Game settings
Choose the challenge, dataset conditions, scoring rules, and learning support.
Live scorecard
Ready. Start a new game to begin matching.
Interactive matching board
Drag a metric card into the answer zone, or use the accessible buttons beside the board.
Keyboard answer list
Select one metric, then submit your match.
Explanation and feedback
See the reasoning, limitation, and companion metric.
The explanation will appear after you submit an answer.
Metric comparison lab
Explore thresholds, imbalance, outliers, performance trends, and category mastery.
Results dashboard
Review mistakes, response times, confusion pairs, and revision priorities.
| Round | Scenario | Your match | Best match | Result | Time |
|---|---|---|---|---|---|
| No completed rounds yet. | |||||
Mastery by category
Your strongest areas and recommended revision topics.
Most confused pairs
Metric reference guide
Search definitions, formulas, suitable tasks, limitations, and companion metrics.
| Metric | Category | Best for | Direction | Key limitation | Companion |
|---|
How to play
Build practical metric-selection judgement through short scenarios.
Correct primary matches earn full points. Suitable secondary metrics can earn partial credit when enabled. Fast answers, streaks, and unused hints add bonuses.
Accuracy can mislead on highly imbalanced data. RMSE increases the influence of large regression errors. PR-AUC often gives more useful information when positive examples are rare.
Frequently asked questions
Common questions about metrics and game behaviour.
Why is accuracy sometimes a poor choice?
A model can achieve high accuracy by predicting only the majority class. Balanced accuracy, F1, PR-AUC, or MCC can reveal failure on rare classes.
When should I prefer MAE over RMSE?
Prefer MAE when each error should contribute proportionally and robustness matters. Prefer RMSE when large errors should receive stronger penalties.
What is the difference between ROC-AUC and PR-AUC?
ROC-AUC compares true-positive and false-positive rates. PR-AUC focuses on precision and recall, making it informative for rare positive classes.
Does a higher metric always mean better performance?
No. Scores such as accuracy, F1, AUC, and R-squared are usually maximised. Error and loss metrics such as MAE, RMSE, and log loss are minimised.
Can more than one metric be correct?
Yes. Real projects usually need a primary metric and supporting diagnostics. The game awards partial credit for defensible secondary choices.
How does adaptive difficulty work?
Strong performance introduces less familiar metrics and closer distractors. Repeated mistakes trigger simpler scenarios and targeted review.
Is progress saved?
Progress, preferences, mastery data, and recent results can be stored in your browser when autosave is enabled.
Can I use the game without dragging?
Yes. Every canvas card also appears as a keyboard-accessible button. Select a metric and press Submit match.
What does the threshold chart demonstrate?
It shows how precision, recall, F1, and accuracy change as the classification threshold moves. The best threshold depends on costs and objectives.