The Meeting Where Two Systems Disagree
The asset register says the server was decommissioned in March. Monitoring says it served four thousand requests this morning.
Someone asks the question that launches a thousand projects: which one is the source of truth?
It feels like the responsible question. It is actually the wrong one — and asking it is how organisations end up spending years, and fortunes, chasing a record that never arrives.
Because the honest answer nobody in the room wants to hear is: both are. The register is telling the truth about the decision. Monitoring is telling the truth about the traffic. Neither is wrong. They are answering different questions and being forced to share one row.
The Golden Record That Never Arrives
The Single Source of Truth is the project that is permanently six months from done.
Every organisation has one running. A programme to reconcile the systems, retire the duplicates, and crown one authoritative record that everyone will finally trust. It is always underway, always nearly there, always quietly slipping.
It slips because it rests on two assumptions, and both are false:
- that there is one right answer waiting to be found, if we just clean the data hard enough
- that everyone will agree to defer to it once we do
You can pour years and a fortune into the first and the register will still disagree with monitoring — because you were trying to legislate agreement between two things that are each telling the truth. And the second assumption never survives contact with reality: the moment one system is declared the winner, everyone whose truth got overwritten quietly reopens their own spreadsheet.
Truth Has More Than One Author
The reason the golden record never arrives isn’t poor discipline or the wrong platform. It’s that different parts of the organisation are legitimately authoritative about different facets of the same thing.
In Truth vs Turf I called these lenses on one graph: Operations owns current state, Enterprise Architecture owns intended state, the PMO owns the money, security owns the posture. Each is right about its own facet. The turf war starts when every lens insists its view is the view.
A Single Source of Truth doesn’t end that war. It just picks a winner and overwrites the losers. And the overwritten truth doesn’t vanish — it goes underground, into side files, footnotes, and the caveats people add out loud in meetings because they no longer trust the record to carry them.
A single source of truth doesn’t end the turf war. It just decides whose truth has to go underground.
Stop Declaring Truth. Compose It.
Here is the shift the whole series turns on.
The mistake is believing the authoritative view is a thing you store. It isn’t. It’s a thing you compose — assembled the moment you ask, from whatever layers have something to say, each contributing only what it is authoritative about.
Nothing is negotiated up front. There is no golden record to keep in sync, because there is no single record at all. When two layers speak to the same question, a precedence rule decides which one wins — set once, applied the same way every time, and visible in the answer.
The authoritative answer to “is this server live?” isn’t kept in any one system. It’s composed, on demand, from the layers that know — with a known rule for who wins when they overlap.
A Worked Example: The Network Says Yes, Security Says No
Take a single service endpoint.
The network layer knows it’s up and routable — packets reach it, the port answers. The security layer knows a policy blocks that port for anyone outside a named group. Ask “can a user reach this service?” and you get two answers that are both entirely true.
In the old world, this is a fight. Which system is the source of truth — the network monitor or the policy engine? You either argue it out, or you anoint one system the winner and bake the other’s reality into a footnote. Either way, someone’s truth is now wrong on the record, and the person who knows it’s wrong stops trusting the record.
In a composed world, you don’t choose. Both layers keep their own truth intact. The answer to the composite question — can a user actually reach it — is composed, with a precedence rule that says the security layer wins on reachability. Result: reachable, no.
And the answer arrives carrying its reasoning: network says yes, security overrides, here’s the rule. You didn’t get a flattened fact you have to take on faith. You got a resolved view that shows its work.
Precedence Beats Consensus
This is the quiet superpower, and it’s worth saying plainly.
Consensus is a meeting you hold forever. Precedence is a rule you set once.
You no longer need everyone to agree on the truth. You need them to agree only on who wins where they overlap — a far smaller agreement, and a far more durable one. Reachability defers to security; cost defers to finance; intended state defers to EA. Each of those is a one-time decision, not a standing argument.
And because the rule is explicit, disagreement stops being a failure to suppress and becomes information. When a lower-precedence layer contradicts the winner, that contradiction is visible — a signal worth looking at, not an error to reconcile away. The record stops hiding its own seams.
Why This Is the SIB in Miniature
Knowledge Isn’t a Story made the case that we don’t need perfect data — we need credible knowledge, information we trust enough to act on right now. Composition is how you get it.
A composed view is current by construction: there is nothing to keep in sync, because it’s assembled at the moment you ask. And it’s trustworthy by construction: it shows the layers it drew from and the rule that resolved them. That is the core move of the Service Intelligence Base — and almost everything else in this series is a consequence of it. History, accountability, commitments, even modelling the future — each turns out to be composition pointed at a different question.
Once you stop trying to declare the truth and start composing it, the reconciliation programme doesn’t get easier. It disappears.
The Challenge
Pick one fact two of your systems disagree about right now — the register versus monitoring, the CMDB versus the cloud console, the plan versus the actual.
Resist the reflex to ask which one is right. Ask instead: what is each of them genuinely authoritative about, and which should win where they overlap? Write that precedence rule down.
What would you stop reconciling?
You’ve just composed your first authoritative view — and you may notice you never needed the reconciliation meeting at all.
If this changed how you look at your next “single source of truth” programme, I’d love to hear it — comment, follow, or share it with whoever owns the reconciliation project that’s always six months from done.