Gameloom
    Sign InGet Started for Free
    LiveGameloom for Unity is live on the Unity Asset Store. Validate a concept inside the Editor, free.See what it does
    Why Gameloom

    You are about to spend a year of your life on a game idea, and the useful second opinion is not the one that agrees with you. So this one is built to disagree, and to show its working when it does.

    Score your concept See how the scoring works
    We don't just collect signals, we process them
    1. 01Live CaptureEvery day we capture the market signals that cannot be back-filled later.
    2. 02Decompose GameDNAEvery game we track is broken down into the design facts that form its GameDNA.
    3. 03Process SignalsYour concept lands among its nearest games, and their outcomes, player sentiment and what lasted are read into the score.
    4. 04Publish ScoreOne score, the same shape every time, with the evidence still attached.
    GameDNA

    We Don't Collect Games.
    We Take Them Apart.

    A GameDNA is a structured genome: what a game is, how it plays, who it is for, how it earns. Every game we track has one, and so does your concept the moment you describe it. Almost everything else on this page is downstream of that one decision.

    What that buys you

    • 01

      Neighbours by design, not by genre tag

      A novel concept still has coordinates, so it can be placed among real games even with no obvious rivals.

    • 02

      “Games like yours” stops being an opinion

      It is the set nearest your genome on your platform, and it is what the market half of your score reads from.

    • 03

      A comparable can be wrong, and you find out

      If a game you named sits nowhere near yours once both are decomposed, you find out before you build towards it.

    And once a game has shipped, we hold three versions of it at once

    Claimed

    What the studio says it is

    A co-op deckbuilding trading game with deep runs

    Store page, tags, description, trailer

    Perceived

    What players said it turned out to be

    A great launch week with nothing to come back for

    The full timestamped review stream

    Rewarded

    What the market actually paid for

    Sold hard for three weeks, then stopped

    Outcome, sentiment and revenue signals

    How It Stays Honest

    The Arithmetic Decides.
    Language Only Narrates.

    Let a model reason freely over raw signal and you get answers that are fluent, confident and wrong. That is the one thing we could not ship. So the two jobs are split, and they never swap.

    The statistical layer

    Produces the claim

    Computes every score from the evidence actually present, sets the band it may ship in, and drops any market input the neighbourhood cannot support.

    The language layer

    Narrates what already passed

    Writes the sentence around a number it did not choose, at temperature zero, and is discarded whole if it says anything the evidence does not allow.

    A claim may never climb higher than the evidence under it allows. That rule runs in the arithmetic, before a sentence exists.

    So every read carries the band it was earned in
    CitedDense, validated neighbourhood
    ComparativeResembles known games
    Not enough signalToo few comparables

    Sales stay ranges rather than facts, because nobody publishes them. A thin neighbourhood is told it is thin rather than quietly averaged over. And the band travels with the finding, so you always know which one you are reading.

    What You Get

    Five Pillars. One Score. Every Point Cited.

    Not a paragraph of advice. One object, broken into the same five questions every time, with the evidence behind each one attached to it. Closer to a credit score than to a chat.

    Market Opportunity: 15.1 of 62 pointsCore Concept: 25.2 of 62 pointsEconomy: 7.1 of 62 pointsExecution: 12.1 of 62 pointsScalability: 2.5 of 62 points
    62/100Score
    Optimize and Validate
    Evidence completeness 80%
    Market OpportunityDo people want it?60/100
    PresentPillar weight 25%
    Core ConceptIs it a cool idea?100/100
    StrongPillar weight 25%
    EconomyWill it make money?35/100
    ThinPillar weight 20%
    ExecutionCan you build it?60/100
    PresentPillar weight 20%
    ScalabilityWill it last?25/100
    ThinPillar weight 10%

    Some of it is answered for you

    You supply most of the evidence, and it is graded on how good it is. The rest is read off what actually happened to games like yours, and those readings land in the number either way. A few of them:

    • Past OutcomesEntirely market-answeredWhat happened to games like yours?What happened to the games nearest yours: how the best of them did, how many had a rough time after launch, and what their reviews looked like at scale.
    • Similar GamesIn part market-answeredWho is your competition?We compare GameDNA, not titles. So the games you name get checked for whether they genuinely sit in your neighbourhood, and anything else that does is surfaced alongside them.
    • HookIn part market-answeredWhat's your hook?How far your pitch sits from the middle of your neighbourhood. It measures how different you sound, which is not the same as how differently you play, and it says so.
    Against a General Model

    Both Sound Confident.
    Only One Is Still True Tomorrow.

    Ask a model why a game failed and you get a story: plausible, well written, and different next week. Causation here is too tangled to recover from text alone.

    A general modelGameloom
    Ask it twiceA general modelTwo different answersGameloomThe same number
    When it cannot tellA general modelAnswers anywayGameloomSays so, and names what is missing
    What it compared you toA general modelNothing you can checkGameloomReal games, and what happened to each
    The weightsA general modelNone to inspectGameloomPublished, and yours to change
    Six months laterA general modelGone with the chatGameloomStill there, and re-runnable
    Open-ended brainstormingA general modelGenuinely betterGameloomNot what this is for
    Questions

    The Ones That Keep Coming Up.

    The argument is above. These are the objections it gets. For the method itself, see how the scoring works.

    Why not just ask ChatGPT to validate my game idea?
    Because it will always answer. Ask any model why a game failed and you get a plausible, well-written story, and a different one tomorrow. Gameloom computes the comparison first, against real shipped games with the same GameDNA, and only then lets language describe what the arithmetic already established. Ask twice and you get the same number, with the reasons still attached.
    Does Gameloom use a language model to produce the score?
    Yes, for one narrow job. Three calls per score, at temperature zero, and between them they read nine of the inputs behind a number, all of them your own text. A model never computes the score, never sets a weight, never decides what applies to your game, and cannot rate a blank field as evidence. If a call comes back malformed, or outside the band its own evidence rating allows, the whole call is discarded and a deterministic rule takes over.
    What is GameDNA?
    A structured genome for every shipped game in the corpus: genre and sub-genre, play modes, tags, how it earns, and a description, turned into coordinates. It is held in three versions at once: what the studio claims the game is, what players actually experienced, and what the market rewarded. Your concept gets the same treatment, which is how we find the games nearest to it in design space rather than by matching names. It also means a genuinely novel concept can still be placed, because it has coordinates even when it has no obvious rivals.
    What happens when there are not enough comparable games?
    The market read switches off. Below 8 platform-matched neighbours its contribution drops to zero weight rather than being reported as a weaker signal, and the score is built from your own evidence alone. A four-game neighbourhood producing full-strength market findings is the failure this gate exists to prevent.
    What if there is too little of my concept to judge?
    Then you do not get a verdict. Under 40% of it answered the score is withheld entirely: you get "Not enough information" and a checklist naming the specific inputs that are missing. A score computed over four answered questions is not a cautious score, it is a made-up one, and a blank scores zero rather than a polite average so that leaving one is never the cheaper option.
    How do you know the scores are right?
    We hold a fixed reference set of design documents and re-score every one of them before any rubric change ships. If the distribution moves in a way we cannot explain, the change does not ship. That is also how we caught our own band boundaries drifting, twice.
    Get started

    A paragraph is enough to start. Score it free, see what the market did to games like it, then strengthen it in Gameloom and carry the evidence through to launch.

    Get Started for Free