Mission control
Choose a learning mode, difficulty, level, and fictional operating objective.
Responsible-use rules
- Every applicant and outcome is fictional.
- Predictions must not support real lending decisions.
- Real systems require legal, ethical, security, and expert review.
- Fairness cannot be guaranteed by removing one attribute.
How scoring works
Interactive application flow
Fabric.js renders the current fictional application, preparation pipeline, probability meter, and decision outcome.
Fictional dataset builder
Generate, edit, import, clean, and inspect applicant records.
Data preparation
Feature selection and engineering
Choose model inputs. Sensitive fictional grouping stays available for fairness auditing but is excluded from training by default.
Applicant records
No dataset generatedClassifier training lab
Train an actual browser-based educational classifier using the selected fictional attributes.
Training log
Evaluation and threshold challenge
Inspect metrics, confusion patterns, calibration, feature influence, and individual decisions.
Explainable prediction
Error analysis detective
| ID | Actual | Prediction | Probability | Pattern |
|---|
Model comparison tournament
| Model | Accuracy | Precision | Recall | F1 | AUC | Fairness gap | Complexity | Recommendation |
|---|
Fairness and bias lab
Audit fictional groups, test mitigation options, and learn why proxy variables can preserve unfair patterns.
Audit notes
Responsible checks
Production and concept-drift simulation
Change the fictional applicant population, monitor decay, and choose a responsible response.
Response decision
Monitoring alerts
| Period | Metric | Value | Status | Suggested action |
|---|
Final game report
Review performance, fairness, efficiency, explanations, and deployment readiness.