Learn quant models through their failures.
Build reinforcement learning, Markowitz, Black–Scholes, and regression in small Python steps, then break one assumption to see what the model misses and why its output can fail.
Start with a picture and a hand calculation. Then run short Python — one formula at a time — and see exactly where the model breaks.
CriticalQuant method
Model failure is not a bonus module. It is where understanding clicks.
Textbook formulas look safe until their assumptions stay invisible. Each lab moves from intuition to a checkable calculation, then to small Python that mirrors the formula — and finally to the single assumption that makes the output lie.
See what the formula means
Story, diagram, and tiny numbers. Every symbol is named before you type code.
Build it in small Python steps
Run one line at a time — covariance, inversion, weights — so you watch the math happen, not memorize it.
Break one assumption
Push a correlation toward perfect, hedge in real steps instead of continuously, or keep only the data that passed a selection filter. Weights explode, signs flip, or labels change — immediately.
Finish with contrast, limits of the setup, and a transfer question without code — so you own the trap, not the syntax.
Launch labs
Learn the model by finding its breaking point.
Not a generic “learn Python” course. Each lab makes a quant model intuitive, builds it step by step, and reveals the assumption that makes its output fail.
Q-Learning & Deep Hedging
Why delta hedging bleeds cash under transaction friction. Code tabular Q-learning, discover No-Transaction Bands, and build J.P. Morgan's Deep Hedging neural policy and CVaR risk engine from scratch.
Why Markowitz weights explode
Expected return and portfolio variance by hand, then in code; see how ρ → 1 and tiny estimate errors create extreme long/short weights.
Where Black–Scholes stops working
Payoff, d₁/d₂, and put–call parity built stepwise; then continuous hedging, constant vol, jumps, and costs as visible limits.
Why regression signs flip
OLS in small Python steps, control the confounder, then see collider selection invent a relationship.
Kelly sizing & ruin
A coin wins 55% of tosses — a genuinely good bet. Watch compounding turn "bet more" into near-certain ruin, find the 10% Kelly stake yourself, and learn why professionals bet only half of it.
Build the next lab with us
Which model should fail next?
Tell us what confused you, what should be clearer, or which financial model you want to build and break next.
Feedback is voluntary. It needs no account or email address; only the answers you choose to send are stored to improve future labs.
Feedback & lab wishes
Choose a lab request, describe a confusing moment, or suggest a model failure worth making visible.