There Is No Single Source of Truth Until Someone Assigns It

Which file is the right one?

On almost every defense and aerospace program, a room full of capable engineers eventually loses an afternoon to that question — and the answer ends up depending on one person's memory. Not because anyone was careless. Because nobody ever decided.

Programs rarely design their data. They accumulate it, one sensible local decision at a time, until they are running on an architecture nobody chose. One team keeps the master spreadsheet. Another team has the newer version in a different tool. A third team exports both, reconciles them by hand, and produces a third version that everyone treats as official until the next review, when someone asks where it came from and nobody can say.

Then someone says the magic words: "We need a single source of truth." As if declaring it makes it so.

It doesn't. A single source of truth doesn't exist until someone assigns it.

Origin is not authority

Every piece of data has an origin — where it was created — and an authority — who decides what it should be. They are often not the same, and confusing them is where most of the trouble starts.

The team that generated a number is not automatically the team that gets to say what the number is. The newest tool is not automatically the official one. The loudest team in the review is not automatically right. Authority has to be decided and written down: for every important value, one owner and one official source. Not inferred. Not implied by org chart or by who spoke last. Assigned, explicitly, where anyone on the program can find it.

Until that happens, every "single source of truth" is just a folder with a confident name.

The cost shows up later

This feels like bookkeeping. It isn't. The cost of unassigned authority shows up later, in places that look like completely different problems.

Design reviews that turn on which number is right — and burn an hour relitigating a value that was settled months ago, except it wasn't, because nobody wrote down who settled it. Migrations that quietly drop business rules because the rule lived in someone's head, not in any record. AI efforts built on data no one can source or verify, where the model is asked to reason over a foundation nobody can defend.

None of these look like data problems in the moment. They all are.

The Monday test

Here is the check you can run on your own program this week. Pick one value your program argues about — the one that comes up in every review. Then ask three questions:

Who owns it? A named person or role, not a team, not a committee. If the answer is "everyone," no one does.

Where is the official source? One location, named explicitly. If there are two candidates, you don't have a source of truth — you have a disagreement waiting for a deadline.

How would you prove it? If someone challenged the value tomorrow, could you show where it came from, who approved it, and what has happened to it since? If the answer depends on one person's memory, the authority was never assigned. It was assumed.

Three questions. If your program can't answer them for the values it fights over, the fix isn't a new tool or a new dashboard. It's a decision — made once, written down, and enforced when the next duplicate shows up. Because it will show up. The next sensible local decision is already being made somewhere on your program, by someone who means well and has no idea they're creating a second source of truth.

Assign it before that happens. Or plan to spend another afternoon finding out which file is the right one.