Drag the text card into a category zone, or use the category buttons. Keyboard players can press number keys, H for hints, and N for the next sample.
Arena setup
PracticeAccessibility
Score
0
Accuracy
0%
Streak
0
Lives / Time
3
Canvas classification arena
Round 0 of 15
Round feedback
Start the arena to receive a text sample.
Rival prediction
—
—
Accuracy by round
Confusion matrix
Per-class F1 score
Confidence distribution
Text preparation and training controls
Preprocessing
Training simulation
Adjust controls, then train the model.
Mistake analysis
Incorrect predictions will appear here.
Model tournament
Compare simulated accuracy, speed, confidence, and training cost.
| Model | Accuracy | Macro F1 | Speed | Cost | Rank |
|---|
Add custom sample
Import and export
CSV columns: text,label,type. Imported files remain in your browser.
Text classification guide
Binary classification chooses between two labels. Multi-class classification selects one label from many. Multi-label classification can assign several topics to one text.
Tokenisation splits text into usable units. TF-IDF emphasises informative terms. N-grams capture short phrases, while embeddings represent semantic similarity.
Precision asks how many predicted items were correct. Recall asks how many actual items were found. F1 balances precision and recall.
A well-calibrated model is correct about eighty percent of the time when it predicts eighty percent confidence. Thresholds can route uncertain cases to human review.
Inspect errors across language varieties and user groups. Protect private information. Keep humans involved where mistakes can seriously affect people.
Session history and achievements
Text RookieCompleted rounds will appear here.