A number is usually unpublished because it is expensive to collect
Unpublished-ness and cheap collectability are in tension. Testing publication rate before hunting incumbents kills bad candidates in two hours instead of two weeks.
- How it was measured
- Derived from four candidates that survived every other check and then failed on collectability. Now run first, with pre-committed thresholds.
- Evidence current as of
The appeal of an unpublished number is obvious: nobody has it, so assembling it creates something new. The trap is equally obvious once stated and almost never stated.
A number is usually unpublished BECAUSE it is expensive to collect.
Unpublished-ness and cheap collectability are in direct tension, and any viable project lives in the narrow band where both somehow hold. Most candidates that look like they are in that band are not — they are simply expensive, and the expense is what has kept everyone else out.
The cheapest possible kill
Sample forty sources and measure what fraction publish the number without a phone call.
| Hit rate | Verdict |
|---|---|
| 50% or more | The series is buildable. The moat is the tedium. |
| 25-50% | Marginal. Model the collection cost honestly before committing. |
| Under 25% | Dead — and dead for the right reason. |
Two hours and no money, against a full competitive sweep. Run it before anything else, because a candidate that fails here fails regardless of how good the rest looks.
Screen decay early too
A second cheap kill: how fast does the number change?
Two candidates died on this alone. One repriced roughly 6% a year, another annually — meaning a six-month-old figure sits inside sampling noise. An annual step function cannot support a refresh product. You update once and the data sits inert for eleven months, and nobody comes back.
What you want is continuous decay, where being three months stale is visibly worse than being current. That is what makes someone check again.
And when it does clear, coverage is not the credibility unit
The best-known published figure in one market we examined rests on 282 responses, a 7% response rate, three years of age, and 93 downloads. A competitor advertising weekly refresh had left 98.6% of its pages untouched for four months, with its flagship page rendering "may be outdated".
Nobody is auditing any of it. Which means the bar is not coverage:
"44 of 44 businesses in this named market, verified on this date" is more defensible than "23,000 properties", and it is what nobody currently offers.
Census-completeness in a named market with a verified-on date. It is cheaper than scale and worth more.
Tracked in our ledger as OPACITY-PARADOX, CREDIBILITY-IS-CENSUS. If a number here is wrong, tell us and we will correct it in place and say what changed.