every frame of the animation. They are also, historically, the studio Ubisoft lets make the strange one.

The release window was a bloodbath. The Lost Crown landed in the middle of January, four days before Palworld appeared from nowhere and ate the entire conversation for a month, and a fortnight before Tekken 8 and Like a Dragon: Infinite Wealth. A mid-priced 2D Metroidvania under a dormant brand, from a publisher whose business is open worlds with towers in them, went out into that and — predictably — struggled to be heard. Ubisoft did release a demo, and the demo is genuinely one of the better ones anyone has shipped; it just needed someone to look at it.

The game deserved better weather.

The parry is the whole conversation

Combat is built on a single flash. Enemies telegraph attacks with a yellow glow, and a yellow-glow attack can be parried on a tight window. Some attacks glow red, and a red attack cannot be parried at all — you dodge, or you eat it.

That’s it. That’s the grammar. And it works because the game commits to it absolutely: the yellow-red distinction is honoured by every enemy in the game including the final ones, the window doesn’t shift depending on the arena, and the parry animation gives you a distinct, tactile, slightly ridiculous counter-flourish that makes you want to do it again.

The obvious ancestor isSekiro, and the debt is unhidden. What Montpellier changed is instructive. Sekiro’s parry is a posture economy — you’re parrying to build towards a break, so the parry is a resource action, and missing one costs you accumulated work. The Lost Crown’s parry is a state action. Hitting it doesn’t build a meter towards a win condition; it opens a window. The pressure is lower and the rhythm is faster, which suits a 2D plane where you can see the whole fight at once and there’s no camera to fight.

That decision is why the game reads as breezy where Sekiro reads as an exam.Nine Sols took the same parry into 2D and kept the exam. Both are right. They just want different things from you.

Layered on top: Athra, a meter you fill by fighting well and spend on special attacks, and an amulet system where you slot buffs into a limited number of sockets. The amulets are the light-RPG layer and they’re fine — a couple of them change how you play, most of them adjust a number. The build depth here is shallow and I don’t think that’s a flaw; the game’s argument is that execution is the interesting axis, and it stakes everything on that.

The map system is the actual achievement

Here is the part I want people to steal.

Metroidvanias have a memory problem. You find a locked thing at hour two, you get its key at hour nine, and in between you are supposed to have remembered a particular ledge in a particular room out of two hundred. The genre’s answers are map markers you place manually, which are a symbol soup you stop reading, and wikis, which are a confession of failure.

The Lost Crown gives you Memory Shards. You stand at the thing you can’t do yet, press a button, and the game takes a screenshot and pins it to that spot on the map. Later, when you have the tool, you open the map and you’re looking at an actual picture of the actual obstacle, and you know instantly whether your new ability solves it.

It’s obvious. It’s so obvious that its absence from twenty years of the genre is an indictment. And it does something subtler than convenience: because logging a puzzle is now cheap and precise, the designers could afford to be much denser with locked content than they otherwise could. Mount Qaf is packed with things you can’t do yet, and the density never becomes anxiety, because the game gave you a filing cabinet.

CompareMetroid Dread, which solved the same problem by narrowing the map into a mostly-linear tube with guided doors. Nintendo removed the navigation burden by removing the navigation. Montpellier kept the sprawl and gave you a better tool. Theirs is the harder trick and the more generous one.

The accessibility options are cut from the same cloth. There’s a platform assistance option, adjustable enemy damage, a guided or exploration mode for the map. None of it is buried in a menu apologising for itself. You can dial the game to the shape of your evening, which for a design this tightly tuned is a remarkably confident thing to allow.

Where it fights itself

The story is the weak axis. The Immortals — Vahram leading, Anahita among them — are drawn broadly, the voice performances swing between committed and stranded, and the plot’s shape is visible from the second hour. Mount Qaf itself is a better character than anyone standing on it.

The art direction is the other place opinion divides. The environments are gorgeous, the animation is superb, and the character designs land somewhere near a mid-budget anime and will not be to everybody’s taste. This got a lot of attention at reveal and it is, having played it, the least interesting thing about the game.

The bigger structural complaint: the back third leans on combat arenas — lock the doors, spawn three waves — as a pacing tool, and after twenty hours of the best 2D platforming Ubisoft has ever produced, being asked to stand in a box and fight is a demotion. The platforming challenge rooms in the late game are the answer the designers already knew was better. There should have been more of those and fewer boxes.

The verdict

The Lost Crown is the best-feeling 2D action game a major publisher has shipped in a decade, and its map system is a genuine contribution to a genre thirty-eight years old. It moves, it’s generous, it respects both your reflexes and your schedule, and it is completely uninterested in wasting your time — which from a company that built its reputation on 200-hour checklists is close to an apology.

Buy it on whatever you own. It runs well everywhere, including Switch, which given the density of what’s on screen is its own small piece of craft.

The tragedy is a scheduling decision. Somebody at Ubisoft looked at a calendar in January 2024 and decided this was the moment, and then Palworld happened, and the best thing the company put out in years went past most people at a distance.

If it hooks you, go and playMetroid Dread for the other end of the design argument, and thenNine Sols when you want the parry to hurt.

Spoilers below

The time-loop material in the back half is where the design and the fiction finally shake hands. Mount Qaf’s temporal instability stops being set dressing the moment you get the abilities that let you manipulate it directly — the shadow clone you place and recall, the frozen platforms — and the game starts building rooms that are essentially a musical bar you have to play in the correct order.

That’s when it becomes clear what the Simurgh powers are really for. They aren’t traversal upgrades with a story hat on. Each one is a new verb in a puzzle grammar, and the late-game rooms conjugate all of them at once: place the clone, dash to it, recall, use the recall’s momentum to reach the thing the dash alone couldn’t. Those rooms are the peak of the whole game and there are maybe a dozen of them.

Vahram’s turn is telegraphed roughly the instant he opens his mouth, and the betrayal lands anyway, for a purely mechanical reason: you fought alongside him in the opening, so the game taught you his moveset as an ally before it made you answer it as an enemy. Nobody says anything clever about it. The game just trusts that your hands remember. That’s the whole design philosophy in one boss fight.

Most games measure difficulty against you. You get better, so they get harder; you fail, so they ease off. Risk of Rain 2 measures difficulty against the clock on the wall, and the clock does not care whether you are ready.

There is a readout in the top-right corner. It tells you how long you have been alive and it tells you, in a word, how much trouble you are in. It starts at Easy. Given long enough it stops using words that mean anything and starts shouting at you. Nothing you do slows it down. You can stand perfectly still in an empty corner of the first map and the number keeps climbing, the enemies keep getting more numerous and more expensive to kill, and the run you were carefully building gets taken away from you by arithmetic.

That single decision — difficulty as a function of elapsed time rather than progress — is what makes this game work, and it’s why it has outlasted a good many slicker things released since.

What it actually is

Hopoo Games released Risk of Rain 2 into Early Access in March 2019 and hit 1.0 in August 2020, on PC and every current console. It is the 3D sequel to Risk of Rain (2013), a 2D side-scroller with the same architecture and a much meaner silhouette. Gearbox published both and later bought the property outright.

The loop is simple enough to explain in a lift. You drop onto a stage as one of a roster of survivors — Commando, Huntress, Engineer, MUL-T and a dozen others, each unlocked by a specific challenge. You kill things, which drops gold. You spend gold on chests, which drop items. Items are permanent for the run and they stack: two of a thing is roughly twice the thing, six of a thing is a problem for whatever is standing in front of you. Somewhere on the map is a teleporter. You activate it, survive the boss and the charge window, and move to the next stage. After five stages you loop back to the beginning with everything harder and everything you own still in your pocket.

Items are the whole texture. A white common item that gives a small chance of chaining lightning is a shrug at one stack and a screen-clearing weather system at twelve. The red-tier legendaries change the rules rather than the numbers — one of them, famously, lets you cheat death once per stage, and the run in which you find it becomes a different run. Lunar items are the interesting ones: they come with an explicit cost, a real downside written on the tin, and picking one up is the game asking whether you understand your own build well enough to pay.

The clock is the design

Here is the thing everything else hangs from. Gold scales with time. Chest prices scale with time. Enemy health and damage scale with time. So the more chests you open, the stronger you are — and the longer you spent opening them, the stronger everything else is.

That’s not a difficulty setting. It’s an economy with an interest rate.

Every decision in a run is the same decision wearing a different hat. There’s a chest on a ledge across the map. Getting there costs ninety seconds. Ninety seconds is worth a certain amount of enemy scaling, and the item in the chest is worth an unknown amount of power, and you have to price that trade with incomplete information while a horde is assembling behind you. Do it right and you leave the stage marginally ahead. Do it four times in a row on a stage where the chests roll badly and you leave the stage behind the curve, and being behind the curve compounds, because a weaker character kills more slowly, and killing more slowly takes more time, and time is the thing that is hurting you.

The genius is that the punishment for greed is never immediate. Spelunky’s ghost arrives at two minutes thirty and tells you off in person. Risk of Rain 2 lets you overstay, gives you the loot, sends you happily to the next stage, and then kills you eleven minutes later with a bill you signed without reading. You almost never die of the mistake you just made. You die of the mistake you made two stages ago, and by the time you understand that, you’ve internalised the pacing in a way no tutorial could have taught you.

Compare the timed doors inDead Cells, which do something adjacent and much more legible: run fast, get a reward, and the reward is gone the instant the timer expires. That’s a clean, honest bargain you can evaluate in a second. Risk of Rain 2’s bargain is smeared across an hour and you’re never quite sure you got it right. The uncertainty is the product.

Why the power fantasy lands

Roguelikes have a structural problem with escalation. If the player gets strong enough to trivialise the content, the game stops being interesting; so most designs cap the player, or scale the enemies to match, or reset the whole apparatus every run. Hopoo went the other way and let the ceiling off entirely.

By the third loop a well-built survivor is a war crime. The screen is a smear of proc effects, ricochets, mortars, satellite lasers and burning ground, and you are killing bosses in the time it takes them to finish their spawn animation. This is the same joy thatVampire Survivors later refined into a pure distillate — the pleasure of watching your own build outgrow your ability to follow what it’s doing.

The difference is that Risk of Rain 2 charges for it. Vampire Survivors ends at thirty minutes and hands you the victory lap. Risk of Rain 2’s clock keeps ticking past the point where your build is godlike, and the scaling curve is exponential where your item stacking is roughly linear. So the god phase is a phase. It has a shelf life. You are always, at every moment of the run, watching two lines on an invisible graph and trying to guess where they cross — and the right play is often to leave a stage with money in your pocket and chests unopened, which feels physically wrong and is correct.

Very few games make walking away from free loot into a skill.

Where it fights itself

The 3D translation cost something real. The 2013 original was a side-scroller, and a side-scroller tells you where everything is. In three dimensions, with a camera behind your shoulder and enemies spawning off-screen in every direction, the fair-fight legibility goes. Late in a run you will be killed by something you never saw, from a direction you had no reason to check, and the game’s answer to that is a minimap you don’t have and a sound mix that is already saturated. The tension between “readable combat” and “the screen should be chaos” never quite resolves.

The other cost is the ramp-in. The first ten minutes of a run are, by design, the least interesting ten minutes. Difficulty is low, items are few, and you are essentially doing paperwork to build a character. Once you’ve had four hundred runs, those minutes are a chore you tolerate. The scaling clock justifies it — you cannot skip the early game without also skipping the item economy — and I still think it’s the strongest argument for the design and the most obvious tax on your evening.

Multiplayer is where the seams show most. Up to four players share a stage, item drops don’t scale cleanly, and the difficulty coefficient rises with player count in a way that makes an uncoordinated four-stack a farm and a coordinated one a rout. It’s a joy to play with friends and it is not a balanced experience, and Hopoo were fairly upfront that it was never going to be.

The ancestor

Everyone reaches for Isaac or Spelunky here, and the resemblance is surface. The real ancestor of Risk of Rain 2 is the arcade timer — the coin-op design where the machine’s job is to end your session on a schedule, whatever your skill, and skill only buys you a longer schedule. Every one of those cabinets had a hidden rank system pushing back against a player who got too good.

The trick Hopoo pulled is exposing that clock, putting it in the corner of the screen, and making it the thing you play against rather than the thing that plays against you.Hades solved repetition by making the loop mean something narratively. Risk of Rain 2 solved it by making the loop mean something economically. Both are answers to the same question: why should I press start again? Hopoo’s answer is that last time you left two chests behind and you’ve spent the intervening hour wondering whether you should have.

The verdict

Risk of Rain 2 is the most honest tension engine in the genre. It gives you one number, tells you the number is your enemy, hands you every tool you need to manage it, and then quietly makes managing it the entire game. The combat is loose, the camera is a liability, the early minutes drag, the multiplayer is lopsided — and none of that matters much, because the thing it does is something almost nothing else does, and it does it for as long as you keep asking.

Play it on PC, where the mod scene has been carrying it for years and where the frame rate survives what the third loop does to the entity count. The console versions are complete and competent; they just wilt at the top end, which is where the game is.

Then, if the loop takes, go and see whatVampire Survivors does when you remove the ceiling and the aiming. It’s the same drug with the difficulty clock swapped for a shorter fuse.

Spoilers below

The reason the loop mechanic works — going back to Stage 1 with a difficulty coefficient that has never reset — is that it makes the map geography a memory test rather than a discovery. You already know where the chests are on Titanic Plains. You know where the shrine spawns cluster. So the second loop asks you to run a route you know at a speed you can’t quite sustain, and the pleasure flips from exploration to execution without changing a single asset.

The obliteration ending is the part I keep thinking about. You can end a run voluntarily, at the obelisk, by choosing to erase yourself. It’s the only exit that isn’t death and it costs you the run’s rewards. A game built entirely on the tension between greed and the clock offers you, as its cleanest ending, the option to stop wanting things. That’s a better joke than it has any right to be, and it took me a long time to notice it was a joke about me.

Every role-playing game I loaded off tape as a kid had the same secret: the character sheet was a list of ways to hurt things. Strength was for hitting. Dexterity was for hitting first. Intelligence was for hitting at range with a fireball. The numbers described a weapons platform with a name on it.

Disco Elysium takes the sheet, keeps all twenty-four numbers, and removes the thing they were for. There is no combat system. Nothing to swing at, no initiative order, no HP bar on a monster. And the sheet ismore dangerous than any of its ancestors, because ZA/UM worked out that if your stats aren’t fighting the world, they can fight you.

Released 15 October 2019 on PC, expanded into The Final Cut in March 2021 with full voice acting and a run of political vision quests, then onto PlayStation, Switch and Xbox — it took four awards at that year’s Game Awards, including Best RPG and Best Narrative. Four years on, nothing has caught it, and the reason is structural.

The premise, briefly

You wake up in a trashed hostel room in Martinaise, a rotting district of a city called Revachol that lost a revolution decades ago and has been paying for it since. You have no name, no memory, and a body that has clearly been in a fight with a week. There’s a man hanging from a tree in the courtyard behind the hostel. You are, it emerges, the detective assigned to that.

Your partner is Lieutenant Kim Kitsuragi, who is competent, patient, and watching. Kim’s presence is the finest piece of characterisation in the medium’s last decade for a single mechanical reason I’ll get to.

You have four days.

The sheet talks back

Four attributes — Intellect, Psyche, Physique, Motorics — with six skills apiece. Twenty-four.

And each of the twenty-four is avoice. They interrupt. Logic offers a deduction. Inland Empire tells you the necktie is speaking. Electrochemistry wants you to find out what’s in the fridge and drink it. Shivers reports what the city itself is doing three streets away. Half Light is convinced everyone in the room is about to kill you. Volition tries to hold the whole lot together and mostly fails.

Here’s the design move, and it’s the one everything else hangs off:investing in a skill makes it louder. Points don’t buy competence, they buypresence. Put six points in Inland Empire and you are a man who receives visions and cannot tell whether they’re true, because the game will keep handing you visions and never confirm them. Put six in Electrochemistry and you will be offered a drink in every conversation, by yourself, from inside your own skull. A high number in a normal RPG means you succeed more. A high number here means that part of youwins the argument more often, and some of those arguments should be lost.

That reframes the build entirely. You are not optimising a tool. You are choosing which of your own tendencies gets a megaphone, and then living in the room with it. It’s the only stat system I’ve played where the min-maxer’s instinct — dump everything into the good number — reliably produces aworse life.

The failure design finishes the job. Skill checks are 2d6 with modifiers, and they come in two colours: white checks, which you can retry once something about you or the world has changed, and red checks, which resolve once and stay resolved. A red check failed is a door closed forever, and Disco Elysium’s whole reputation rests on those doors being more interesting from the outside. Failing to open a fridge in this game generates a better forty minutes than succeeding at most quests elsewhere.

Morale is a health bar

The bit I don’t think gets enough credit: there are two damage tracks. Health, which the world takes off you, andMorale, whichideas take off you.

Be humiliated in a conversation and Morale drops. Have a memory surface that you weren’t ready for and Morale drops. Run out of Morale and the game ends — you are dead, in the full and final sense, having been killed by a thought.

That single system does more thematic work than every grimdark narrative in the genre put together, because it makes the abstractionliteral at the mechanics level. This is a game about a man held together with string, and the string has a number, and you can watch the number go down while somebody is being unkind to you about your jacket. You heal it with cigarettes, with a drink, with a small act of self-mythology — each of which is a bad long-term idea that the systems reward in the short term, which is the most honest model of self-destruction anybody has shipped.

And Kim. Kim is a mechanic. He is standing next to you during almost every conversation, he reacts, and the game tracks how he’s reading you. There’s no visible approval meter to game — you just have a colleague, and you can feel him deciding. That absence of a bar is why he works. Every RPG companion since Baldur’s Gate has been a slot machine you feed dialogue options into for a romance payout. Kim is a person whose respect you can lose without a notification.

The ancestors

Planescape: Torment (1999) is the acknowledged parent and the comparison everyone reaches for, and it’s fair: a text-first RPG about an amnesiac, where the great question is what can change the nature of a man. Torment had combat, and Torment’s combat was the worst thing about it — a Infinity Engine obligation bolted to a novel. Disco Elysium is what you get when a studio finally has the nerve to cut the obligation.

The deeper ancestor is thirty-nine years old and ran on the machine under my telly.Ultima IV: Quest of the Avatar, 1985, C64 among others — the RPG that removed the final boss and replaced him with an ethics exam. Richard Garriott’s game watched what you did, tallied it against eight virtues, and the win condition was becoming a particular kind of person. It was clumsy, it was gameable, and it was the first time a computer RPG asserted that the interesting variable was the player’s character rather than their arithmetic. Everything Disco Elysium does with the Thought Cabinet — where you internalise an idea over real hours of play, and it changes your stats permanently, sometimes for the worse — is Ultima IV’s proposition with forty years of writing craft applied to it.

The Thought Cabinet deserves its own paragraph, actually. You pick up an idea. It occupies a slot. Itcooks — for an hour of play, while you do other things — and then it resolves into a permanent modifier, and you can’t easily get rid of it. That’s the best model of how convictions actually work that I’ve seen in software. You don’t decide to believe something. You carry it around until it sets.

Where it fights itself

The Final Cut’s voice acting is magnificent and it slows the game down, and I’ve never fully squared that. Disco Elysium is areading machine — a million-odd words, a prose rhythm that rewards skimming back and forth, an internal monologue that works at the speed your eye moves. Full VO puts a metronome on it. Lenval Brown’s narration is extraordinary and it is also a pace limiter on a game whose original virtue was letting you set your own. I’d still recommend the Final Cut, and I understand people who hard-disagree.

The four-day clock is the other tension. The game presents time pressure and then, for long stretches, doesn’t actually apply it — you can wander Martinaise indefinitely within a day, and the deadline turns out to be softer than the framing implies. That’s a mercy for the player and a small dishonesty in the design.

And the save button undermines the red checks. ZA/UM built a game where failure is the best content, and then shipped it on platforms where reloading takes four seconds. The design’s central bet requires a discipline it can’t enforce, and most players will quietly welch on it at least once. I did.

I’ll note the obvious and stop: ZA/UM’s ownership has been through the Estonian courts since 2022, and a planned expansion was cancelled alongside layoffs in 2023. That’s the public record. What it means for anything after this game is a matter for lawyers, and it doesn’t change what’s on the disc.

The verdict

Disco Elysium is the most important RPG since Torment, and its importance is almost entirely mechanical rather than literary — which is a strange thing to say about a game with a million words in it. The writing is extraordinary. Plenty of games have extraordinary writing bolted onto a combat engine that resents it. What makes this the one is that ZA/UM turned the character sheet into a psychology, made the damage tracks include your dignity, and then had the nerve to delete the swords.

You will fail at things. The failures will be the parts you talk about at dinner. That’s the design working exactly as specified, and four years on nobody has copied the trick properly, because copying it requires giving up the thing every publisher thinks a game is for.

Play the Final Cut. Take your time. Let Kim decide about you.

What to play next:Citizen Sleeper, which takes the “your stats are your circumstances” idea and makes it a clock;Norco for the same species of rotting-place writing; andBaldur’s Gate 3 if you want to see what the genre looks like when it keeps the swords and gets everything else right.

Spoilers below

The Thought Cabinet’s cruellest entry is the one that makes the whole system legible: some thoughts are traps. Internalise the wrong conviction and you take a permanent penalty and a permanent identity, and the game charges you skill points to forget it. Nothing warns you. The idea looked good when you picked it up, the way ideas do.

The political vision quests added in the Final Cut are the sharpest thing in the expanded version and the most misunderstood. Each ideology — communist, fascist, moralist, ultraliberal — will let you commit, reward the commitment with a mechanical bonus, and then show you the sad little end of that road. The quests aren’t lectures. They’re the game applying the same logic it applies to Electrochemistry: hold a belief loudly enough and you become a man who holds that belief loudly, and there’s a scene waiting at the end where you get to look at him.

The Deserter and the phasmid are where the whole architecture pays out. The case resolves through police work — real, sourced, tedious, sit-down-and-do-the-interviews police work — after fifty hours in which the game has been feeding you visions, hunches and Inland Empire’s psychic nonsense. And then the phasmid arrives, and it’sreal, and the game refuses to say which of your two epistemologies was correct. Both. Neither. Harry gets to keep his string.

That’s the closing argument for the no-combat design. A game with a fight in it would have had to pick a winner. Disco Elysium ends on a man standing in reeds looking at something impossible, having solved the murder by asking questions, with a colleague who has decided about him. There’s nothing to hit. There never was.

You are sitting at a table in a cabin. The man opposite you has no face, only two eyes floating where a face should be, and he is dealing you cards. You will play cards for a while. Eventually you will realise you can stand up.

That is the moment Inscryption is built around, and everything else — the found footage, the ARG, the three genre swaps, the card game that turns out to be four card games — descends from it. Daniel Mullins Games released it on 19 October 2021 for PC through Devolver Digital, with PlayStation following in August 2022 and Switch that October. Two years on, it’s still the best answer I know to a question most games never ask: what is thetable for?

The card game is real

Start with the part people skip, because it’s the part that makes the rest possible.

Inscryption’s deckbuilder is a genuinely good deckbuilder. Four lanes, creatures with power and health, a scale between you and your opponent that tips when damage gets through — win by tipping it far enough your way. Simple enough to read in a minute.

The costing is where it gets its teeth. Most creatures costblood, and blood comes from sacrificing creatures already on your board. So playing a strong card means killing a weak one you played earlier, which means every turn is a small act of asset-stripping your own tableau. Other cards costbones, which you get from things dying — yours or his. The economy is entirely built out of loss. You cannot spend anything in this game that you didn’t first destroy.

Compare that to Magic’s lands, which are a resource you set aside and grow. Inscryption’s currency isattrition, and it makes the play feel morally different on a turn-by-turn basis. You get good at feeding squirrels into a stag. It becomes routine. The game notices that it has become routine for you, and that noticing is the first crack in the wall.

The frame, and the breaking of it

Act 1 is the cabin. You’re playing a roguelike run against Leshy, the eyes across the table, who narrates your journey across a little map board and cheats when he feels like it. Lose and you die; die and you make aDeathcard — a custom card built from your failed run, sigil and stats and a portrait you draw yourself — which gets shuffled into the pool for future attempts. Your failures become furniture in the next game.

And you can stand up. You can walk around the cabin. There’s a safe, a clock, a wolf pelt, a carving. The escape room is not a bonus; solutions found in the room change the card game, and cards found in the game unlock the room. The two layers feed each other, which is the trick that makes the framing device structural rather than decorative.

Then Act 2 happens and the entire visual register drops to 8-bit — an overworld, four rival card-masters, four wholly different rule-sets grafted together into one deck. Then Act 3 happens and it’s a different thing again, with a different currency and a different narrator and a different joke.

Threaded through all of it is a man called Luke Carder, filming himself, opening card packs on a webcam, playing the disc he shouldn’t have. The found footage sitsoutside the cabin the way the cabin sits outside the card game. Frames all the way down.

Why it works, mechanically

Here’s my systems read, and it’s the reason I keep recommending this over the other meta-games.

Fourth-wall breaking is cheap. It has been cheap since Psycho Mantis read your memory card in 1998 and since Eternal Darkness faked a TV channel change in 2002 — both brilliant, both agag, a thing that happens to you once and then the game resumes. The gag works by violating an expectation, and an expectation can only be violated so many times before the violation becomes the expectation.

Inscryption solves this by making the framea mechanic with a cost. When Leshy interferes, it changes your odds. When you stand up from the table, you’re spending time you could have spent playing. When the room gives you something, it goes in your deck and has stats. The meta layer never gets to be free commentary, because everything it does has to be paid for in the currency of the card game underneath. That’s the discipline Mullins learned between Pony Island in 2016 and here: Pony Island is a magic trick about a game, and Inscryption is a game that happens to contain magic tricks.

The genuine ancestor, though, is older than either, and it’sHacker — Activision, 1985, which I loaded on a C64 as a kid and which opened with no instructions whatsoever, just a login prompt and the flat lie that you’d broken into something you shouldn’t have. There was no manual to consult because theabsence of the manual was the design. Inscryption is doing the same thing with forty years of extra technique: withholding the frame so that finding the frame is the reward.Tunic is playing the same game from the opposite direction, handing you a manual you can’t read;Hypnospace Outlaw does it by making the interface itself the level. All three descend from the era when a game could justlie to you and there was no wiki to check.

Where it fights itself

Act 1 is the best hour of card game released this decade, and Inscryption spends the next several hours moving away from it.

That’s a design decision, and I understand it, and I still think it costs the game real weight. Act 2’s four-scrybe mash-up isbroad — every mechanic from every act, all at once — and breadth is the enemy of the tension Act 1 built. The blood economy worked because it was the only lever. Give me bones and energy and mox gems and a pixel overworld, and I’m no longer making the one horrible decision over and over; I’m managing a system. The horror was in the narrowness.

Act 3 recovers some of it by imposing a new constraint, and by then the story has enough momentum to carry a weaker table. But the shape of the game is a diminuendo in mechanical terms while the plot escalates, and the two curves fight.

The evidence for that reading is Kaycee’s Mod, the free update Mullins put out in March 2022: Act 1’s cabin, extracted, made endlessly replayable, with escalating challenge modifiers and no story at all. It exists because players finished the game and wanted thefirst part back. When your own post-release content is an admission that your best system was in the opening act, the review writes itself.

The other cost is structural and unavoidable: this is a game that can only be new once. Deckbuilders are usually re-playable machines — that’s the genre’s entire economic proposition. Inscryption’s power lives in the reveals, and reveals don’t survive a second viewing. Kaycee’s Mod is the patch for that, and it’s a good one, and it’s also a card game with the meaning surgically removed.

The verdict

Inscryption is a game about the difference between playing and being played, and it earns that sentence by being a genuinely excellent card game first and a conceptual stunt second. The order matters. Plenty of games have broken their frames; almost none of them built something strong enough underneath that the breakhurt.

It’s uneven. It peaks early, sprawls in the middle, and ends on a note that some people will find inevitable and some will find like a rug being pulled for the fourth time in six hours. None of that stopped it from being the thing I’ve thought about most since 2021, and the eyes across the table are still there when I close mine.

Go in knowing nothing. Then come back and play Kaycee’s Mod for a hundred hours like the rest of us.

What to play next:Tunic, for the other great modern game about withheld information, andHypnospace Outlaw, which builds a whole world out of an interface that pretends to be furniture.

Spoilers below

The Luke Carder framing is the piece that turns Inscryption from a good trick into a coherent argument, and it took me a second run to see how carefully it’s wired. Luke isn’t a narrator — he’s aplayer, and specifically a player of the kind the internet manufactures: a collector, a completionist, a man who cannot leave a mystery alone and who films himself failing to leave it alone. Every terrible decision he makes is the decision the audience is loudly telling him to make, which is the joke and also the indictment.

The Karnoffel Code and the OLD_DATA business is the weakest strand, honestly. It’s the ARG scaffolding, and it exists to give the found-footage layer a reason to have stakes. It works well enough to hold, and it’s the one place where Mullins reaches for a conspiracy plot because he needs a plot rather than because the plot is the idea.

What redeems the final act completely is the deletion. Inscryption ends by having the gameremove itself, and it commits — cards get erased, systems get switched off, and the last thing you do is the thing that ends the possibility of doing it again. For a genre built on the promise of infinite runs, closing on permanence is the single most aggressive design choice in the game. Leshy’s whole tragedy is that he only ever wanted to keep dealing. Kaycee’s Mod, arriving five months later to give the cabin back forever, plays like Mullins conceding the point to him.

There’s a moment, some way into Alan Wake 2, when the game stops being a game and becomes a music video — a full-blown, choreographed, live-action rock number with the studio’s own creative director hoofing about in it, staged inside a survival horror published by the company that makes Fortnite.

Nobody made Remedy do that. That’s the whole review, really. Alan Wake 2 is a studio taking the biggest swing available to it, with somebody else’s money, on a sequel to a game from 2010 that most publishers would have quietly filed under “fondly remembered”.

It shipped on 27 October 2023 for PC, PlayStation 5 and Xbox Series X/S, published by Epic Games Publishing, digital-only, no disc. Directed by Sam Lake and Kyle Rowley. It is Remedy’s first actual survival horror after twenty-odd years of making third-person shooters that were secretly about something else, and it is the least compromised thing they have ever put out.

Two games in a trench coat

You play two people.

Saga Anderson is an FBI agent who arrives in Bright Falls to investigate a series of ritual killings, and her half is the procedural: rain-soaked Pacific Northwest towns, a caravan park, a flooded resort called Watery, a lot of walking through woods with a torch. Her signature system is theMind Place, a mental room she can step into at any moment to pin evidence to a case board, connect threads, and profile suspects by sitting in an imaginary chair opposite them.

Alan Wake has been in the Dark Place for thirteen years, writing a story to get out, and his half is a nightmare version of New York where the geometry lies. His system is theWriter’s Room, where he swaps out plot elements — pick a different angle for a scene, and the physical space rearranges to match the new draft. He also carries an angel lamp that can lift light out of one place and drop it into another, changing the whole state of a level between two versions of itself.

You can switch between them almost freely after the opening hours. Two protagonists, two structures, two visual languages, one story that only closes if you’ve been in both.

The reason this works, when it works, is that the two halves arearguing. Saga’s method is evidence: things happened, they can be established, a board can hold them. Alan’s method is authorship: things happen because someone wrote them, and the board is where you decide what’s true. Putting a detective and a novelist in the same plot and giving each of them a corkboard is a genuinely good joke, and the game is smart enough to know it’s a joke and serious enough to build both systems properly anyway.

The Mind Place problem

Except one of those boards doesn’t actually work, and it’s worth being specific about why, because it’s the clearest case in the game of the design fighting itself.

Saga’s case board cannot be wrong. You collect a piece of evidence, you go into the Mind Place, you drag it to the slot that lights up, and a thread appears. The game will not let you connect the wrong things. There’s no failure state, no dead end, no bad theory. It is aritual of deduction rather than deduction — the pleasure of tidying, dressed as the pleasure of thinking.

Compare it withReturn of the Obra Dinn, where being wrong is the entire texture of the experience and the game only confirms you in batches of three so you can’t brute-force it. Obra Dinn treats you as a reasoning adult who might fail. The Mind Place treats you as a reader who needs the plot restated in a nice room.

And yet — I don’t hate it, and here’s the honest complication. Alan Wake 2’s story is deliberately hard to hold, and the case board is where Sagasays out loud what she’s just worked out. It’s a comprehension aid with a lovely UI. As a puzzle system it’s theatre. As a narrative instrument, it’s the reason a plot this strange stays legible for twenty hours, and I’d rather have the legibility than a fake puzzle I’d have looked up anyway.

The Writer’s Room is the better system for the reverse reason: swapping a plot beat visiblychanges the level. The idea has consequences you can walk through. It’s the same trick Alan’s angel lamp pulls with light, and the same trickControl pulled with the Oldest House rearranging itself behind you. Remedy have been building the “the building is the plot” mechanic for years, and here it finally has a novelist in it, which is what it always wanted.

The horror, and the shooting

Survival horror, properly. Small inventory, real ammunition scarcity, save rooms with a locker, enemies that require a light-then-shoot two-step that makes every encounter cost something.

The light mechanic from the first game returns and is finally load-bearing. Enemies wear a shadow shield; you burn it off with the torch, and the torch has batteries. So a fight has a resource sequence — light, then bullets, then run — and the horror comes from the arithmetic breaking down halfway through. That’s a good system, borrowed knowingly from the genre, and the sound design sells it: the Taken don’t shamble so much asmutter, and the muttering arrives before they do.

The shooting itself is the weakest thing here, and Remedy know it. The guns are deliberately heavy and imprecise, dodging is a lurch, and after eight hours the encounters stop escalating — you fight the same handful of shapes in the same handful of ways, with more of them. Remedy have made some of the best third-person combat ever built. They have chosen, correctly, to make the combat in their horror game feel bad. The problem is that they didn’t cut enough of it, and a system you’ve deliberately made unpleasant should be rationed like the ammunition is.

The other soft spot is the middle. Alan’s Dark Place chapters are the most inventive material in the game and also the most repetitive traversal — the light-swapping is a puzzle you solve about nine times, and by the sixth it’s a chore with a beautiful skybox on it.

Why the swing lands anyway

Because of the confidence. Alan Wake 2 is full of things that should not survive a pitch meeting.

It has live-action FMV cut into it constantly — Ilkka Villi’s face, Matthew Porretta’s voice, a whole in-fiction talk show, an in-fiction TV anthology. FMV has been a punchline since the CD-ROM boom in the mid-90s, when every studio with a camcorder shoved grainy actors into a DOS game and called it cinema. Remedy have spent twenty years quietly refusing to let the idea die, from Max Payne’s photo-comic panels onward, and Alan Wake 2 is where it stops being a stylistic tic and becomes the actual grammar. The live-action isn’t cheaper than the engine. It’sanother layer of the fiction, deployed because a story about an author trapped in his own draft should keep showing you the seams.

The nearest recent relative isImmortality, which also understood that footage of a real face carries a charge no rendered model can fake. Immortality is the more rigorous experiment. Alan Wake 2 is the one that got a AAA budget and used it to stage a musical.

And the Remedy Connected Universe finally justifies itself. Alex Casey — Sam Lake’s face, James McCaffrey’s voice, a hard-bitten cop from a series of novels inside a game whose protagonist wrote them, played by the man who made Max Payne — is the kind of joke that only works if you’ve been paying attention for two decades. Control’s Federal Bureau of Control is here too, and it lands as a payoff rather than a homework assignment.

The verdict

Alan Wake 2 is the most interesting big-budget game of the year and it is not the mostenjoyable one, and I think Remedy would take that trade every time.

The combat drags, the middle sags, the case board is a magic trick pretending to be a mind. Against that: a studio that got handed serious money and spent it on a wordless twenty-hour argument about authorship, with a rock opera in the middle and a live-action talk show host doing the exposition. It is strange in a way games at this budget essentially stopped being around 2012, and the strangeness is not decoration — it’s structural, it’s the point, and it is worth more than another well-tuned shooting gallery.

Play it in the dark, in chunks, and let Saga’s board do the remembering for you.

What to play next:Control for the same studio building the same idea with concrete instead of ink, andImmortality for the purest version of the live-action gamble.

Spoilers below

The structural gag is that Alan is writing Saga’s story and Saga is investigating Alan’s, and both of them are right. Remedy commit to the loop hard enough that the game’s own chapters start behaving like drafts — the Initiation and Return labels aren’t flavour, they’re the manuscript’s table of contents, and the whole thing folds into a spiral where the ending is a beginning that’s been through a rewrite.

The chapter that makes the case for the whole project is We Sing. You walk into the Old Gods of Asgard material expecting a set-piece and get “Herald of Darkness”: a full musical number, live-action, that recaps thirteen years of a fictional writer’s biography in verse while you’re still holding a torch. It is the most expensive thing in the game and it has zero mechanical purpose, and it’s also the only sequence I’ve seen this year that made me put the controller down and grin at a wall. That’s the swing. A studio with a publisher breathing on it does not make We Sing.

The Dark Place’s rewriting of New York is where the Writer’s Room finally goes from clever to frightening — the moment the plot element you swap starts changing things you didn’t intend, and the tool you’ve been using to escape becomes the thing keeping you in. Alan’s arc lands because the mechanicis the theme: an author whose only power is authorship, in a place that grants it too literally.

The ending is a hinge into whatever comes next, and it’s the one place the ambition cashes a cheque it hasn’t earned yet. Remedy have built a universe and now owe it a resolution, and Alan Wake 2 closes on the confidence that they’ll get to make it. On this evidence, they’ve earned the benefit of the doubt.

Six slots. That’s the number everybody argues about, and it’s the right thing to argue about, because Signalis is a game where the number of pockets you have is a statement of intent.

rose-engine — Yuri Stern and Barbara Wittmann, two people — released Signalis on 27 October 2022 for PC, PlayStation 4, Xbox One and Switch, published by Humble Games with Playism handling Japan. A year on it has quietly become the survival horror that other survival horrors get measured against, and it managed that without a marketing budget, a famous engine, or a single frame of hardware showboating. It looks like something that fell off a shelf in 1998. It thinks like something written in 2022.

What it actually is

You are Elster, a Replika — a mass-produced synthetic worker unit, model LSTR-512 — and you wake in a wrecked ship on a frozen world with a photograph and a hole where your memory used to be. You are looking for someone. The game will not tell you clearly who, or why, or whether you have done this before.

The camera sits above and behind, roughly top-down, in the way the fixed cameras of the original PlayStation era sat: composed rather than convenient. The art is deliberately low-resolution and heavily dithered, the whole thing bathed in CRT scan and red emergency light. Enemies are other Replikas, and they shamble.

Mechanically it’s Resident Evil’s architecture, and it doesn’t pretend otherwise. Locked doors, keys in the wrong wing, save points that cost you a walk, a map that fills in as you go, an inventory that runs out constantly. Elster carriessix items. A pistol takes a slot. Its ammunition takes another. A health item takes another. The key you need is a fourth. Do that arithmetic in a corridor with something walking toward you and you’ll understand the entire design philosophy inside twenty minutes.

The one genuinely novel system is the radio. Elster carries one, you can tune it, and the world broadcasts on frequencies — puzzles are solved by finding a number somewhere and dialling it in, and the radio hisses and sings and occasionally says something it shouldn’t. It’s a diegetic puzzle interface that also functions as a mood machine, which is exactly the sort of double-duty this game does everywhere.

And the dead don’t stay dead. Down a Replika and it will get back up after a while, unless you burn it with a thermite flare — and flares take a slot too. That mechanic has an ancestor, and it’sResident Evil proper: the Crimson Heads of the 2002 remake, where a corpse you didn’t burn became a faster, angrier problem an hour later. Signalis takes the idea and makes it structural. Every kill is a debt with interest.

Why the constraint is the point

Here’s the thing I keep coming back to, and it’s why the six slots matter beyond the grumbling.

A limit in a game is usually a difficulty dial. Signalis uses its limit as agrammar. Six slots means the game can guarantee you are always making a sentence out of the same small vocabulary: what you carry, what you leave, what you walk back for. Every room becomes a decision about what you’re willing to be defenceless for. That’s the survival-horror bargain from 1996, and it works now for the reason it worked then — scarcity makes spacemean something, because space has a price.

But rose-engine push it further than the genre usually does, and this is the bit that earns the headline. The whole game is built out of compression. Fragments of text that don’t resolve. A word repeated in three different registers until it warps. Scenes that recur with one detail changed. Imagery lifted openly from Robert W. Chambers’The King in Yellow — the sign, the play, the thing you shouldn’t read — and from Stanisław Lem, whoseSolaris haunts the game’s central question about whether the person you’re looking for is a person or a projection of your own wanting.

That is how a poem works. It gives you less than you need and makes the shortfall do the labour. Signalis’s six-slot inventory and its six-fragment story are the same design instinct pointed at two different systems, and the fact that both land is why the game is more than a very good pastiche.

The mechanics, and where they fight the game

I’m generous to ambition and merciless about padding, and Signalis has one real problem: the back half asks you to carry the six-slot bargain further than the bargain can hold.

Early on, scarcity is tension. Later, when the puzzle chains get longer and the item you need is four rooms and two loading screens away, scarcity becomesadmin. You will do laps. You will stand in a save room playing inventory Tetris with a keycard and a flare, and the fear will drain out of the game while you do it, because nothing is chasing you in a save room and the horror doesn’t survive a spreadsheet.

This is the oldest tax in the genre and Signalis pays more of it than it should, largely because it’s honest to a 1996 template that had reasons — disc-loading reasons, memory-card reasons — that stopped existing decades ago. Resident Evil 4’s remake spentan entire redesign figuring out how to keep scarcity while deleting the errand, and it’s the one lesson Signalis declines to take.

Combat is the other soft spot. Aiming is stiff by design, and stiffness is a legitimate horror tool —Dead Space’s 2023 remake knows exactly how much friction to leave in a weapon so that firing it feels like a commitment. Signalis’s shooting is stiffand thin: the guns don’t have much character, and the enemies don’t react enough to make you feel the hit land. You end up avoiding fights, which is correct survival-horror behaviour, arrived at for slightly the wrong reason.

The thing it does that nothing else does

And then it does something I haven’t seen another game do this well.

Signalis makesthe interface itself unreliable. The screen is a device Elster is looking through, and the game knows it. Things get into the frame that shouldn’t. The presentation glitches in ways that are plainly authored, and because you’ve spent hours trusting the HUD as a neutral instrument, the moment it lies to you lands like a floorboard giving way.

The ancestor for thelook is older than the PlayStation. Watching Signalis, I kept thinking about Team17’sAlien Breed on the Amiga in 1991 — the same top-down corridors, the same doors that ate keys, the same feeling of a small bright thing moving through a big dark ship. Alien Breed was cheerfully arcade about it. Signalis takes the identical camera and uses it for dread, because from above you can see everything in the room and still not see what’s coming through the door.

The ending — there are several, and they hinge on things the game never tells you it’s counting — is where the poetry either takes or doesn’t. Signalis will not explain itself. It has a theory of what happened, it has left it in fragments across a dozen rooms, and it fully expects you to either assemble it or go and read someone else’s assembly. That’s a genuine choice with a genuine cost, and I respect it more than I enjoy it.

The verdict

Signalis is the best thing to happen to survival horror since the genre started remaking itself, and it came from two people who understood that the constraints were the art itself. It is a game about memory built out of a system that punishes you for carrying too much, which is a joke so good I’m not sure it’s a joke.

It’s also stiff, it’s fiddly, and it will make you walk a corridor eleven times. The last third is a puzzle box wearing a poem’s coat. Every one of those complaints is true, and none of them touched how the thing sat in my head afterwards, which is the only measurement I trust.

Go in cold. Take the flares. Don’t look anything up until the credits.

What to play next:Resident Evil 4 (2023) for the version of scarcity that respects the clock, andDead Space (2023) for the industrial-horror sound design Signalis is clearly listening to.

Spoilers below

The Replika question is the engine, and it’s more carefully built than the fragmentary telling suggests. Elster is a mass-produced unit; the Nation stamps out identical bodies with identical memories; and the game’s horror is that identity here is a manufacturing tolerance. When you meet another LSTR, the correct reaction isn’t fear of a monster. It’s the much worse realisation that the monster and you are the same product with different wear.

The Ariane material is where the poem earns its structure. The repetitions across the game — the same room reassembled, the same photograph degrading, the same promise resurfacing in a different room’s handwriting — read as glitchiness on a first pass and asiteration on a second. You’re being shown drafts. The story has been run before, and what you’re playing is one attempt among many, which retroactively makes the respawning enemies feel less like a difficulty mechanic and more like a thesis.

The endings sort by hidden metrics — how much damage you took, how often you fiddled with the radio, whether you kept certain objects — and I think this is the one place the design overreaches. Signalis is asking you to earn a reading of its text by behaviours it never taught you were text. The Promise ending is the one the game clearly loves, and the one hardest to arrive at by instinct, which means most players get their interpretation assigned rather than chosen. For a game this deliberate about constraint, that’s the one constraint that fires backwards.

There is a particular kind of tiredness that comes from modern puzzle games. It isn’t the puzzles. It’s thedistance between them — the walk back to the thing you already solved, the second cutscene explaining the lever, the collectible you’re meant to want. Most puzzle games are a good idea wrapped in forty minutes of administration.

Cocoon has none of that on it. Geometric Interactive’s debut, released on 29 September 2023 across PC, PlayStation, Xbox and Switch under Annapurna Interactive, runs somewhere around five hours and spends all five of them doing the one thing it exists to do. I have played puzzle games since the C64 was the only machine in the house, and I can count on one hand the ones this disciplined.

The idea, stated once

You are a small winged thing. You walk. You have one action button. You can pick up a glowing orb, and it sits on your back like a rucksack.

Each orb is a world. Set it on a pad and step in, and you are inside it — a whole biome, with its own colour, its own creatures, its own puzzles. Step onto another pad and you are spat back out, standing next to the orb you were just living in.

That’s the sentence. Everything Cocoon does for five hours is a consequence of it.

Because an orb is a worldand an object, you can carry a world into another world. You can be inside orb A, holding orb B, and set B down and enter it — you are now two worlds deep. Later you’ll be carrying a world that contains the world you need to be in, and you will have to think about what “inside” means for about ninety seconds before it clicks into place.

And because an orb is an object, it can also be atool. Each one lends a power to whoever’s carrying it. So an orb is simultaneously a place you can go, a key you can hold, and a thing that has to be somewhere for a puzzle in a different place to work. Three roles, one noun, zero explanation.

Teaching with the level instead of the text

Cocoon has no dialogue. No text. No tutorial pop-up telling you to press the button. No hint system, no HUD, no map, no journal.

This is easy to describe as minimalism and get wrong. Minimalism is what it looks like. What it actually is isteaching — every mechanic gets introduced in a room where only one thing can happen, in a place where the wrong answer is visibly the wrong answer. You learn the orb-as-tool idea in a corridor where the tool is the only variable. You learn nesting in a chamber that gives you two orbs and a very short leash. By the time the game combines the ideas, you’ve already been quietly examined and passed.

Jeppe Carlsen was the lead gameplay designer on Limbo and Inside at Playdead, and this is the same craft with the safety off. Limbo taught you by killing you; the rooms were short, the deaths were instructive, and the restart was instant. Cocoon barely kills you at all, and it doesn’t need to. It teaches bygeometry — a ledge you can see and can’t reach, a pad that’s clearly a pad, a door that clearly wants something. The lesson arrives through your eyes before it arrives through your hands.

That’s an old skill and a rare one. The real ancestor here is Geoff Crammond’sThe Sentinel (1986), which I played on a C64 in a room with the curtains shut and no idea what I was doing for the first hour. The Sentinel had a handful of verbs and no story at all, and it taught you its entire logic by putting you on a hill and letting you look. Cocoon has that same faith: that a player looking at a well-built space will work it out, and that working it out is the whole product.

Portal (2007) is the other obvious relative, and Cocoon is more austere than Portal — no voice in your ear, no jokes, no character to be charmed by. What it keeps is Portal’s teaching curve, where the tutorial and the game are the same object and you can’t see the seam.

The loop, and why the loop holds

Here’s the mechanical thing I keep turning over.

Most puzzle games have a currency problem. They need to hold back abilities so the difficulty can rise, so they lock things behind doors, gate them behind progress, hand them to you on a schedule. The gating becomes the pacing, and you can feel the designer’s hand on the tap.

Cocoon solves this by making the abilityportable and physical. You don’t unlock the power — you’recarrying it, and you can put it down. So the difficulty rises through logistics rather than through permissions. The question stops being “do I have the tool” and becomes “where does the tool have to be standing while I’m somewhere else”. The game never has to take anything away from you. It only has to build a room where you need to be in two places at once.

That’s a beautiful bit of economy, and it has a downstream effect the whole game rides on:backtracking stops being punishment. In a Metroidvania, going back is a tax you pay for the design’s shape. In Cocoon, going backis the puzzle — the return trip is the move, the carrying is the thinking. Walking is never dead time, because the thing on your back is the reason you’re walking.

The other place the fat gets trimmed is failure. Boss encounters are pattern-reading with instant retries; there is no resource attrition, no inventory management, no lives system left over from an arcade that closed thirty years ago. When you die, you are back a few seconds later, and the game has correctly assumed that your punishment was already the fact that you were wrong.

Where it costs itself something

I’d be a bad witness if I said the discipline is free.

Cocoon is a game with almost no friction, and friction is where a lot of people findfeeling. There’s nothing here to be furious at, nothing to grind against, nobody to like. The world is gorgeous — Erwin Kho’s biomechanical architecture is genuinely strange, all chitin and wet metal and machines that look grown rather than built — and it stays at arm’s length. You will not carry a character out of this game, because there isn’t one.

The difficulty sits in a narrow band, too. The puzzles areclean — nearly every one lands with the small satisfying click of a well-made lid — and clean means the ceiling stays low. Nothing in Cocoon will hold you for an hour the way a late Obra Dinn deduction will, or leave you filling a physical notebook. If your favourite feeling in a puzzle game is being genuinely stuck and slightly insulted by it, Cocoon will feel like it’s letting you off.

And the ending is where the restraint runs out of road. A game this wordless has to land its close on shape and sound alone, and Cocoon’s finish is more of a chord than a sentence. It resolves. Whether itmeans anything is between you and the credits.

The verdict

Cocoon is the best-engineered puzzle game since Return of the Obra Dinn, and it wins on completely different ground. Obra Dinn is a mountain you climb with a pencil. Cocoon is a machine with every unnecessary part removed, and the pleasure of it is watching a designer refuse — over and over, for five hours — to pad his own work.

That refusal is worth more than it sounds. This is a game that could have been twelve hours. It could have had a collectible chime and a lore codex and a second act where you revisit the first orb with a new hat. All of that was available, and all of it would have sold, and Carlsen threw the lot away. What’s left is five hours where every single minute is the good part.

Take the afternoon. It only wants one.

What to play next:Chants of Sennaar, which came out the same month and teaches you a language with no lectures either;Tunic, for the other great modern experiment in telling the player nothing; andReturn of the Obra Dinn if you want the version that fights back.

Spoilers below

The nesting reaches its proper depth in the back half, when you’re moving orbs between worlds that are themselves sitting inside orbs, and the game asks you to hold a mental stack three deep. What’s remarkable is that Cocoon never renders this as a diagram. There’s no map of the nesting, no visual aid, no “you are here” — the structure lives entirely in your head, and the game trusts that it will fit there. It does, mostly, and the two or three moments where itdoesn’t quite fit are the most alive the game gets, because you have to stop and physically reason about containment.

The bosses are the one concession to convention, and they’re better than they need to be. Each one is built out of the orb power you’ve just acquired, so the fight functions as an exam on the mechanic — read the tell, apply the verb, repeat with a variation. It’s the Limbo idea again: the encounter is a puzzle wearing a monster costume, and it dies when you’ve understood it rather than when you’ve out-twitched it.

The final stretch collapses the orbs together, and it’s the one time Cocoon reaches for grandeur. It’s a fine ending and a slightly hollow one — the game has spent five hours teaching you that everything means something mechanically, and then closes on something that means somethingthematically, in a register it never taught you to read. I’d have taken one more puzzle over the awe. That’s a small complaint about a game that got almost everything else exactly right.