Gallery
Discover apps people built on Collabby by chatting. Run them, then remix to make them yours.
Doing Two Things at Once
It may feel like you’re doing two things at once, but you’re really switching quickly—and every switch takes time. 🎯 First, measure yourself: press whether each number is odd or even. Try eight rounds and you’ll see that each round takes a different amount of time, so we use the average from several rounds. 🔀 Next, mix it up. Add a second rule to the same task so «odd/even» and «size» take turns. Move the mixing handle to 0, 25, 50, or 100 and measure again—at 0, there’s no switching, so the difference disappears. ⏱️ Finally, find where the time went. Compare rounds with «switches» to rounds where you «kept going»: the slowdown comes from switching, not from knowing two rules. That difference is the time it takes to switch once. This app was made with code too · Remix it and measure our class’s results!
How Much More Interest for a Longer Term
💸 “A longer term makes each payment lighter” — true, but there’s a cost. First, explore the monthly payment. With 0% interest, it’s “borrowed amount ÷ term”; with interest, the payment grows even for the same term. 📈 Then add up all the interest. Each extra month lowers the payment but raises the interest, shown side by side on one screen — that’s the trade-off. There’s nowhere that extending the term reduces interest. 📊 Finally, view the repayment schedule. The payment stays the same, but at first more goes to interest and later more goes to principal. So the month when the debt is cut in half is “always” later than the midpoint — exactly at the midpoint only when interest is 0%. This is a learning model, so it uses no real products or rates.
How Credit Scores Rise and Fall
📉 Credit scores rise slowly but can drop all at once. Slow to rise — use less of your limit and pay on time, and you’ll gain 8 points a month. Raising your score by 80 points takes ten months. Fast to fall — one late payment can cost 250 points, and getting them back takes 32 months. It takes 12 months of on-time payments to gain 96 points. Is this true? We test four common claims, like “Never getting a card gives you the highest score.” ⚠️ This isn’t a real credit-scoring formula, but an educational weighting model showing what moves in which direction and by how much.
Insurance Changes the Worst Case
🤝 Insurance does not change the «average» — it changes the «worst case». Carrying it alone — Accidents happen rarely. But when they do, they come all at once. Pooling it together — Everyone puts in a little, and whoever is hit takes the whole pot. Is this true? — Things people often say about insurance. Are they true?
Progressive Tax Goes Step by Step
🪜 Moving up a bracket does not put that rate on «all» of your earnings. Step by step — A tax table from a made-up country. Each step has its own rate, and it applies only to the money that lands on that step. Does earning more hurt? — Pay the tax and look at what stays in your hands. What happens if you earn a bit more? What is my rate? — There are two rates — the one on your last coin, and the one you get by dividing your whole income.
What You Get When the Exchange Rate Moves
💱 Money is different in different countries. The rate for exchanging it is the exchange rate, meaning how many won it costs to buy «1 dollar». Move the handles to change the exchange rate and the money exchanged, and see how the dollars you receive are decided. Exchange back to check that you return exactly to where you started — exchanging money itself doesn’t create or destroy money. 📉 But here’s the tricky part: when the exchange rate has «risen», the number is bigger, which may sound good, but it means 1 dollar costs more, so the same amount of won buys fewer dollars. Paying more won for a toy that still costs 20 dollars is another way to see the same thing — the won has «weakened». 🧳 Finally, choose who wins and who loses in four situations: travel, exports, studying abroad, and foreign visitors. The key question is simple — are you the one «paying» dollars or «receiving» them?
The First Number Pulls Your Answer
⚓ Seeing a “totally unrelated” number first can pull your estimate toward it. Instead of hearing about someone else’s experiment, try it yourself with six questions—draw one card from a bag with 3 large-number cards and 3 small-number cards for each question, decide whether the answer is larger or smaller than that number, then write your estimate. At the end, compare the estimates from the “three large-number” questions and the “three small-number” questions as ratios to the true answers. 🎚️ Use the slider to see why: people “start” from the first number and move toward the true answer, but don’t go all the way. Turn the movement all the way up, and the difference between the large- and small-number groups becomes “exactly 0.” Anchoring happens because we “don’t move far enough.” 🏷️ Finally, find the numbers acting as reference points in list prices, opening offers, the highest menu price, and donation suggestions. ⚠️ The app also explains the limits of your experiment—with six questions, you can’t say that the chance of getting a result like this is less than 1 in 20.},
When It Becomes Ten Times Bigger
📈 There’s one question: “Does anyone have the same number?” The answer never changes, but the number of steps can be completely different. 🤝 Compare every pair, and the squares stack into a triangle—that triangle is “people × (people−1) ÷ 2.” When the number of people doubles, the steps grow to nearly four times as many. 🏷️ Make a table once (20 setup steps), then it takes one step per person, so with 8 people the two methods meet exactly at 28 steps. With fewer people, you can even check by hand that the “slow method” is actually faster—slow isn’t always bad. 🔟 Finally, make the number of people ten times larger. The number of comparisons grows by a little more than 100 times, and gets closer to 100 times as the group grows, but it is never exactly 100 times.
Pointing to the Same Box
🔗 “I fixed one side—why did the other change too?” Explore a programmer’s most common trap hands-on. 🔢 Numbers and text keep their own «share» under each name, so copying moves the value and they don’t stay connected. That means you can’t make “changing a changes b too” with numbers. 📦 Names, on the other hand, only «point» to boxes, so if two names point to one box, changing either name changes the same box—we show it clearly with two arrows joining one box. 🧪 Finally, try three ways to copy things: «pointing» changes both the outside and inside, «copying only the outside» separates the shells but still shares the inner box (the shallow-copy trap), and only «copying everything» separates them completely. You’ll also see that there’s no method where the outside follows changes while the inside stays the same.
Input Order, Output Order
📚 Even when you add the same things in the same order, the order they come out depends on the “container.” A stack, like a plate pile where you add and remove only from the top, comes out exactly backward. A queue, like a ticket line where people join at the back and leave at the front, comes out in the same order. 🎯 Mixing adding and removing can change the order “to some extent,” but not any way you like — a stack can’t make 「3 · 1 · 2」, and a queue can’t make 「3 · 2 · 1」. 🖨️ At the end, try choosing the right container for print jobs, undo actions, ride lines, and matching brackets. 🧾 Brackets can fool you if you “only count”: even with 2 opening and 2 closing brackets, the pairs can still be wrong—and stacks catch that.
Same Ballot, Different Order
🔀 The three groups’ «preferences» never change: Group 1 is 🧑🦲>🧑🦰>🧑🦱, Group 2 is 🧑🦰>🧑🦱>🧑🦲, and Group 3 is 🧑🦱>🧑🦲>🧑🦰. Instead of choosing three proposals all at once, compare them two at a time and carry the winner forward. Then, depending on the order of comparisons, «each» proposal can pass. 🪑 So in a meeting, the person who decides what to compare first is effectively deciding the result. Try changing the group sizes to see that this happens only when «no group is larger than the other two combined». 🎯 At the end, predict the winning proposal using only the group sizes and the first matchup.
Same Hearts, Different Results
🙋 The “hearts” of 1,000 voters never change—the number of people in the three villages and which side they prefer stay fixed. If everyone votes, the result is always 🧑🦱 560 : 🧑🦲 440. But move just one “turnout” handle, and 🧑🦲 wins. No one changed their mind. 🚪 Next, count the people who “didn’t go”—if supporters of the losing side stayed home, the result would have flipped even if only they had voted. 🎯 Finally, look only at those who showed up and guess the winner (one round is exactly tied).
Same Votes, Different Seats
🧮 The vote totals stay fixed: 🧑🦱 47 · 🧑🦰 33 · 🧑🦲 20. Line up three ways to turn votes into seats (quotients and remainders · divide by 1, 2, 3 · divide by 1, 3, 5), then change only the number of seats to see the results split — with some seat totals, «divide by 1, 2, 3» gives 🧑🦲 exactly 0 seats. 📉 And with «quotients and remainders», there’s a spot where adding one seat makes a party lose one instead (7 seats → 8 seats). Even though nobody changed their vote. 🎯 At the end, guess which method was used just by looking at the seat allocation.
Same Votes, Different Results
⚖️ 100 members. Keep the yes, no, and abstention votes fixed, and change only the «passing rule» to see approval or rejection — simple comparison, majority of those present, majority of all members, or two-thirds of those present. First, ask whether the meeting is valid (a majority of all members are present). 🤐 Without changing a single yes or no, adding abstentions flips the strictest rules first — under some rules, an abstention has practically the same power as a no. 🎯 At the end, choose which of four vote counts passes under one given rule.
Same Votes, Different Seats
🗺 25 people are seated, with the vote fixed at 🧑🦱 15 : 🧑🦲 10. Simply changing the “lines” dividing them into five groups of five can produce seat results of 5:0 · 3:2 · 4:1 · 2:3 — without changing a single vote. 🗑 “Wasted votes” explain why: no matter how the map is divided, 10 votes are always wasted, and the side that wastes fewer of its own votes wins more seats relative to its vote share. 🎯 At the end, guess which map created each result just by looking at the outcome.
Same Ballots, Different Winners
🗳 Three groups ranked three candidates. Move only the group sizes, and plurality, runoff, and head-to-head voting produce different winners—without changing a single ballot. In the default setting, head-to-head voting goes in a full circle, so there is no “candidate who beats everyone.” 🧑🦳 Add a new candidate, and the winner changes even though none of their supporters wins. 🎯 At the end, choose how the votes must have been counted to produce this winner.
The Question Already Holds the Answer
❓ “Are you still late for meetings these days?” — Whether you answer yes or no, “You used to be late” stays built in. There is only one of the three answers that removes what was planted. 🚫 Turn the sentence into a negative, and two things split apart: one disappears as soon as it is negated, while the other stays firm — the one that remains is the “presupposition.” Left as it is, both remain and cannot be told apart. 📋 At the end, choose the survey question that plants nothing in the respondent’s mind.
Who Did It Disappeared
🫥 Write the same event in four ways. Turn two dials: sentence form (active or passive) and whether to name or hide one person. Whether you can answer “Who did it?” depends only on whether a person is named—not on active or passive voice. Even in passive voice, naming them keeps the answer; even in active voice, hiding them makes it disappear. 🔲 When you see all four boxes at once, the answerable ones split vertically, not horizontally. 📄 At the end, choose the apology that takes responsibility from three apologies containing “I’m sorry.”
How Will You Check?
🔍 What separates facts from opinions isn’t the tone—it’s whether there’s a «way to verify» them. Judge the six cards one by one: only the three with a way to check are facts—baking time records, an ingredient list, and sales records. 🧪 When four pieces are mixed into one sentence, only two of the four can be verified; turn off the opinion pieces and it’s 2/2, but turn off the fact pieces and it’s 0/2—a long sentence that says nothing. 📣 At the end, choose which of the three ad lines can be checked later.
Not Knowing What It Is
👉 “It” can refer to any object mentioned earlier. Change the scene from 1 to 4 objects and the number of times “it” appears from 1 to 2, and the possible readings “multiply” to 1, 2, 3, 4, or 9. With only one object, it always has exactly one meaning, no matter how often it is mentioned. 🪞 Even in a short scene with just two objects, both interpretations can be grammatical; repeating the name makes only one meaning clear. 🧾 Finish by choosing sentences that aren’t confusing.
One Word Picks a Side
🔗 Tell the same bus story with two facts. Keep every word the same and swap only the link—“and,” “but,” or “so”—and the impression can tilt as far as 22:78. Switch the order, and the winning fact flips exactly; choose “and,” and the bias disappears. 🔀 With three facts, the one placed last becomes the conclusion—this happens in all six possible orders. 📰 Finally, choose which of three sentences using the same two facts “recommends this bus.”
Why Top Scorers Do Worse the Next Time
🎯 A hundred people sit the same test twice. A score is «skill + luck». Keep only the ten who did best in round one, look at round two, and the average drops sharply — yet nobody got worse. The ten who did worst go up instead. Both simply came back toward the «middle». Picking the top scorers does not pick skill alone; it also picks «whoever was lucky that day», and that luck does not follow them. 🎲 Second, the check. Turn «luck's share» down to zero and the two rounds become identical, and the drop is exactly zero. The more luck there is, the further it falls on its own — and skill was never touched. 📰 Third, the judgement. You praised someone after a brilliant day and they did worse; you scolded after an awful one and they improved. Did the praise ruin it? This app was made with code too · Remix it and change luck's share.
Why Winning Every Part Can Still Lose the Whole
🏥 Two hospitals' surgery success rates. For mild cases it is A 93% against B 87%; for severe cases A 73% against B 69% — split, A leads in both. Pool them and it becomes A 77%, B 83%: the ranking flips. Not one of the four success rates was touched. All that changed is «who took more of which patients». 🔀 Second, the reason. Severe cases succeed less at either hospital, and 80% of A's patients are severe against only 20% of B's. The pooled number shows that gap, not skill. Set the two shares equal and it can «never» flip — what flips it is not the rates but the mix. 🔍 Third, the judgement. If you are a severe case, read the number for «your own group», not the total. This app was made with code too · Remix it with different shares.
A 99% Accurate Test Can Still Be Wrong
🧪 You test positive on a 99%-accurate test. What is the chance it is real? Line up ten thousand people and the answer appears — if 1 person in 1,000 has it, then 10 of the 10 real cases test positive, and 100 of the 9,990 clear people test positive too. That is 110 positives in all, of which only 10 are real: 9%. There are so many clear people that the false positives swamp the real ones. 📉 Second, the curve. Without touching the test at all, changing only «how rare it is» moves the answer from 9% to 92%. Meanwhile raising «the share of true cases caught» from 80% to 100% barely moves it — what decides the answer is not the test but the rarity. 📬 Third, spotting it. With three tests that are all equally 99% accurate, 1-in-1,000, 1-in-100 and 1-in-10 give 9%, 50% and 92%. This app was made with code too · Remix it with the numbers of a test you know.