- Explore rare events, shared birthdays and coin-flip streaks in three interactive probability experiments.
- Compare exact theory with 100, 1,000 or 10,000 simulated experiments and inspect sampling variation.
- Replay a seed, share an experiment link, and preview or save a clearly labelled result card.
- A Million Chances Still Leave Room for Nothing
- Expected Once Is Different From Guaranteed Once
- Why 23 People Can Be Enough for a Birthday Match
- Coin Streaks Are Less Simple Than They Look
- Read the Tiles Before Arguing With the Percentage
- What the Confidence Interval Does and Does Not Say
- Share a Result Someone Else Can Reproduce
- Probability Simulator Questions
- Keep Following the Numbers
A one-in-a-million event sounds like something you can safely forget about. Try it a million times, though, and intuition starts arguing with arithmetic. Is a hit guaranteed? Surprisingly, the chance of at least one is only about 63.21%.
ContentBASE One-in-a-Million Lab turns that argument into an interactive probability simulator. Explore rare events, the birthday problem and coin-flip streaks. Compare mathematical probabilities with repeatable simulated experiments, then share the result that surprised you most.
One-in-a-Million Lab
Explore rare-event probabilities, the birthday paradox and coin streaks with exact mathematical models and repeatable simulated experiments.
Loading the probability lab… Enable JavaScript to experiment.
Start with free Canva bundles
Browse the freebies page to claim ready-to-use Canva bundles, then get 25% off your first premium bundle after you sign up.
Free to claim. Canva-ready. Instant access.
1. A Million Chances Still Leave Room for Nothing
The default experiment gives each independent attempt a one-in-a-million chance of success, and puts one million attempts in a batch. A teal tile means that entire batch contained at least one success. An amber tile means it contained none.
You might expect every tile to turn teal. Instead, about 37% of batches miss completely. That does not contradict the expected number of successes being one: some batches have several successes, while others have none.
Change the odds or attempts, then select Run experiments. Try 100, 1,000 or 10,000 batches. The theoretical chance is calculated from the model; the observed hit rate comes from the particular sample you drew.

2. Expected Once Is Different From Guaranteed Once
For a constant probability p repeated over n independent attempts, the expected event count is np. The probability of at least one event is 1 − (1 − p)n. These answer different questions.
Expectation describes the average count across repeated batches. A probability describes how often a specified outcome occurs. Multiplying a small probability by the number of attempts can give an expected count; it is not a general formula for the chance of a hit.
The lab also calculates how many independent attempts are needed to reach a 50%, 95% or 99% chance of at least one event. Even 99% leaves a possibility of missing. Ordinary repeated attempts with a nonzero failure chance do not become a guarantee at a particular finite count.
3. Why 23 People Can Be Enough for a Birthday Match
Choose Shared Birthdays. Under the standard model of independent birthdays spread uniformly across 365 days, a group of 23 people has about a 50.73% chance that at least two people share a birthday.
The tempting mistake is to compare everyone with one particular person. The experiment asks whether any pair matches. Twenty-three people give you 253 pairs to think about.
Those pairs are not independent, so the lab calculates the complement: the chance that every birthday is different. Each simulated group draws individual birthdays and checks for a match. With 366 people and only 365 possible birthdays, a match is certain.
Real births are not evenly distributed across the calendar. This mode explains the classical birthday problem; it does not use a population’s actual birth records or include leap day.
4. Coin Streaks Are Less Simple Than They Look
In Coin Streaks, ask whether a sequence contains a run of consecutive heads. Change the number of flips, the target run length and the probability of heads.
Getting ten heads in ten specified flips is different from finding ten heads somewhere within 200 flips. A longer sequence gives the streak more places to occur. But the candidate starting positions overlap, so treating all those opportunities as independent gives the wrong answer.
The theoretical calculation tracks the length of the current trailing run of heads. Once a path reaches the target, it stays counted as a hit. The simulated experiments flip their own coins until they reach the target or run out of flips.
A long losing streak also does not make the next independent flip more likely to be heads. Change the heads probability and the odds change; simply feeling overdue does not change them.
5. Read the Tiles Before Arguing With the Percentage
Demanding experiments calculate in short cancellable tasks, with a visible calculating state. Start another run to replace one in progress. Very small model rates and interval endpoints use scientific notation when ordinary percentage rounding would hide them.
Each tile represents one complete experiment, not one individual attempt, person or coin flip:
- Rare events: one tile is an entire batch of the chosen number of attempts.
- Shared birthdays: one tile is a new group of the chosen size.
- Coin streaks: one tile is a full sequence, counted as a hit if the target streak occurs anywhere.
The mosaic reveals its sample over four seconds. Reveal all shows it immediately. Switching tabs does not pause the reveal; an explicit pause stays paused. Devices requesting reduced motion see the completed result immediately.
The curve below the tiles comes from theory, independently of the sample. A small sample can wander visibly above or below it. Drawing more experiments usually makes the observed rate more stable, while still leaving sampling variation.
6. What the Confidence Interval Does and Does Not Say
The lab displays a 95% Wilson score interval around the observed proportion of experiments with a hit. It describes uncertainty from drawing a finite Monte Carlo sample.
The mathematical probability is already determined by the chosen model. The interval does not make that exact calculation uncertain, certify a real-world risk estimate, or prove that attempts in an actual situation are independent.
A 95% confidence method refers to its long-run coverage under repeated sampling. It is not a promise that every displayed interval contains the model probability.
7. Share a Result Someone Else Can Reproduce
Run experiments chooses a fresh seed. Replay seed keeps the visible seed and settings, producing the same sample again.
Copy experiment link preserves the mode, inputs, experiment count and seed. Use it when sharing an unusually lucky or unlucky result so someone else can inspect exactly what happened. The result card includes the event definition, theoretical chance, observed hit rate and seed.
The random generator is designed for repeatable educational experiments. It is not a cryptographic generator, a gambling system or a source of measured real-world odds.
8. Probability Simulator Questions
8.1 Does the Lab Perform a Trillion Individual Attempts?
No. In rare-event mode it calculates the chance that a whole batch contains at least one hit, then samples that batch outcome directly. That is sufficient for the hit-or-miss question and keeps huge input counts practical. Birthday and coin modes simulate their groups or sequences directly.
8.2 Why Is the Simulated Percentage Different From the Theoretical One?
Theory describes the probability under the assumptions. A sample is one possible collection of outcomes. Replaying the same seed repeats that collection; running with a fresh seed produces another one.
8.3 Can I Enter Actual Risk Figures?
You can explore a chosen one-in-N probability, but the tool does not verify its source or whether it remains constant and independent. Changes in age, exposure, behaviour or shared causes can make that simple model inappropriate.
8.4 Is This a Revival of an Older Website?
No. One-in-a-Million Lab is an original ContentBASE experiment. It shares the collection’s interest in making abstract numbers feel tangible.
9. Keep Following the Numbers
For a related thought experiment about repeated annual risks, try the ContentBASE Life Simulator. Put your own lifetime in perspective with Your Life on Earth, or investigate how a tiny disturbance spreads in Traffic Jam Lab. Find the collection in ContentBASE Tools.