If project leadership is distributed, what keeps the project coherent?

Hassan, I think visibility is a critical part of the answer.
I would add, though, that visibility alone may not be enough to produce coherence.

Two teams can understand each other’s priorities, dependencies and decisions and still make locally legitimate choices that interact in ways neither can resolve alone.
At that point, the issue is no longer simply whether information is shared, but whether the system can recognize material interdependencies, surface the resulting trade-offs and determine who has legitimate authority to resolve them.

I also think your point about escalation raises an important distinction. Ideally, integration should not begin only when decisions visibly conflict.
By then, their effects may already have propagated elsewhere in the project.
The stronger capability may be detecting consequential interdependencies early enough for coordination or intervention while meaningful choices still remain.

So I agree that distributed leadership does not require one person controlling everything.
But it does require more than distributed visibility. It requires an architecture through which locally legitimate decisions can remain coherent as they interact across the system.

That, for me, is where distributed leadership either remains genuinely distributed or gradually turns into fragmented leadership.