Foundations track 🧭 · six sessions, no code
For anyone who prices anything, argues for a budget, or reads a number a model produced. Prices as information, incentives and elasticity, marginal thinking and opportunity cost, when markets fail, money, inflation and interest, and games and moats. The marketplace's real counts appear on every page; the code does not.
Prices are information
Supply and demand as a description of where a price settles, not who sets it. A price as a message carrying knowledge nobody holds in full. Along the curve, or the curve moved. Of 32,216 real products, 203 moved price often enough to say anything about response.
📏 Foundations session 2Incentives and elasticity
How much people respond. Elastic versus inelastic, the revenue rule, who really pays a tax, and the Olist number read as an elasticity: a range, -1.5 to -2.2, and why a range is the honest shape.
⚖️ Foundations session 3Marginal thinking and opportunity cost
Decisions at the margin, sunk costs and the fork, the next best use, comparative advantage, and why a physical unit, a copy of software and one more prediction have three different marginal costs.
🍋 Foundations session 4When markets fail
Externalities, public goods, the market for lemons, adverse selection and moral hazard, and the metric that became a target. Applied to a data purchase where the quality cannot be seen before paying.
🕰️ Foundations session 5Money, inflation and interest
Money as three jobs, inflation as a change in the price of money, real versus nominal, interest as the price of time, discounting, and why measured productivity lags a technology.
♟️ Foundations session 6Games and moats
Strategic interaction and the prisoner's dilemma, network effects as a claim with a published rebuttal, switching costs, and when a data advantage compounds and when it does not.
Data and AI deep-dive track 🔬 · ten sessions, Python
For the people who compute the number and have to defend it. From 112,650 real order items to a panel, five ways to fit the line and why four of them are wrong for a pricing decision, then the economics of predictions, networks, auctions, labour, information and platforms, and a bench that refits everything live.
Demand from data
One row is one unit at one price. The three decisions, the panel, the three sieves that leave 203 of 32,216 products, the log-log fit, and why one product's slope of -10 is not an elasticity.
🪜 Deep-dive session 2Why the naive number lies
Same rows, five fits: +0.008 across products, -0.108 pooled, -1.788 within product, -1.854 with months, and a placebo at zero. What a fixed effect removes, and what nothing removes.
💳 Deep-dive session 3Pricing with the number
The revenue rule with the real slope: +8.9 percent or -7.2 percent for the same 10 percent rise. Price discrimination in three degrees, versioning, and freemium as a sorting menu.
🧮 Deep-dive session 4The economics of a prediction
Fixed versus marginal cost of inference, per-token pricing as pure marginal cost, utilisation, and the break-even volume between renting a model and owning one.
🕸️ Deep-dive session 5Network effects and data moats
Direct and indirect effects, Metcalfe against n log n, two-sided platforms, and when more data compounds and when the learning curve flattens.
🔨 Deep-dive session 6Auctions: how ads and compute are priced
Why auctions exist, second-price and the dominant strategy of telling the truth, the generalised second-price auction that is not truthful, and spot markets for compute.
🧑🔧 Deep-dive session 7Labour economics of AI
Tasks versus jobs, displacement and reinstatement, the three task-level studies with their exact figures, and what a task study cannot say about employment.
🔍 Deep-dive session 8Information economics of models
Lemons in data and model markets, benchmarks and evals as signals that decay, principal and agent with a metric between them, and the measure that became a target.
🏗️ Deep-dive session 9Build, buy or rent a data platform
Make-or-buy as transaction costs, total cost over a horizon, the option value of renting, and the switching-cost line the vendor did not draw.
🎛️ Deep-dive session 10 · bench nightThe elasticity bench, and defending a number
2,216 real rows, five rungs, every slope computed in your browser. The anti-lever that prices on the market-wide number and gets the sign wrong, the break button, and the sentence for the meeting. Final scorecard.
Your passport 🎫
Clear all three quiz questions in a session and it gets stamped. Sixteen stamps means you can read a price as a message, say what an elasticity was compared with, and defend a number with its range, its mechanism and its caveat. Stamps live in this browser only.
What this is not, so you land in the right place 🧭
| If your question is | Go here instead |
|---|---|
| How do I read a P&L, a cash flow statement, or size a funding gap? | Finance: Reading the Numbers ↗ |
| How do I prove the price caused the change, not just moved with it? | Causal Inference ↗ |
| How do I design the price test that settles it? | Experimentation ↗ |
| How do I choose the best plan under constraints, and what is a constraint worth? | Prescriptive Analytics ↗ |
| What is a customer worth, and what did it cost to get them? | Customer Retention ↗ |
| I price things, or I compute numbers about prices, and I want to understand how people respond and defend what I find. | you are in the right place |
Economics as a data habit. The mechanisms are standard; the numbers are computed from a real marketplace for these pages; and where the literature disagrees, the page teaches the disagreement rather than picking a side.
What is honest here 🧾
The numeric spine is real public data, recomputed for these pages and refitted live on the bench. Where a figure comes from a paper, the page names it and says whether the paper was read at source; where nothing citable exists, the page says that instead.
| Claim | Status |
|---|---|
| Every Olist count and every slope: 32,216 to 203, the five rungs, the placebo, the category numbers | Really computed from the Brazilian E-Commerce Public Dataset by Olist (Kaggle, CC BY-NC-SA 4.0), delivered orders before 1 September 2018; the bench recomputes the deterministic rungs in the browser to three decimals of the offline build |
| The revenue projections for a price change | Modelled under a constant-elasticity assumption, labelled so on every page that shows one |
| Whether the Olist slope is causal | Not claimed. An association with a placebo behind it; product effects, month effects and the placebo remove three things, and none removes "the seller chose the price" |
| Published elasticities (gasoline, food) and the three AI task studies | Reported from named papers, with the exact figure and whether the paper was read at source or through its abstract |
| Inference prices | One vendor's list prices as published on 25 September 2026, three models named, with "re-check before you decide"; no GPU price anywhere |
| Beira Goods, the inference workload, the auction bidders, the learning curve, the TCO table | Constructed, and the page says so each time |
| Whether AI raises or lowers employment | Not claimed. The task studies measure tasks in experiments, not jobs in an economy; the page teaches the disagreement |
The knowledge map 🧠
Both tracks at a glance - hover a session to spotlight its concepts, click any node to jump in.