Probability, Randomness & Coin Flips — Explained Simply

A truly random event is one where the outcome can't be predicted from prior information. In practice, "random" splits into two categories:

1. True randomness — driven by quantum-level processes (radioactive decay, thermal noise). Used in cryptography and lottery draws.
2. Pseudorandomness — generated by mathematical algorithms that produce sequences appearing random but derived from a starting "seed". Used in most software (including our decision generators).

For everyday decisions, pseudorandomness is functionally identical to true randomness. The difference matters for cryptography, not for whether you pick pizza or curry tonight.

Close, but not exactly. A 2007 Stanford study by Persi Diaconis showed that a coin caught in the hand (not on a table) lands on the same side it started about 51% of the time — a tiny bias caused by the physics of the flip.

For a coin caught + slapped onto the back of the hand, the bias is closer to 50.8%. Still effectively 50/50 for any single decision, but statistically detectable across 100,000+ flips.

A 2023 replication (Bartoš et al.) confirmed the finding with 350,757 coin flips — landing "same side up" 50.8% of the time. Fun trivia, doesn't affect any real-world decision.

In 100 coin flips, the probability of seeing a run of at least 6 in a row (all heads or all tails) is roughly 80%. In 200 flips, it's effectively 100%.

Human intuition treats such streaks as "not random" — but they're exactly what random looks like. The "hot hand" belief in basketball, the "due for a win" belief in gambling, and the "we've had 3 girls, next one must be a boy" belief in genetics all stem from this misreading of randomness.

Random sequences always contain clusters. A truly evenly-distributed sequence (H T H T H T...) would look suspiciously non-random.

Named after a famous 1913 incident at the Monte Carlo casino where black came up 26 times in a row on a roulette wheel. Gamblers bet increasingly on red, convinced it was "due". They lost millions.

The truth: each spin is independent. The odds of red on spin 27 are the same as spin 1 (18/37 on European roulette, ignoring the zero pocket). The wheel has no memory.

Rule to remember: when events are independent (coin flips, dice rolls, roulette spins), what happened before tells you nothing about what happens next.

Some "random" processes have memory:

• Card decks — as cards get dealt, the probability of remaining cards changes. This is why card counting works in Blackjack.
• Weighted samples without replacement — pick a name from a hat, don't put it back, and the odds shift for the next pick.
• Population sampling — as you interview more people, the probability of interviewing another person with the same trait changes.

For everyday decisions, most people confuse these two categories. A coin flip has no memory. A shuffled Spotify playlist (which uses "shuffle without replacement") does — that's why "shuffle" sometimes feels non-random.

Spotify famously changed their shuffle algorithm in 2014 after users complained it "wasn't random enough". The original was truly random — which produced the streaks and clusters described above. Users perceived this as broken.

The new algorithm deliberately spreads songs out — same artist doesn't repeat within N tracks; genre is roughly balanced through the queue. It's less statistically random, but feels more random.

Lesson: for user-facing "random" features, perceived randomness often matters more than statistical randomness.

Three doors. Behind one, a car; behind the other two, goats. You pick door 1. The host — who knows what's behind each door — opens door 3 to show a goat. Should you switch to door 2?

Yes, always switch. Sticking with door 1 gives you a 1/3 chance of winning; switching gives you a 2/3 chance. This is counter-intuitive because most people treat the revealed information as "50/50 between the two remaining doors" — but the host's knowledge changes the maths.

Marilyn vos Savant published this in 1990 and received 10,000 letters of protest, including from PhD mathematicians. All wrong; she was right. Simulations settle it: switch and win 2/3 of the time.

The single most-common probability mistake in everyday reasoning: ignoring base rates.

Classic example: A medical test for a rare disease (base rate 0.1% in the population) is 99% accurate. You test positive. What's the probability you actually have the disease?

Intuitive answer: 99%. Correct answer: about 9%. Here's why: in 100,000 people, 100 have the disease (99 test positive). But 999 of the healthy 99,900 also test positive (1% false-positive rate). So of 1,098 total positive tests, only 99 are true positives — 9%.

Bayes's theorem formalises this. Most humans get it wrong without conscious training.

Every RDG picker uses the browser's crypto.getRandomValues() — a cryptographically-secure pseudorandom number generator. In practice this is more than random enough for any decision-making purpose.

Some of our pickers (Spotify-style) deliberately weight against consecutive repeats — a Random Picker pulling from a list of 20 items won't pick the same one twice in a row, even though "true random" would occasionally do so. This is a UX choice, not a technical one.

Our Coin Flip, Random Number Generator and Magic 8-Ball are all true 50/50 (or evenly-weighted N-way) — each event independent of the last.

This resource is designed for classroom use. Print, share, cite — all we ask is a link back to the source. Suggested citation: "Random Decision Generator — Probability, Randomness & Coin Flips Explained Simply (2026)".