30 films · 64 live pages

The picture is the proof.

Short films about everyday puzzles and slow changes, where nothing on screen is an illustration. Either a working simulation is running and the numbers are measured from it as you watch, or real open data is shown exactly as it was recorded. Every film has a page here where you can run the same thing yourself.

running in your browserfilm 09 · the shower problem

Too hot. Turn it down. Now it is freezing.

The water takes six seconds to reach you. Make the hand quicker and watch what a delay does.

Open the full model
30films so far, each three to six minutes
35working simulations
30real records, shown as recorded
1rule: you can check it

What this is

Two kinds of film, and nothing in between.

Explainer videos usually ask you to trust the narrator and enjoy the animation. These films are built the other way round: the picture is the evidence, and the voice only tells you where to look.

Simulation

A model small enough to read

  • The rules are stated in full, on screen and at the top of the code. A lift that never stops. A shop with five tills. A tap and six seconds of pipe.
  • Every percentage, wait and temperature is measured from the model while the film runs. Nothing is drawn to look right.
  • Before a number is used, the model has to agree with an exact result from the textbooks.
  • The same model file runs the film and the live page, so the film cannot show anything the page would not.
Real data

A record, shown as it was recorded

  • One satellite photograph a year. One frequency reading a second. One temperature a day since 1772.
  • The source and the date are on screen. One colour stretch for every photograph, and nothing redrawn.
  • Areas and counts are measured from the record itself, with the method said out loud.
  • Blemishes stay in: a damaged satellite, cloudy years, instruments that could not hear small earthquakes.
If a claim can be run, it should not have to be believed.

Why it is made this way

The narration is an AI voice, and every film says so. A generated voice has no reputation to stake, so the trust has to come from somewhere else. Here it comes from the picture: a model you can run, or data you can look up.

Surprising results deserve a way to check them. That 45% of people in a supermarket queue are overtaken by someone who arrived later, or that the earliest sunset comes a week before the shortest day, sounds like something a video would make up. So each film ships with the thing itself.

Limits are part of the result. Every film ends with what the model leaves out or what the data cannot show. A number without its limits is an advertisement.

How a film is made

Six steps, in this order.

The script is written last but one, from numbers that already exist. That order is the whole method.

  1. Write the rules, or find the record

    A simulation starts as a short list of rules in one file. A real-data film starts with a public source that anyone can download.

    queue/model.js · 5 tills · arrivals at random · service 2 min on average · nobody switches lines
  2. Make it agree with something exact

    The model is run against a result that is known for certain. If they disagree, the model is wrong and nothing is filmed.

    Erlang's formula: 3.05 min
    this simulation: 3.06 min ok
  3. Measure what the film will say

    Long runs produce the figures. For real data, a script counts pixels, readings or earthquakes and writes down the method.

    Lake Mead, water in frame
    1998: 536 km² · 2026: 270 km² counted from the photographs
  4. Write the script from the numbers

    Only now is there a narration, and every figure in it comes from step three. The voice is generated; the words are checked against the measurements.

    "Forty-five per cent of the time." ← overtaken: 45.1% of 540,037 customers
  5. Film the model itself

    The film is the same page, stepped frame by frame, with the numbers computed again as it plays. Even the sound is calculated from the film's own events.

    one page → 1920×1080 film, 1080×1920 shorts, and the live page you can open
  6. Say what it is not

    Each film closes on its limits, and each live page lists every rule. Where a spoken line turned out slightly off, the screen shows the measured figure.

    "What the model leaves out" is a chapter in every film

The films

Pick one, then run it.

Each card opens the live page: the same model or the same data as the film, with the controls handed to you. The newest pages are up before their films; those are marked.

01simulation

Why the lift goes the wrong way

From floor 3 of a 30-floor tower, the first lift to arrive is going down 93% of the time. Your wait for one going up is the same on every floor.

03simulation

Why buses come in threes

Six evenly spaced buses collapse into one convoy by themselves. Doubling the fleet barely helps. One small rule does.

25simulation

Ten more on the clock

A car at 50 km/h stops in 24 metres. One at 60 that brakes at the same instant is still doing 38 there.

31simulation

The 37% rule

A hundred options, seen once each, no going back. Look at 37, then take the next one that beats them all: you get the best 37% of the time.

Run itfilm to come
35simulation

Why an outbreak overshoots

Herd immunity arrives at the peak, not the end. With each case infecting two others, 50% is the threshold and 80% catch it.

Run itfilm to come
36real data

The word “the”

Count every word in a book and rank them. In seven books and five languages the counts fall on the same straight line.

Explore the datafilm to come
37simulation

Fair coin flips, one winner

Two hundred people swap small fair bets. No one cheats and no money leaves, yet one of them ends up with nearly all of it.

Run itfilm to come
39simulation

100 prisoners, 100 boxes

Guessing frees them once in a number with 31 digits. Following the numbers round in loops frees them 31% of the time.

Run itfilm to come
41simulation

Win 51% of the points

Tennis scoring magnifies a small edge: 51% of the points wins 61% of best-of-three matches.

Run itfilm to come
43simulation

Can you fake a coin toss?

A hundred real tosses nearly always hold a run of five. People inventing tosses break their runs up. Try it.

Run itfilm to come
50real data

Peak child

The number of children in the world peaked in 2020. The population keeps growing until 2084.

Explore the datafilm to come
51simulation

Whoever asks first wins

A hundred a side, everyone with their own preferences. The side that asks gets its 5th choice. The other side gets its 20th.

Run itfilm to come
52real data

Coal to zero

Great Britain's electricity every month since 2009: coal from 42% in 2012 to nothing in 2024.

Explore the datafilm to come
53simulation

Why a thrown phone flips

A box spins steadily about two of its axes and tumbles about the third, however carefully it is thrown.

Run itfilm to come
57simulation

Why time runs one way

Molecules hop at random between two halves of a box. Every hop can be undone. The whole never goes back.

Run itfilm to come
62simulation

Look at two queues, not one

Jobs sent to servers at random pile up. Let each look at two and take the quieter, and the worst pile collapses.

Run itfilm to come
63simulation

Two losing games that win

Game A loses. Game B loses. Two rounds of each in turn wins, and so does picking between them at random.

Run itfilm to come
65simulation

The winner's curse

Every bidder guesses honestly and the winner still overpays, by more the more bidders there are.

Run itfilm to come

Watch

The films are on YouTube.

Three to six minutes each, with chapters. Every description links back to its live page here.

In plain terms

What is and is not claimed.

The voice is generated

Narration is an AI voice reading a written script. The pictures and numbers are not generated: they are computed or recorded.

Models are simple on purpose

A lift with no passengers, buses that never fill up. Simple enough to check, and the films say what a real system adds.

Measurements have error

Satellite areas are good to a few per cent. Where a year is cloudy or an instrument changed, the film marks it.

Data says what, not why

A temperature record or a shrinking lake is shown as measured. Causes are outside what the picture can prove.

Sources

Where the real data comes from.