Hunor.digital
← All writing
Strategy · Analytics honesty

Why your "single source of truth" dashboard keeps lying

The dashboard did not fix the disagreement, it centralized it. Why truth layers lie: colliding definitions, silent joins, and time zone crimes. And the fix that works.

Lázár HunorDigital Fixer
The short answer

A 'single source of truth' dashboard lies when its inputs disagree underneath: different definitions, different time zones, different dedup rules feeding one confident chart. Fix the metric definitions and the joins before the visualization, and publish the definitions next to the numbers, or the dashboard just centralizes the disagreement.

Somewhere in your company there is a dashboard that was supposed to end the arguments. It was probably even called that in the kickoff deck: one place, one truth, no more "whose number is right" meetings. And yet here you are, in the meeting, and marketing's revenue is not finance's revenue, and the dashboard shows a third number with total confidence.

The dashboard did not fail to end the disagreement. It laundered it. Here is how that happens, mechanically, and what actually fixes it.

A dashboard is a mouth, not a brain

Every chart is the last step of a chain: source systems, extraction, joins, definitions, aggregation, display. The dashboard only performs the final step. When people call it a source of truth, they are attributing to the mouth what was decided much further up the throat. If two upstream systems disagree, the chart does not resolve the disagreement; it silently picks a side, and its confidence is typographical, not epistemic.

That is the core lie of the genre. Everything below is just the specific ways it happens.

The five ways truth layers lie

  1. Definition collisions. "Revenue" is gross in the shop, net of refunds in finance, and net of tax in the warehouse query someone wrote in 2023. All three are correct by their own definition; the dashboard shows one without saying which. The same word, wearing three different numbers.
  2. Time zone crimes. The shop closes its day in UTC, the ad platform in your account's local time, finance at midnight Berlin time. Any daily comparison across them is systematically off, worst on the exact days (sales, launches) when people stare hardest.
  3. Join fan-out and dedup drift. Orders joined to sessions multiply rows when a match is not one-to-one; customers deduplicated by email here and by customer ID there produce different counts of the same humans. Nobody sees the join, so everybody blames the chart.
  4. Attribution baked in silently. The revenue-by-channel widget uses some attribution model. Which one? The dashboard rarely says, and the ad platforms' own numbers use different ones, which is why GA4, Facebook and your store never agree. A model choice presented as a fact is an opinion in a suit.
  5. Decay without alarms. The pipeline broke in March, a column changed meaning in May, and the dashboard kept rendering beautifully. Wrong data does not look wrong. It looks like a slightly surprising quarter, and by the time someone investigates, three decisions already shipped on it.

Why "more dashboard" never fixes it

The instinctive response to a distrusted dashboard is a better dashboard: new tool, new charts, this time with the good BI vendor. But every failure above lives upstream of visualization, so the new tool imports the old lies with better fonts. This is the build-vs-buy dashboard question's dirty secret: the tooling debate is a comfortable proxy war, fought to avoid the boring work of agreeing what the words mean.

What actually works

None of this is glamorous, which is exactly why it works and why it is skipped.

The honest reframe

Stop asking for a single source of truth. Ask for a single source of definitions, with numbers that carry their assumptions visibly. That is achievable in weeks, not quarters, and it ends the arguments not by hiding the disagreement but by making it inspectable.

If your dashboard is currently in the "confident but contradicted weekly" phase, the fastest exit is a tracing exercise: take the three numbers leadership quotes most, trace each from chart to source, and write down every definition and join along the way. It fits in a week. I do it as a standard audit, and I have never once traced three metrics without finding at least one silent lie, which is precisely why it is worth doing before the next quarter gets planned on top of one.