Is It a Game Yet? Six Questions for a Vibe-Coded Game
Your project is a game when the player has a goal, does something again and again, can fail, sees at once whether a move worked, and the game can be won or lost. Many projects built with an AI stop one step short: the character moves around nicely, but nothing can go wrong and nothing ever ends. That is a toy.
What makes something a game?
The game researcher Jesper Juul compared seven older definitions in 2003 and named six features of a classic game: it has rules, a variable outcome, some outcomes count as better than others, the player has to make an effort, the player cares how it ends, and real-life consequences are optional. Our six questions build on that and turn it into something you can check in a game you are making. Feedback and a reason to come back are added from everyday game design. The point about real-life consequences is left out, because it is about stakes, not about building.
| Question | What it asks | It does not count when |
|---|---|---|
| Goal | Does the player know what to reach? | the goal is only in your head, not shown in the game |
| Action | What does the player do again and again? | the character only walks around |
| Resistance | What can go wrong, so the player can fail? | a timer never runs out, or enemies cannot hurt |
| Feedback | Does the player see at once whether a move worked? | points go up somewhere nobody looks |
| Outcome | Can the game be won and lost, or come to a result? | points pile up with nothing they lead to |
| Again (bonus) | Is there a reason for one more round? | it is a bonus: a game without it is still a game |
The first five are the core. The sixth is what makes people come back, but a game without it is still a game.
Toy or game? An example
Cloudbank Couriers, an invented balloon delivery game, has a goal (six parcels before dusk), an action (pick up, steer, drop), feedback (a bell and a tick on the list) and an outcome (a summary when all are delivered). But nothing can go wrong: the sun never sets, storms do nothing, and a parcel on the wrong roof just stays there. And while that was missing, an inventory of balloon parts was built.
So it is not a game yet: four of five core parts. Only the next step is worth building, and it is not the inventory: let the sun set, and let storms push the balloon.
Keep the scope in check
Put every feature in one of three places: core, later or cut. The rule is simple: nothing from later while the core is missing. An AI is happy to build whatever you ask for next, so the list has to say no for you.
Paste this to your AI
Check whether my project is a game yet. Go through my real code and answer these six questions:
1. Goal: does the player know what to reach?
2. Action: what does the player do again and again?
3. Resistance: what can go wrong, so that the player can fail?
4. Feedback: does the player see at once whether a move worked?
5. Outcome: can the game be won and lost, or come to a result?
6. Again (bonus): is there a reason for one more round?
For each, write one plain sentence, the file and function where it happens, and a short way for me to see it in the game. If you cannot point to a place in the code, write "missing". Do not count anything as there because it is planned.
Then draw the game's flow as a small Mermaid state diagram: where the player starts, what they do, how they win and how they lose. Show me any state the game can get stuck in, and any state from which it can never end.
Then list every feature as core, later or cut. Tell me if anything from later is built while the core is missing.
Finish with: how many of the five core parts are there, and the one thing to build next. Do not tell me whether it is fun: only players can tell that.
This text lets your AI judge on its own. Connected to Spielerando, our checker tests its game card and walks your game's loop, and your project page shows how many core parts are there, version by version.Connect your AI
What this cannot tell you
Whether it is fun. A game can have all six parts and still be dull. The questions tell you whether it has the parts of a game, so that a playtest is worth doing. Then let real people play, and measure where they quit, for example with a drop-off funnel.
How to check it worked
- Try every probe yourself. For each part, your AI gave you a way to see it. If you cannot see it, it is not there.
- Look for the place in the code. Open the file and function it names for one part.
- Ask what is missing first. The answer should be one thing from the core, not a new feature.