Inside the vault
11 databases. Each one dated, sourced and traceable to the file it came from.
Open: method published
Census Archive
203 Census and BLS source files, 2015 to 2025, each dated and fingerprinted. The ground every other database stands on.
What is in it →Entrepreneur Map county database
3,142 counties, each set beside its 60 most similar peers and rated on up to 23 business types.
Method →Opportunity Index
Version 0.1: 33,314 scores across 2,946 counties, built from five parts that are always shown.
Method →Leisure Wage™
The hourly wage a household needs in each of 3,144 counties for basic costs, savings and a modest leisure budget.
Method →Announced Projects
812 announced projects of 100 or more jobs. Every source dated, no row ever deleted, saved in checksummed snapshots. Documented, not complete.
The Map Brief →Named Employers
Employers with 500 or more workers, each with its source and grade. Arkansas pilot: 37 employers so far. Documented, not complete.
Method →Startup Industry Database
910 business concepts in 14 groups, each with the cost of a first year.
Open the Startup Lab →Demand Map inputs
For 23 business types, what tends to open when a large employer comes to town.
Open the Demand Map →Startup Library
Research dossiers and startup kits by business type, tailored to a city and state.
See the Library →Held: named, not opened
Judgment Models
Four frozen models for weighing a business idea, and the database behind them. Never published.
Follow one number
Take a figure on a county page: the median household income in Pulaski County, Arkansas. Here is its whole trail.
- 1. PublishedU.S. Census Bureau, American Community Survey 5-year estimates, 2019–2023, table B19013, Pulaski County (FIPS 05119): $60,385, margin of error ±$1,648.
- 2. Held in the vaultFile
acsdt5y2023-b19013.dat, 17,884,943 bytes, fetched October 31, 2026. SHA-256 fingerprint:c9007017b7bb29ba758c77612ec9a4a0b897a4ed931d527754629866f580a110 - 3. Reshaped, never alteredOne panel row: vintage 2023, county 05119, variable B19013_001, estimate 60385, margin 1648. Nothing filled, adjusted or rounded.
- 4. Put to workThe Peer-County Method compares it with the 60 most similar counties, and the Opportunity Index sets it against the county’s Leisure Wage to score spending power.
- 5. Shown to youOn the Pulaski County page, with its source and the date we read it.
The six tests every database passes
Hold the evidence
Public data moves. Census restates old years, renames industries and redraws counties. We keep a dated, fingerprinted copy of every file we use, so any figure can be traced to the file as it stood the day we used it.
Check a file yourself →Never fill a blank
When Census hides a small count, we say so on the page. We do not estimate it, smooth it or borrow a neighbor’s number. A gap in the data is information, and we show it.
Read the Map method →Compare like with like
A county is judged against the 60 counties most like it, not against a national average. A mountain town is not measured against a port city. That is the Peer-County Method.
See how peers are chosen →Show the grade
Every figure carries its grade: Official (as the government published it), Modeled, or GWM, a Groundwork Measure. You always know whose number you are reading.
See the Opportunity Index method →Rebuild it or do not ship it
A result has to be reproducible from stored files. We rebuilt the Entrepreneur Map for Arkansas, Texas and Connecticut from the vault and the live Map matched it exactly, byte for byte.
See the proof →Never delete history
A project that is delayed, paused or cancelled keeps its row and gets a status. Every change is a dated event, and the whole record is saved in checksummed snapshots.
See the Map Brief →What is in the Census Archive
| Source | Years | Files | Publisher |
|---|---|---|---|
| County Business Patterns Businesses with employees, by county and industry: counts, jobs, payroll, size classes | 2015 to 2023 | 9 | Census Bureau |
| Nonemployer Statistics One-person businesses, by county and industry | 2015 to 2024 | 10 | Census Bureau |
| Population estimates County population, three vintages (Census restates earlier years each time) | 2010-2019, 2020-2024, 2020-2025 | 3 | Census Bureau |
| American Community Survey, 5-year Population, age, household income, labor force, rent, schooling, earnings and occupations | 2015 to 2024 | 56 | Census Bureau |
| Business Dynamics Statistics Business openings, closings and job creation, 1978 to 2023 | 2023 | 1 | Census Bureau |
| Quarterly Census of Employment and Wages Private jobs, establishments and pay (county extract of the annual file) | 2015 to 2024 | 10 | Bureau of Labor Statistics |
| Local Area Unemployment Statistics County labor force and unemployment, annual averages | 2015 to 2025 | 11 | Bureau of Labor Statistics |
| Business Formation Statistics New business applications by county | annual file, fetched November 2026 | 1 | Census Bureau |
| LEHD LODES 8 Where people work and where they live (all jobs; county totals kept) | 2021 to 2023 | 100 | Census Bureau (LEHD) |
| NAICS concordances How industry codes changed in 2017 and 2022 | 2012-to-2017, 2017-to-2022 | 2 | Census Bureau |
All of it is published by the U.S. Census Bureau or the Bureau of Labor Statistics. The raw files run to about a gigabyte and are held in a private repository. The fingerprint list is public: download the manifest.
What we have proved
- We re-downloaded sample files from Census and matched their fingerprints to our copies.
- We rebuilt the Entrepreneur Map for Arkansas, Texas and Connecticut from stored files. The output matched the live Map exactly.
- We rebuilt the Map’s business-applications and jobs-and-commute layers for the same three states from stored files. They matched exactly.
What we publish and what we keep
Open
- Every source, year and file fingerprint
- Each method, its weights and its rules
- Known breaks in the data, written down
- How to check a file and how to cite us
Held
- The contents of the Feasibility Database and the Judgment Models
- Client work and client files
- Working files and outreach lists
Check a file yourself
- Download the manifest. Each row names a file, the Census or BLS web address it came from, its size, its SHA-256 fingerprint and the day we fetched it.
- Download the file from that web address.
- Compute its SHA-256 fingerprint (on a Mac or Linux computer:
shasum -a 256 filename). If Census has not changed the file, it matches the manifest. - If it does not match, Census changed the file after our copy. That is why the vault exists: our figures were built from the version recorded here.
What we did to the data, and what we did not
Raw files are never edited. The county panel only reshapes them into one layout: it fills nothing, adjusts nothing and estimates nothing. Where Census hides a value, the panel keeps the hiding flag.
Known breaks in the data
- County Business Patterns 2015 and 2016 carry about twice the rows of 2017 and later. County totals are consistent; Census does not say why the file changed.
- Industry codes changed in 2017 and 2022. We keep a code map and a list of 919 six-digit industries whose meaning did not change from 2012 through 2022. Industries that were split or merged are flagged, never divided.
- Connecticut moved from 8 counties to 9 planning regions at different times in different Census and BLS products. We do not convert between them; a table records which geography each source and year uses.
- American Community Survey 5-year windows overlap. The 2022 file covers 2018 to 2022, the 2023 file covers 2019 to 2023. Consecutive files are not independent years.
- 2025 unemployment is an 11-month average: October 2025 was not collected during the federal shutdown.
How we keep it current
Every quarter we check Census and BLS for new releases and re-check sample fingerprints. New files are added with their own date; older files are never overwritten.
Cite this
Groundwork Business Advisory. (2026). The Groundwork Data Vault: Census Archive manifest and code maps (version 2026-11-02) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23906283
Web page: https://groundworkbusinessadvisory.com/data-vault/
Use and license
The underlying Census and BLS data are public. Our manifest, code map and county panel are free to use for non-commercial work with credit to Groundwork Business Advisory. Commercial use, including resale or use in a paid product, is by license: info@groundworkbusinessadvisory.com. For chambers, economic development groups, lenders and newsrooms.
Permanent copy
The manifest, the code maps and a readme are also deposited with a permanent DOI on Zenodo, the open research repository: doi.org/10.5281/zenodo.23906283. Cite that record in papers, grant applications and news stories. It is licensed CC BY-NC 4.0.
Version history
| Date | Change |
|---|---|
| November 2, 2026 | Permanent DOI assigned through Zenodo for the manifest and code maps. |
| November 2, 2026 | First version. American Community Survey 2015 to 2022, Business Formation Statistics and LODES added to the vault. |
Groundwork Business Advisory. Plain and sourced.