One HTML page: a coffee shop simulated as dots, 8 inside at most, where every Frappuccino has an independent 1-in-6 chance of being a Neon Unicorn and three held in the shop at once sets off a catastrophe. The disaster has to come from the dice rather than a timer, and the model is asked to predict its own catastrophe rate before writing the page, so the live counters check its arithmetic.
Leave it running on fast-forward for a minute before reading the answer. The counters are the answer key.
Watch the unicorn rate. Over a few hundred drinks it should sit near 16.7%. A rate that drifts well away from that, or a readout that is not computed from the actual drinks, means the die is rigged or the stat is decorative.
Is the catastrophe earned? Count the glowing drinks inside the shop when the event fires. It should be three or more every time. Events arriving on a steady beat, or firing with two unicorns in the room, are the scripted disaster the prompt ruled out by name.
Check the escalation tiers. One unicorn inside and two should each look different from zero, and from each other. The common failure is a page that goes from calm straight to explosion.
Count heads. At most eight inside, a visible line outside, and customers who leave carry their unicorn out with them, which lowers the count. A page where unicorns stay counted after their holder has left is miscounting the one number the challenge is about.
Does it rebuild and carry on? A page that ends at the first catastrophe, or rebuilds into a broken state, has done half the brief.
Compare The odds against the catastrophe counter. The prediction depends on the model's own dwell time and service rate, so there is no single right number; the question is whether the paragraph and the page agree. A model that predicts one event per 40 customers while its page shows one per 400, and does not notice, did the arithmetic for the prose only.
You write simulations where the interesting behavior is supposed to come out of the dice rather than be animated by hand.
Build me a single HTML page: a Starbucks-style coffee shop running a bizarre promotion, simulated live.
data: URIs are fine.localStorage, sessionStorage, cookies or indexedDB — sandboxed frame, they throw.html block.One paragraph, before the file. Given your own timings — how long a customer stays inside holding a drink, how fast the counter serves — roughly how many customers should pass through between catastrophes? Show the working, not just the number.
The single fenced html block.
Having written it, what does the page do in its first few minutes? Name anything you tuned, special-cased or nudged. Say what it does badly — I am going to leave it running and watch the counters.