United States
Data Methodology
This site is a classification reference, and its credibility rests on a short list of verifiable practices: definition text restated from the public manual with its vintage, figures drawn only from named public datasets with per-value provenance, and source limitations — including suppression — shown rather than smoothed over. This page documents those practices as they actually operate.
The definition layer: restated, vintaged, uninterpreted
The classification content on this site — the 621210 definition and its cross-references — is restated from the publicly published NAICS manual and stored as a sourced record with the manual vintage (currently the 2022 revision) attached. The definition panel on the definition and data pages renders from that record through one code path, so the quoted text and its citation cannot drift apart page by page. Editorial writing around the panel explains how to read and apply the classification, but the classification text itself is presented as the manual states it: this site does not paraphrase definitions into friendlier claims, extend boundaries by analogy, or rule on edge cases the manual leaves open. Where the manual is silent, the honest answer is that the manual is silent.
The authoritative source for the classification is the published manual, always. This site's role is a well-lit path to it — accurate restatement, stated vintage, and a link out. Any conflict between this site and the manual is resolved by the manual, and would be treated here as a correction.
The data layer: named sources, per-value provenance
Figures on the data pages come from a small set of public sources — Census Bureau business and population data, Bureau of Labor Statistics occupational wage data, and federal shortage-designation data — each the primary publisher for its measure. Provenance is stored per value, not per page: every data point carries its source identifier, the dataset's reference vintage, the ISO date of retrieval, its unit of measure, its geography level, any transformation note, and a deep link to the source table or API query it came from. Citations are generated from those records by a single shared formatter, which is why every figure's citation line has the same shape: source, vintage, unit, geography, retrieval date.
- Ingestion writes dated, versioned filesA script retrieves data from the named public sources and commits it to version-controlled files, stamping each value's full provenance record, including when it was pulled.
- Builds read only the committed filesPages are built from those files and never call live endpoints at render time. What a visitor sees corresponds exactly to a recorded retrieval — reproducible, diffable, and stable between visits.
- Derived values disclose their parentsThe few computed figures (such as establishments per 100,000 residents) are simple disclosed derivations whose citations name both parent sources. There is no modeling layer above the sources.
- Updates are deliberate eventsNew source vintages enter by re-running ingestion, which updates values, vintages, and retrieval dates together in version control. Nothing updates silently.
Honest absence: suppression as a first-class state
Public sources withhold values — for disclosure protection in small cells, or because a measure isn't published at a given geography. In this site's data layer those conditions are explicit states on the value itself: reported, suppressed with the source's reason, withheld, or not collected. The type structure obliges every rendering component to handle the non-reported states, which is a stronger guarantee than editorial care: there is no code path by which a suppressed value displays as zero or as a blank that resembles measurement. On the page, an unavailable value appears as a stated limitation with its citation, because 'the source withheld this' is itself sourced information worth showing.
Most data-quality failures are not lies; they are defaults — a join that fills missing cells with 0, a chart library that plots absent data as a baseline. Making suppression structurally unrepresentable as a number means the site's honesty doesn't depend on every future page author remembering the rule.
Boundaries: what this site refuses to publish
Standing refusals, so readers know what absence means
- No invented or modeled statistics — if a figure appears, it traces to a named public source
- No scores or rankings of businesses, markets, or geographies built on the classification
- No assignment of NAICS codes to specific named businesses, and no sale or endorsement of classified lead lists
- No presentation of the classification as a credential, quality signal, or legal status
- No silent updates — value changes arrive with changed vintages and retrieval dates
These boundaries are the methodology as much as the provenance mechanics are. A classification reference earns its keep by being reliably boring: the definition as published, the data as published, the gaps as they actually are, and a clear line where the code's authority ends. Corrections are welcome — a factual error in restated definition text or a mis-cited figure is treated as a defect and fixed with the update dates to show it.
Frequently asked questions
Where does this site's definition text for 621210 come from?
It is restated from the publicly published NAICS manual, stored as a sourced record with the manual vintage attached (currently the 2022 revision), and rendered through a single panel component with its citation. The manual remains the authoritative source; this site's job is accurate restatement with a stated vintage and a link out, not interpretation of the classification text.
Does this site calculate any of its own statistics?
Only simple, disclosed derivations from cited parents — for example, dividing an establishment count by a population estimate to show density, with both sources named where the figure appears. There is no modeling, interpolation, gap-filling, or scoring anywhere on the site. If a source didn't publish a value, the page says so instead of estimating one.
How do I verify a figure I found on this site?
Follow its citation: every figure links to the exact public table or API query it was retrieved from, and the citation states the vintage, unit, and geography level. Checking takes one click plus a comparison. If you find a mismatch between a figure here and its source, that is a correction-worthy defect in this site — the source wins by definition.
Why does the site show 'suppressed' instead of just leaving those numbers out?
Because absence has meanings, and hiding it picks one for the reader. A suppressed value tells you the source's disclosure rules were triggered — often because the underlying count is small — which is materially different from 'not measured' or 'zero.' Rendering the state with its reason and citation preserves that information and keeps the site's totals and comparisons honest about their coverage.
Will the data on this site change when NAICS is next revised?
A revision would enter the same way everything does: deliberately, through re-ingestion, with the new manual vintage stated on the definition record and any affected dataset figures updated alongside their own vintages. Because vintage is attached to every value and to the definition itself, a revision produces a visible, documented transition rather than a silent redefinition — and comparisons across the boundary would be flagged as cross-vintage.
Related on NAICS 621210
How we handle this information
We keep material limitations visible, separate advertising from editorial judgment, and avoid inventing live scores or recommendations when the underlying evidence is not available.
Related in this network
Related properties may share common ownership. A cross-property link is not an endorsement — see our ownership disclosures.