LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
The lecture's open-source game, built in HTML, CSS, and JavaScript and hosted on GitHub Pages, gives you two seats in a tiny market. As a taker you buy or sell against a posted bid and ask, using three dice of which some faces stay hidden. As a market maker you post a middle price and a spread, then live with whoever hits your bid or your offer. Every 15 seconds a new supply, demand, and set of values appears, and the previous set settles. The point of the rounds that were played is the difference between a price edge and an inventory problem.
Why it matters
A market maker does not need a view that the hidden total will be high or low. The maker hopes to buy on the bid and sell on the ask, keep the two sides matched, and keep the spread. A taker does the opposite: trade only when the posted price is on the wrong side of the value implied by the dice you can see. The game makes both jobs short enough to repeat. It also shows the failure mode in one round. Quote a middle that nobody wants to sell into, and you own a one-sided position when the dice resolve.
The idea in plain words
Three dice produce a total. You see some of the faces and not others, so your estimate of the total is an expectation, not a known sum. Supply and demand are posted as prices you can trade against. There is no cap on size. The lecture treats size as a risk dial you set yourself, including steps of 10, rather than as part of the scoring rule. In one taker round the working estimate was 10 and the buy price was 10, so one unit was bought. The dice resolved at 6. The loss was 4 dollars. In the next round the offer was 13 and the bid was 12. Ten units bought at 12 resolved at 16, for a gain and loss of 40. A later estimate of 13.5 met a market posted at 18 on both sides, which the lecture called a clear sale. Another round showed 3 and 5 and was bought at 5. A round showing 3 and 7 was called an obvious buy.
How to use it
Play the taker seat until the arithmetic is boring. Estimate the hidden total from the faces you can see. Buy if your estimate is above the offer you must pay. Sell if it is below the bid you can hit. Do nothing when the posted prices already surround your estimate and the gap is smaller than the uncertainty you still have. The lecture wants later versions that condition the hidden faces on what you have seen, which would make that mental estimate harder. Those versions were a wish based on feedback, not a feature of the build being demonstrated. Then switch seats. Set the middle near your estimate and choose a spread wide enough that you are paid for being wrong, and narrow enough that both sides still trade. After each settlement, look at the sign of the inventory before you look at the gain. A gain with no offsetting sale is the 350 result: you were a bettor who happened to be right.
What usually goes wrong
The tempting error is to call any profitable round a good quote. The 12-with-a-one-point market made money and was still called terrible, because nothing was sold and the whole position was directional. The next time that imbalance resolves the other way, the same quote loses the spread it never collected and the inventory as well.
A second error is sizing up because the edge looks obvious. The game allows any quantity. Ten bought at 12 worked when the total was 16. The same ten loses 40 if the total is 8. The lecture's one-unit loss of 4, from buying at 10 into a 6, is the same arithmetic at a smaller size. A third error is treating the 15-second clock as the lesson. The clock only forces a decision. The lesson is whether you traded against a price or tried to earn a two-sided spread.
Questions people ask
What is the spread in this game?
The gap between the bid you buy at and the ask you sell at, as the maker. A one-point differential around a middle of 12 means those two quotes sit half a point on either side of 12, which is how the lecture's 12.5 offer appeared.
Why did the balanced round not need a forecast?
Because buys and sells arrived in matching size. The inventory from one side funded the other, and what remained was the difference that had been quoted. The dice still resolved. They did not have to move your net position.
Is the code a trading system?
No. It is a practice page. The lecture describes it as open source on GitHub Pages and says harder modes, which would tie the hidden values to the faces already seen, might come later. Nothing in the demo places orders in a real market.