Head in one hand, elbow on the desk, two monitors still running behind it.
That is the state the trial set out to measure, and it is the one that moved.
But the first thing that changed at these companies was not the calendar.
It was the meetings. Weeks of them, cancelled, shortened, merged, or reduced to a written note.
Handoffs got rewritten, and interruptions got actual rules.
Only then did anybody stop coming in on Friday, so the trial measured both things at once.
What was actually being tested
The intervention was not a day off. It was a work reorganization phase, and only after that a shorter week at full pay.
Companies spent the run up stripping out the parts of a working week that consume time without producing anything.
Then the shorter week started, and average hours fell by around five. Not eight, which is the figure that keeps getting attached to this.
So the thing under the microscope is a package. Nobody ran a version where the day disappeared and the meetings stayed.
Which means the four day week in this study is inseparable from the housekeeping that preceded it.
That is not a flaw in the design. It is the design, and it changes what the finding can be used to argue.
A company that runs the cleanup and keeps five days would presumably also see something. Nobody measured that arm, so nobody knows how much of the gain it would have taken with it.
The chain the statistics actually found
Well being did not improve by magic. The analysis identifies three things carrying the effect.
Self reported work ability rose, which is people saying they can do their job properly.
Sleep problems fell, and fatigue fell.
Those three sit between the shorter hours and the outcomes, and when you account for them the rest of the improvement largely follows.
Burnout dropped, job satisfaction rose, mental health improved, and physical health improved by the smallest margin of the four.
Reductions at the individual level tracked bigger gains than reductions at the company level, which points at personal hours rather than policy as the active ingredient.
The numbers and how they were gathered
The sample is 2,896 employees at 141 organizations across Australia, Canada, Ireland, New Zealand, the United Kingdom and the United States.
Data came from surveys before and after the six month trial.
There were also 12 control companies, and the same pattern of improvement did not appear in them.
That control group is the most important sentence in the paper and it goes missing in almost every retelling.
Without it this is a before and after study, which proves very little. People generally report feeling better when somebody is paying attention to how they feel.
With it, the comparison has somewhere to stand, even if the controls were not randomly assigned either.
It was published in a behavioral science journal in July of last year, by four researchers based in Boston and Dublin, and roughly 9 in 10 companies stayed on the schedule afterward.
What the authors say it does not cover
Every company volunteered. Firms that put themselves forward for a well being trial are probably already better than average at well being, which the authors state plainly rather than bury.
It is not a randomized controlled trial, and the authors say so.
Everything is self reported. No wearables, no clinical measures, and no output data, which matters because the productivity claim rides entirely on people rating their own work ability.
The sample leans heavily toward professional services and office work in English speaking countries, and the authors warn against sweeping claims from it.
The occupations missing are the ones where fatigue is already a measured safety input, as it is when both pilots fall asleep in a cockpit at cruising altitude.
What sleep loss does to a body is well documented elsewhere, which is exactly why the untested cases are the interesting ones.
Why a plant and a ward are the real question
A continuous operation line does not have meetings to cancel. Its week is already stripped, because the machine sets the pace and somebody has to stand at it.
A hospital night shift is the same. You cannot merge a handover into a written note when the thing being handed over is a patient.
For those workplaces a shorter week is not a reorganization. It is a headcount decision, and it costs money that the office version does not.
The published paper makes no claim about them, which is the correct position and also the frustrating one.
Its own data stays private until 2027, so nobody can re run the analysis in the meantime, as the coverage noted at the time.
The honest summary is short. For office work the gains are real and measured against controls, and for everyone else the experiment has not been run.
