The data lied.
Your parents paid the price.
Now you read it better.
You are Epoch, a student building an AI named Entropy to clear your parents' names. Each chapter drops you into a real data mystery — from a rigged school reading contest to a conspiracy that frames the innocent — and asks you to clean datasets, train models, audit algorithms, and decide what evidence is strong enough to publish. You are not memorizing concepts; you are using them under pressure to protect people the data failed.
What is Epoch & Entropy?
It is a story-driven data-science strategy game. You play as Epoch, a teenager who discovers that the same data systems used to convict their parents are being sold to schools, cities, and courts. With your self-built AI partner Entropy, you investigate cases, challenge flawed analyses, and learn when to trust the model — and when to trust your own judgment.
A look inside Epoch & Entropy
Entropy runs alone. Until it doesn't.
Watch a live demo of Entropy solving a case without a human partner — and see exactly where an overconfident AI goes wrong.
Three things make this a game
Not a course. Not a visual novel. A real strategy game where the data is the battlefield.
An AI partner you cannot trust
Entropy runs analyses in seconds, but it is literal and overconfident. It will suggest shortcuts, spot spurious patterns, and push you to publish before you have checked the data. Your job is to supply the judgment it lacks: verify assumptions, question labels, and decide when the evidence is strong enough.
Cases with real stakes
Each case is a self-contained data puzzle with real consequences: a bus route cut by a biased survey, a hiring algorithm that proxies race through zip code, a vaccine shipment spoiled by bad sensors. Your analysis decides who gets hurt and who gets saved.
Skills that transfer
You will use real tools and concepts — train/test splits, confusion matrices, ROC/AUC, SQL joins, and fairness audits — not as quizzes, but because the case demands them. The game builds intuition; your job is to know when to doubt the output.
What to expect
Every chapter blends narrative, strategy, and hands-on data work. Here is what you will actually do.
Real data workflows
- Import and inspect messy CSVs and tables.
- Clean missing values, duplicates, and bad joins.
- Write SQL queries and build train/test splits that hold up.
An AI partner with opinions
- Entropy runs models and suggests interpretations.
- It will sometimes bully you into conclusions.
- You choose to accept, override, or retrain — and live with the result.
Consequences that stick
- Publish sloppy findings and the wrong person loses a job, route, or trial.
- Get it right and the story unlocks new leads and allies.
- There is no reset button on a bad recommendation.
Skills you can use tomorrow
- Bias, overfitting, validation, and fairness audits.
- Neural nets, NLP, computer vision, and reinforcement learning.
- Every concept maps to a real decision you had to make.
The same models that framed them are sold to cities and courts.
Meridian Data Systems shapes the answers power wants to hear. To beat them, you have to understand their tools better than they do.
— Budo
Every case teaches one thing
Start small. End on the case that started everything.
The Reading Contest
A duplicated row threatens to give a sixth grader a trophy he did not earn. Learn to spot duplicates and check the primary source.
Duplicates · Source checkingThe Late Bus
The city wants to cut a bus route. The survey only asked people who drive. Learn sampling bias and who the data forgot.
Sampling bias · Selection effectsThe Credit Model
A model never sees race, yet it rejects one zip code. Learn proxy variables, disparate impact, and fairness audits.
Fairness · Proxy discriminationThe Cold Case
Turn every skill back on the verdict that put your parents away. Find the deleted transactions. Prove the frame.
Network analysis · Evidence synthesisWhat you will actually learn
Concepts you can use on any dataset, taught under pressure.
Supervised vs Unsupervised
When to label, when to let patterns emerge
Confusion Matrices
Accuracy is not enough
ROC / AUC
Trade-offs under uncertainty
Neural Architectures
CNN, RNN, LSTM, GRU
Overfitting
Train/test splits and validation
NLP
Sentiment, context, lazy labels
Computer Vision
Image forensics and synthetic media
Reinforcement Learning
Decisions that learn from feedback
From schoolyard mystery to federal case
The campaign escalates one honest analysis at a time.
Training Wheels
School cases
City Desk
Algorithms & buses
Public Trust
Hospitals & workplaces
Shadow Files
Fraud & terror
Global Scale
Pandemics & climate
Machine Learning
Models that frame
Cold Case
Exoneration
The Epoch effect
Stories start changing after you touch them.
Loading feed...
One honest analysis at a time.
Free to play. No ads. No premium. Just you, Entropy, and the data everyone else stopped reading.