Bias Detector Game

Score0

Fairness0%

Evidence0%

Streak0

Hints3

Time--:--

Mission briefing

Generate a mission to begin investigating.

Dataset audit Beginner

Mission setup

Choose a play mode, difficulty, and generated dataset settings.

Game mode

Difficulty

Dataset size 80

Group imbalance 35%

Bias severity 55%

Decision threshold 0.50

Missing-data rate 4%

Label noise 6%

Random seed

Group analysis

Investigation questions

Build an evidence-based conclusion instead of relying on overall accuracy.

1. Which group appears most affected?

2. What is the most likely bias type?

3. Which metric best exposes the problem?

4. Which first mitigation is most suitable?

Collect evidence, compare group results, and submit your audit.

Fairness dashboard

Compare rates, errors, calibration, and performance across fictional groups.

Overall accuracy: --

Dataset inspector

Sort records, filter groups, reveal missing values, and inspect possible proxy features.

0 rows

Mitigation laboratory

Apply a strategy, compare before-and-after metrics, and observe remaining trade-offs.

No mitigation applied yet.

Bias audit report

Generate an evidence summary with metrics, findings, mitigation, and limitations.

Complete an investigation to generate a report.

Mission checklist

Complete every investigation step.

Metric guide

Plain-language fairness definitions.

Achievements

Earn badges through careful analysis.

Local leaderboard

Scores remain inside this browser.

Investigator name

Learning notes

Important responsible-AI reminders.

Fairness is contextual.Different fairness definitions can conflict. The relevant choice depends on harms, goals, and affected people.

Accuracy can conceal harm.A model may look accurate overall while failing a smaller or underrepresented group.

Removing a field is insufficient.Other features may reproduce sensitive information through hidden proxy relationships.

Human review is not automatically fair.Review processes also need clear criteria, training, monitoring, and accountability.