# How Kiwari playtests your game

Most of building a game is checking that the thing you just made actually works. In Studio that means pressing Play, walking over, clicking the button, reading the Output, stopping, fixing, repeating. Kiwari runs that loop for you, inside your open place, and tells you what it saw.

## What a playtest looks like

You ask in chat, in plain words:

> Playtest the shop: walk to the green pad, buy the sword, and confirm the leaderstat goes up.

Kiwari starts Play mode itself, spawns in, and drives the game the way a player does. It presses keys, moves the mouse, walks the character, and clicks buttons. Every step is proven with evidence: a console line, a measured value, a capture. "I pressed E and the sword's damage went from 10 to 25" is a result. "It should work now" is not, and Kiwari is built not to say it.

When the run ends, Kiwari stops Play and cleans up its test helpers, so your place is left as it was. The playtest appears in chat as a grouped log you can expand, with a short sped-up clip of what happened.

## It checks what screenshots cannot show

After building or moving things in the world, Kiwari measures the map instead of eyeballing it:

- Parts stacked on the same footprint, the z-fighting shape that flickers as you move.
- A floor laid exactly on top of another surface.
- Anchored props touching nothing, floating off their walls.
- Parts reaching below the floor line.
- A light census: how many lights, how bright, and the ambient.

These are the bugs players spot in the first minute, and the ones source code never mentions. Geometry and light are exactly what a model cannot reason about from scripts, so Kiwari measures them directly.

## It only claims what it saw

A fix counts as verified only when Kiwari observed the result after the change: a printed check, a measured value, a clean world pass. When it cannot observe something from its side, like how dark the scene feels or whether the pacing is fun, it says so plainly and names what you should check yourself. If you report the same bug again, it treats your report as new evidence instead of defending its earlier fix.

## Getting good playtests

- **Ask for the goal, not the script.** "Confirm the buy button works" beats ten micro-steps.
- **Mention timing.** If a monster wakes up after 25 seconds, say so, so the run waits long enough to see it.
- **Say what state to set up.** "With the sword equipped, press E on the dummy" gets a real answer. The same key press without the setup proves nothing.
- **Keep one place open.** Playtests run in your open place, and Kiwari asks you to pick a window when several are open.

## What stays with you

Kiwari checks that things work. Whether they feel good, look right, or are tuned well is still your call, and it will tell you which parts it could not judge. That split is the point: you stop spending your session on the mechanical verify loop and spend it on the game.

[Start for free](https://kiwari.dev/download) and ask for a playtest on your own place.

Related: [How to playtest Roblox games with AI](https://kiwari.dev/blog/how-to-playtest-roblox-games-with-ai) · [How to get started with Kiwari](https://kiwari.dev/blog/how-to-get-started-with-kiwari) · [Review every change before it lands](https://kiwari.dev/blog/review-every-change)