Company-reported or model estimate
How to read the numbers in a report, and when to trust them.
Every figure in a TractionAtlas report carries one of two labels. It is the single most useful thing on the page, and the easiest to skip past.
Company-reported
The business published this number itself — on a pricing page, an open startup dashboard, an annual recap post, a conference talk, an interview.
These are precise and they are not neutral. A company publishes the number that flatters it. "10,000 users" may count trial sign-ups from three years ago; "$2M ARR" may be run-rate from the best month. We record what was said, where it was said and when, and leave the number as stated rather than adjusting it.
Read it as: accurate about what the company chose to disclose.
Model estimate
Derived from third-party sources — traffic providers, directory listings, app store ranks — rather than from the company.
These do not flatter anyone, but their error bars are wide, and they get wider the smaller the business is. Traffic estimates for a site with 2,000 monthly visits are close to noise; the same method at 500,000 visits is usually within a sensible range.
Read it as: reliable for direction and order of magnitude, not for a decimal place.
Why they are never averaged
A blended number would be worse than either input, because you would no longer know which kind of error you were holding. So a field shows one or the other, with its label, and when both exist they appear as separate fields rather than one reconciled figure.
The most useful comparison
Not "how big is this company" — the labels make cross-company comparison of absolute size unreliable, and you should distrust any chart that stacks them.
Instead: the same company across versions. Both labels are produced the same way each month, so the month-to-month change within one report is far more trustworthy than the absolute value is. That is the comparison the version history is built for.