Perfect Score — the greatest games of all time, taken apart into ten ingredients and rebuilt on a budget

FEEL

The half-second loop: input, animation, impact, recovery. Feel is why one platformer is beloved and its clone is unplayable with an identical design document. It is the only ingredient a player judges before they have understood anything.

100 means: A jump that people describe as "perfect" without being able to say why.

Floor across the canon: 30

SYSTEMS

Depth, interaction, emergence. High SYSTEMS means the rules produce situations the designers never placed by hand — the barrel, the guard and the slope combining into a story nobody wrote.

100 means: Players routinely do things the developers have to be shown.

Floor across the canon: 40

WORLD

Coherence, atmosphere, environmental storytelling. WORLD is high when the geography is an argument: when the shortcut that unlocks back to the bonfire tells you more about the design than any cutscene could.

100 means: People draw the map from memory years later.

Floor across the canon: 5

STORY

Written narrative, character, theme, dialogue. Deliberately not the same as emergent story: the tale of your Sims family is SYSTEMS, not STORY. This axis is what somebody sat down and wrote.

100 means: The writing would survive being moved to another medium.

Floor across the canon: 0

ART

Art direction. Not polygon count, not resolution, not budget. A 1997 sprite game and a 2018 photogrammetry showcase can both sit at the top of this axis, and one of them ages better.

100 means: A single frame is recognisable at thumbnail size.

Floor across the canon: 30

SOUND

Music, sound design, and the audio that does mechanical work — the sound that tells you the parry landed before the animation has finished. The most under-argued ingredient in every list ever written.

100 means: The soundtrack outlives the console it shipped on.

Floor across the canon: 60

PACING

Structure and tempo: how fast it teaches, when it escalates, whether it knows to end. High PACING is ruthless — it is mostly a record of what got cut.

100 means: Nothing in it overstays, including the ending.

Floor across the canon: 40

MASTERY

The distance between playing it and being good at it, and whether that distance is legible. High MASTERY games are still being learned decades later, usually by people who were not born when they shipped.

100 means: A competitive scene exists that the developers did not fund.

Floor across the canon: 24

SURPRISE

Invention, at the moment it arrived. This axis is historical and it is unfair: doing it second is worth almost nothing here, no matter how much better the second one is.

100 means: Everything after it is described using its name.

Floor across the canon: 62

RETURN

Endurance. Whether there is a reason to open it again once you know how it goes. Content runs out; rules do not, which is why the two highest scores on this axis are a puzzle game from 1984 and a box of cards.

100 means: People are still playing it after the servers are gone.

Floor across the canon: 20

Five terms, all visible while you drag. Mean is how much game there is. Peak pays for the distance between your best ingredient and your average one. Synergy and Conflict apply to pairs, and only once both are above 60 — two mediocre things do not add up to a good one. Deficit charges for going below the lowest level anything on the all-time list shipped with.

The contentious term is Peak. It says a game that is 70 at everything is worse than one that is 95 at a single thing and 45 elsewhere, at identical cost — that the round build is the failure mode rather than the safe option. That is a claim about how greatness works, and the evidence offered for it is that none of the 42 games here is round.

They are an argument, written by hand. Nothing here is scraped or averaged from review aggregates — an average of other people’s opinions is still an opinion, and putting two decimal places on it would only disguise that. What the decomposition buys you is a precise disagreement: not "you rated it too low", but "SOUND is not a 66 in that game, and here is the track that proves it".

There is not one, and the chart is why. Give a hill-climber a budget and it finds a peak; give it a different starting point and it finds a different peak of almost exactly the same height. Give it every point there is — 100 on all ten ingredients, the literally perfect game — and the score collapses, because a game that is maximally story-driven and maximally emergent is at war with itself. Greatness is a set of incompatible local maxima, not a summit.

Because the good directions are occupied. The match is a correlation on the shape of a build rather than its size — each build is centred on its own mean first, so what is being compared is which ingredients it favours over its own average, not how much it could afford. Plain similarity would have put every pair above 95%, which is a decoration rather than a match. At 92% or above the page says it already exists, and it means it: across the whole canon the median game correlates with its own second-nearest neighbour at about 88.

A perfect game is 1000 — 100 on every ingredient. You get half. Everything on the all-time list cost more: the cheapest here is 637 and the most expensive is 862. Discovering that by loading a masterpiece into the forge and watching it blow through the budget is the fastest way to understand what those games actually were.

The lowest any of the 42 games goes on each ingredient, taken from the data rather than decided. It is the most interesting number on the page: STORY has a floor of 0 — five games here have no story at all and are still canonised — while SOUND has a floor of 60. There is no such thing as a beloved game with bad sound. Together the floors cost 311 of your 500 points before anything has been made good.

Probably. The model is opinionated on purpose and it has a favourite failure: it puts Deus Ex last, because Deus Ex wants deep systems and a written story and endless replay at once, and every one of those pairs is taxed. Whether that is the model catching something real about ambition or simply being unfair to the most ambitious game on the list is the argument worth having, and the numbers are all on screen so it can be had precisely.

No. It is an unaffiliated design study. No art, audio, code or data from any game appears here — every cover on the page is drawn at runtime from the ten numbers underneath it, which is why two identical builds produce an identical cover. The games are named and described in order to argue about their design, which is what criticism is.

It is not a ranking. The list at the top is what one opinionated model does with forty-two games somebody else canonised. If it disagrees with you, the numbers are all on screen and the argument is available.

It is not measured. The only measured numbers on this page are the counters under THE CROWD, and even those are soft — the endpoint is unauthenticated, because the people worth counting have no account and never will.

It is not affiliated with anything. No art, audio, code or data from any game is used here. Every cover is arithmetic.