Read a content report closely and you'll find two species of number living side by side, dressed identically.
The first was counted. Something observed an event and incremented a total. The second was asserted. Somebody, or some model, produced a figure and wrote it down.
Both render as a number in a box. Only one of them is a fact.
The distinction
Measured means a system observed the thing and counted it. Views. Likes. Comments. Spend. Impressions. Watch time. You can trace it to an event log. If you asked "where did this come from", there is a mechanical answer.
Reported means someone stated it. A creator's audience size on a rate card. An agency's "3× engagement lift". A tool's "estimated reach". An AI summary's "your engagement is up significantly". Nobody counted anything at the moment that sentence was produced: the number came out of an assertion, an estimate, or a model.
Reported isn't a synonym for false. Plenty of reported numbers are accurate. The point is that they carry a different burden of proof, and once the two are mixed in the same table, the burden of proof for everything on that table drops to the weakest item.
Where the two get mixed
Four places, in roughly descending order of how often it happens.
Screenshots. The most common one, and the most invisible. A screenshot of an analytics dashboard is measured data. But which metric, over what window, on what account, is entirely under the control of whoever chose the crop. The pixels are honest. The framing is a claim.
Summaries over data. "Your Reels are outperforming your posts." That sentence sits on top of measured data, but it is itself reported: someone chose the comparison, the window, and the word "outperforming". Every one of those choices is a judgement that the sentence doesn't disclose.
Language models. An LLM handed real analytics will happily write "engagement rose 34%". It didn't calculate that. It produced the most plausible-looking sentence given the surrounding text. This is the single sharpest version of the problem, because the output is fluent, confident, and formatted exactly like a computed result.
Rate cards and case studies. "180k followers." "We drove a 4× return." Nobody in the conversation is in a position to verify either. They're reported, they're often approximately right, and they should be treated as claims until something measures them.
The rule
This is the doctrine Acumin is built on. Numbers come from code: computed by the pipeline, from real fetched data. Words come from the model: the interpretation, the phrasing, the argument. Confidence comes from the data: how much of it there was, and how consistent. The model composes sentences around figures it is not permitted to originate, which is why it cannot invent a statistic even when a convincing-sounding one would fit.
That's a constraint, not a feature, and it costs something: sometimes the honest output is "there isn't enough here to say". That's the trade being made deliberately.
What to do about it
Ask "who counted this?" If the answer is "the platform", good. If it's "an agency", it's a claim. If it's "an AI wrote the summary", find the underlying figure before you act on it. It takes five seconds and it changes what you do surprisingly often.
Separate them visually. If you produce reports, put measured figures and interpretation in visibly different places. A section of counted numbers, then a section of what you think they mean. Mixing them into flowing prose is what makes the two indistinguishable to a reader.
Insist on the denominator. "Engagement up 40%" is nearly meaningless without knowing what it's 40% of. Forty percent of a base of five is two. A measured number without its denominator is functionally a reported one, because you can't check it.
Treat rate cards as claims. Not as lies. As claims. When a creator says 180k followers and 6% engagement, that's a starting point for a conversation, not an input to a spreadsheet. The Creator Network exists partly because of this: ratings there only come from brands who actually booked and paid for delivered work, which is the difference between a reputation someone asserted and one that was earned in public.
The uncomfortable part
Applying this properly will make your reporting less impressive. Numbers you were fond of will turn out to be reported, and the honest version of the sentence will be quieter.
That's the correct outcome. A team that knows which three of its numbers are real makes better decisions than a team with thirty numbers of mixed provenance, because the second team can't tell which ones to trust and so ends up trusting whichever one supports the plan they already had.
The quiet version is also more persuasive to anyone senior enough to have been burned before.
How to use this tomorrow
Open your most recent content report. Go through it line by line and mark each number M or R (measured or reported).
Then look at the decisions that came out of that report and ask which ones were resting on an R.
That's it. You don't need to remove the reported numbers. You need to know which ones they are.
Related: What your evidence is actually worth covers the other axis: where a measured number came from and what it's therefore able to prove.