What does the public record actually say about this business?
1,293,701 rows of United States federal public records, held locally and served free: how many operators exist, what they earn, who is licensed where, what firms in the field actually bill, and how fast it is forming or dying. Every figure names its source. Every gap is stated rather than filled.
That band is not a design flourish. It is what happened when the same candidate was run through the same rubric ten times: 53 at the low end, 57 at the high end, standard deviation 1.6. It cost $4.76 to find out. A score that straddles a gate is not a verdict, and this is the only way to see that.
Start with the question you have
Which industry should I be in?
Screen 261 industries on operator count, licensing, wages and jurisdiction — ranked by the one signal backtested against 45 years of outcomes.
Is this specific deal ordinary?
Place a business by headcount and revenue against its industry, see the labour share its peers run, and read what the public record cannot check.
What is the record on this industry?
956 industries, every US jurisdiction — operators, wages, licensing, federal buyers, formation rate, and what firms there actually earn.
I have an idea, not an industry.
Describe it in a sentence. We match it to an industry and show the record — free — or underwrite the idea itself through 26 stages including an adversarial pass.
What is held, and where it came from
1,100,961 establishment records
Census County Business Patterns 2023 — how many businesses of each kind operate in each county, and their payroll.
27,067 occupational licences
CareerOneStop, US Department of Labor — 54 of 56 jurisdictions. Free and ungated here permanently, by agreement and by choice.
13,248 years of business dynamics
Census BDS, 1978-2023 — firm births, deaths and survival. This is what every backtest on this site is graded against.
Revenue for 970 industries
Census SUSB 2022 — firms by receipt size, so a business can be placed by revenue and not only headcount, and labour share of revenue becomes computable.
Why this is not like the other tools
The error bar is measured, not asserted
±3.1 at 95%, from running the same input 10 times. Not one competitor publishes one.
Predictions are frozen before the outcome
63 recorded with a metric, a threshold and a date, uneditable afterwards, and graded when they resolve.
We publish our own failures
70 defects, generated from the regression suite rather than written by hand — a defect reaches that page only after acquiring a test that holds it shut.
A dimension with no evidence is not scored
It returns nothing and the covered weight is published beside the score. A score built on 55 points and presented as 100 is the failure this whole system argues against.