The Working Memory Stress Test
Everyone knows the number seven. Almost nobody has measured themselves. In the next few minutes you will run the same memory task twice — once in silence, once while something small and stupid interrupts you — and watch your own capacity fall between the two. Guess your interrupted score first. Then find out.
You will be shown a list of words, one at a time, then interrupted by a short sum, then asked to repeat the words back in order. How long a list do you think you will manage?
Not your best day. Not the number you have read somewhere. The number you would bet money on, right now, on this device, with whatever noise is around you.
Hold the list.
This is your capacity, measured twice.
The longest list you repeated back perfectly with nothing in the way.
The same task, with six seconds of arithmetic dropped in the middle.
What the interruption cost you — the only number on this page that matters.
You have met this limit before. You called it something else.
Nine numbers on one slide
A slide with nine figures is not a dense slide, it is an unreadable one: the room can hold about four, so it holds the four nearest the top and argues about those. The fix is not a better speaker. It is fewer numbers, pre-grouped.
"Wait, what did we already rule out?"
Twenty people, three theories, someone joining every ninety seconds. Every join is an interruption for whoever is holding the thread. This is why incident command writes state down — the whiteboard is not admin, it is the memory.
Everything the leaver knew
An hour of talking transfers roughly nothing, because the receiver is at capacity by minute two. What transfers is an artifact they can re-read when the question actually arrives.
The gap between meetings
Back-to-back scheduling assumes the cost of a switch is zero. In a controlled experiment, interruptions averaging 2.8 seconds — about the time to perform one step of the task — doubled the rate of sequence errors, and 4.4-second ones tripled it. The interruption here lasted six seconds. A meeting lasts thirty minutes and arrives with its own unfinished business.
What this is, and what it is not
This is a demonstration, not a measurement instrument. One run each way, no practice trials, on your own device, with whatever is happening around you. Your two numbers would move if you ran it again — so do not write either of them down as a fact about yourself.
What is robust is the shape. The modern account is not really about storage at all: capacity is controlled attention — what you can hold on to while something else competes for you — which is why it predicts performance on real work, and why it collapses the moment anything interferes. The number usually put on it is about four chunks, and "about" is doing real work in that sentence: it is an average across people and tasks, not a count of slots. The older "seven, plus or minus two" is looser than its fame suggests, and Miller said in the same paper that chunking changes what counts as one item — which is why a phone number is easier than seven unrelated words.
The design here is a complex-span task, which exists precisely because interleaving a processing task with the items lowers the span, reliably, in everyone. And if you are wondering whether you could practise your way to a bigger number: across 87 publications and 145 experimental comparisons, working memory training improves the trained task and close relatives of it, with no convincing transfer to reasoning, comprehension or arithmetic.
So the interesting comparison is not your number against anyone else's. It is your number against your other number.
Everything above, and where it comes from.
Each entry says what the finding actually shows and where it stops. If you are going to repeat one of these numbers in a room where someone might check, read the note first.
You cannot train your way past this. There is no discipline that adds a fifth slot, and the people who believe they have one are simply the people who have never measured. What you can change is what you spend the slots on — which is the whole of the practice: write state down so it stops costing you memory, group information before you hand it over, and defend the runs where the list is allowed to stay in your head. Capacity is fixed. Consumption is a choice.