Methodology
How an abstract is assembled
Federal agencies publish enforcement and filing records under the employer name and address each office happened to capture — the same firm appears as “ACME TRUCKING, INC.” in one file and “Acme Trucking LLC” in another. FirmStanding resolves those variants into one entity per firm and links every source record to it with an explicit match tier and confidence. Nothing is inferred beyond the match itself: the abstract reports what the records show.
Step one
Name & address normalization
Names are uppercased; punctuation collapses to spaces; ampersands become “AND”; a leading “THE” and trailing legal suffixes (INC, LLC, L.L.C., CORP, CO, LTD, and their stacked combinations) are stripped repeatedly. Addresses receive USPS-style abbreviation (STREET→ST, SUITE→STE, directionals), and ZIPs reduce to five digits. Trade names (DBAs) are normalized separately, so an entity is reachable by either its legal or trade name. Placeholder names (“N/A”, “UNKNOWN”) are treated as no name at all.
Step two
Deterministic match tiers
Records cluster into entities on exact agreement of normalized fields — there is no fuzzy or probabilistic matching in v1. Every link between an entity and a source record carries the tier that produced it; a dossier's overall confidence is its weakest link.
Normalized name + 5-digit ZIP agree
The strongest signal available without registry data: same name at the same ZIP.
Normalized name + city + state agree
Same name in the same city. Robust for distinctive names; weaker for chains.
Normalized name + state agree (distinctive names only)
Only names with two or more tokens that are not on the generic-name stoplist may match on name+state alone. This is the weakest tier and the one to scrutinize.
Generic names (“QUALITY CLEANERS”, “MAIN STREET DELI”, single-token names) are barred from Tier 3 and can only cluster when ZIP or city also agrees — a common name in the same state is not evidence of a common owner.
Read this before relying on a dossier
Known limitations
- Brand + generic-name risk at Tier 3. Names like “TEXACO GAS STATION” or “QUALITY CLEANERS” pass the two-token distinctiveness test yet may describe unrelated businesses in the same state. Tier-3 spot checks show matches dominated by legitimate multi-site employers, but residual false merges concentrate exactly here — which is why Tier 3 carries 0.60 confidence and why every dossier discloses its tier mix.
- No registry ground truth yet. v1 uses federal enforcement files only — no Secretary of State or UCC data — so corporate families, mergers, and successor entities are not modeled. Two legally distinct firms sharing a name and address may merge; one firm filing under different addresses may split.
- Entity IDs rotate. The entity layer is rebuilt from scratch after each source refresh; permalinks are the name+state slug, not the internal ID.
- Coverage windows differ. OSHA inspections are loaded from 2015 forward; WHD is full history; exclusion lists are current snapshots (terminated SAM exclusions drop out of the file); FMCSA carriers join only when they match a non-FMCSA record.
- Records are as-filed. Agencies correct and restate their files; a record here reflects the vintage shown, not necessarily today's docket. Absence of a record is not a clean bill — it may mean the agency never inspected, or the match missed.
Provenance
Source systems
Vintages are the source-file dates recorded by the ingest ledger, not the dates we happened to load them. Every API response repeats this attestation per section.