What does a “single source of truth” actually require?

The phrase is used to sell software and it is not a software property

Buying one platform does not give you a single source of truth. It gives you one place where the truth could live, which is necessary and nowhere near sufficient.

The failure is always the same: the system holds the drawing, and somebody emails a copy to a subcontractor because it is faster, and now there are two. Not because the system was bad. Because emailing was easier than explaining, and nobody was accountable for currency.

What it actually requires

Five things, and only the first is about software:

• One authoritative location per fact — and everyone knows which. A drawing’s current revision lives in the register. A worker’s hours live in the timecard. Not "usually", not "should".
• A status on everything. Work in progress, shared, published, superseded. A document with no status is a document whose currency is a matter of belief.
• Active withdrawal, not passive replacement. Issuing revision C does not remove revision B from a site office, a laptop, or a subcontractor’s inbox. Somebody must withdraw it, and be able to prove they did.
• A rule about who may change what. If two people can both edit the authoritative record, you have two sources of truth that agree today.
• And it has to be FASTER than the workaround. If getting the current drawing from the system takes longer than asking a colleague to email it, people will ask the colleague. Every time. This is not a training problem.

The test

Pick a fact — the current revision of a drawing, the hours a crew worked last Tuesday, the value of committed cost on a package. Ask three people, independently, where the authoritative answer lives.

If they name three places, or if two of them name the same place and are wrong, you do not have a single source of truth. You have a system, and a set of habits that route around it.