Build a mushroom. Ask the model.

Describe what you see. Anything you leave out stays a wireframe.

Drag to turn

Solid parts are what you observed.

Wireframe parts are unknown, and the model treats them as unknown too. Press Predict when you are ready.

Scanning

Asking the model.

Probability that the mushroom is poisonous

0%

chance it is poisonous

About this answer

Inputs changed since this answer. Predict again to update it.

Never decide whether to eat a mushroom with this tool.

It is a student model trained on a deliberately noisy dataset, where about 2% of the labels are flipped. It is tuned to catch 90% of poisonous mushrooms, and pays for that by flagging roughly two thirds of the edible ones as poisonous too.

API settings

Leave the address empty on the live site: it forwards /api to the backend itself. Fill it in only to point this page at another API, for example http://localhost:8000 during development. Stored in this browser only.