The first version of the morning edit always had something to tell you. Five items, every day, without exception. It tested well and it felt terrible after two weeks.
The problem was not the content. It was the promise the format made. Five items every morning tells you, implicitly, that every morning contains five things worth knowing. Most mornings do not. So the product started padding, and once a product starts padding you stop reading it carefully, and once you stop reading it carefully it may as well not exist.
Padding is a design decision
It is tempting to treat this as a content problem, fixable with better ranking. It is not. It is a decision about what the interface is willing to admit.
A fixed-length list cannot say nothing much happened. The shape forbids it. If you want a product that can be honest about a quiet day, you have to build a layout that looks correct when it is nearly empty, which is a harder design problem than it sounds, because empty layouts tend to read as broken.
So the rule we ended up with: the morning edit renders between zero and five items, and the zero state is not an error message. It is a sentence.
Nothing needs you before eleven. The rest can wait.
What it costs to say less
Three things got harder.
Ranking became unforgiving. When you show five items, the fifth-best item is a rounding error. When you might show one, that one has to be right. We spent longer on the threshold than on the ranking.
The product feels less impressive in a demo. A screen with one line on it does not photograph well. We lost something in the first thirty seconds and gained it back in the second week, which is the correct trade and an uncomfortable one to make.
Trust became load-bearing. A product that sometimes says nothing is only useful if you believe its silence. The first time it stays quiet on a day that actually needed you, the whole arrangement collapses. That raised the stakes on everything underneath it.
Where restraint should not go
Restraint is not the same as withholding. There is a version of this idea that becomes precious: an interface so minimal it hides things you needed, in the name of calm.
The test we use is simple. Quiet is for things the product decided were not worth your attention. It is never for things you asked for. Search returns everything. History is complete. Context controls show you the full list of what is connected, not a curated subset.
Restraint is a rule about what the product volunteers. It is not a rule about what it will tell you when you ask.
The typography of not very much
One practical note, since this is where most of the work actually went. A screen with little on it exposes its type. Every flaw in the scale, every slightly wrong measure, every inconsistent baseline is visible when there is nothing else to look at.
We ended up with a narrower measure than we started with, a larger step between the label and the body, and a hairline rule doing the work a card used to do. None of that is visible as a decision. It is only visible as the absence of a feeling that something is off.
Which is more or less the goal.