What’s inside:

• Part 1 · Sample City — 100 hidden residents in a 10×10 grid. Survey 10 at RANDOM reveals a fair sample and scales it up (“6 of 10 → ≈60% of the city”), with a running history of estimates so students see sampling variation (50%, 70%, 60%…). The ⚠️ ONE street only button surveys Elm St. — where neighbors think alike — and overshoots every time, making bias tangible. Reveal the whole city shows the true 60% and compares it against both sampling methods.
• Part 2 · The Carnival Games — four rotating booths (double-heads, spin-&-flip, at-least-one-head, double-green). Students commit to one of three probabilities, then the booth opens its full sample space as a table with winning cells in gold — feedback specifically names the collapsed-list trap (HH/”one of each”/TT) and the adding trap (⅓ + ½ = ⅚). After the reveal, “Play it 200 times ▶” runs a live simulation showing experimental probability wobbling around theory.
• Part 3 · Behind the Booths — 🔍 Inference Machine (checks randomness first, finds the sample proportion, scales up with the “not k, and not the percent” warning, and reports honestly with wobble language), ⚔️ Data Duel (three preset matchups engineered to separate the concepts: A-better-and-steadier, B-clearly-better, and same-center-different-consistency — comparing mean and range with plain-language verdicts), and 🌳 Sample Space Builder (three two-step experiments: count each step, multiply the counts, list every outcome, extract P).
• Part 4 · Quiz — 10 questions across sampling design, inference scaling, sample variation, center-and-spread comparison, the probability scale, the 50/50 fallacy, the gambler’s fallacy, compound sample spaces, AND-multiplication, and experimental vs. theoretical.

Misconception coverage summary: Questions 1–3 target sampling and inference errors — trusting friends, the pizza line, or a single team as a sample, reporting the raw sample count or the percent as the population answer instead of scaling, and misreading normal sample-to-sample variation as error or false precision. Question 4 targets comparison traps: judging a group by its single best score and believing more varied scores mean more consistency. Questions 5–7 and 10 target probability-model errors: probabilities above 1, the “two outcomes = 50/50” fallacy, the gambler’s fallacy in both “due” and “hot streak” forms, and over-reading small experiments as loaded dice or permanently changed probabilities. Questions 8–9 target compound-event traps — the three-outcome collapsed sample space for two coins and adding instead of multiplying for AND events.