EXP-04 counts one thing: the act of changing attention. Not what it cost, and not where it landed.
I work in public. Large areas, a lot of people, and phones go off that aren't mine. Other people's messages, calls, whatever.
I check mine anyway.
I know it's their phone. I check mine. Even though the surface isn't on, that is the digital surface. By instinct, you almost stop what you're doing when you hear a phone go off. You check if it's yours when you know it's not.
Where I work is what made it visible to me. That's my condition, not everyone's.
Last week's experiment ended on something I never worked out. My thumb opened the wrong app on a phone gone gray, and I don't know why. The cause stayed hidden. The act didn't.
This week's experiment counts the act.
The instrument is a piece of paper.
You carry it for one waking day and make one mark every time your attention changes to, from, or between a digital surface. A switch counts whatever set it off. Another person's phone going off counts if it pulls you from what you were doing to your own phone. You don't sort the marks. You don't timestamp them. You don't stop to work out what any of them meant. At the end of the day you write down the total, and the total is the finding.
That's the whole experiment. Day seven of the protocol. A folded piece of paper and a pen. Mine goes in the clip on a marker lid. I always have a pen or a marker on me.
Earlier this year we published a week of content on context switching, called The Cost of Switching. Every day in it priced a switch as a loss.
It kept one column.
The first thing the count made visible was how small the smallest ones are.
The first time I ran it, I didn't count looking for my phone. I remember saying to myself, no, that doesn't count, I was looking for my phone. When I ran it again, I counted it. Not really a panic all the time. Mostly a pocket check. Is it over there where I left it charging. Where's my phone.
It's one a lot of people don't count. I mean, why? It pulls your attention from whatever you're doing.
Not every pull netted out negative.
One kind I mark is the just-in-case check, while I'm waiting on somebody. You text a person, you give them a couple of minutes, you check. Email runs a little longer. A couple of hours, then you look. And if you leave the thread open, it won't give you an alert, because you're in an active chat. It's very situational.
Most of those checks find nothing. Maybe one in ten finds something. That's my estimate, not a count.
The one that paid off: I'd pressed the volume down to the bottom without realizing it, and the phone was silenced. I don't use vibrate. I hadn't heard back in a while, so I checked, and there was an icon saying a message was waiting. The phone was muted. That's where it paid off.
I'm not going to build a rule out of one instance. But it is the instance, and a one-column count has nowhere to put it.
The misses do have a cost. The cost is attention residue. The check turns up nothing and it's on my mind now. I'll check it later. I'll check it in a minute. It's in the back of my mind.
So what separates a pull that nets something from one that doesn't? I'm not really sure anything does. The closest I have is where you ended up.
Notice what that is not. Our own prompt in August asked people to sort each switch as it happened: chosen, requested, automatic, avoidant, unclear. Every one of those is about what started it. Mine is about where it landed.
The audit doesn't ask either question. It pulls forward the act of changing attention, and in some cases focus. The consequence is a separate issue. The task you switched into is a separate issue. The audit counts the act.
The totals move a lot between runs. No two days are exactly the same, whether it's a work day or a relaxed, non-structured one.
Where it does bleed is when marking the switch becomes its own metric. I absolutely changed my behavior after the first run, once I had a baseline to compare against. The background thought was: is this more than last time. Does that count, or does it count as one long switch or multiple. I always went with my first instinct, to not taint the sample as much as possible.
There is no good or bad number here. The audit is observe and report. It scores nothing, assigns no countermeasure, and asks you to continue nothing afterward.
The full conditions and what to record are on the EXP-04 Context Switch Audit page.
This is a standalone experiment. The number has no consequence, so be truthful.
Others have looked at returns rather than hours. In 2026, researchers at Aalto University followed the web and app use of 277 people for seven months and paired it with surveys on how overloaded they felt. Repetitive, short-duration use of devices set the most overloaded people apart from everyone else, with a large effect. The most overloaded weren't the heaviest users.
One study, and an observational one. It can't say the returns cause the overload, and it didn't test an audit like this. A one-day paper tally and months of logged sessions don't confirm each other.
Lassila, H., Aledavood, T., Kulshrestha, J., & Lindqvist, J. (2026). Stop Fiddling With Your Phone and Go Offline: People Experiencing High Information Overload Have Sparse Online Sessions. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26). ACM. https://doi.org/10.1145/3772318.3790822