← Back to the findings · Methodology v3
What counts as a small business?
There is no universal employee cutoff. A company’s federal small-business status depends on its industry, its affiliates, the applicable size standard, and the date and purpose of the determination. This study classifies public evidence; it does not make binding eligibility determinations.
The legal framework
The Small Business Act describes a small business concern as independently owned and operated and not dominant in its field. SBA regulations generally require a business concern to be organized for profit, with a place of business in the United States and qualifying U.S. operations or economic contribution. Registration is evidence about an entity, not an exemption from these requirements. See 15 U.S.C. § 632, 13 CFR 121.101, and 13 CFR 121.105.
For federal procurement, the contracting officer assigns a NAICS industry code to the procurement. The corresponding SBA size standard determines the threshold. A provider can be small for one procurement and other than small for another. FedRAMP certification itself does not assign one procurement NAICS code to every use of a product. See 13 CFR 121.402.
Selected current standards relevant to cloud and software services, checked October 30, 2026:
| NAICS | Industry | Average annual receipts ceiling |
|---|---|---|
| 513210 | Software Publishers | $47 million |
| 518210 | Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services | $40 million |
| 541511 | Custom Computer Programming Services | $34 million |
| 541512 | Computer Systems Design Services | $34 million |
| 541513 | Computer Facilities Management Services | $37 million |
| 541519 | Other Computer Related Services | $34 million |
These are selected industry standards, not a complete list; exceptions and other NAICS codes may apply. Historical code 511210 is retained for older software-publishing records. Standards and code definitions change, so current thresholds cannot establish historical eligibility. Source: 13 CFR 121.201.
Receipts generally use the average over five completed fiscal years, with special rules for newer concerns and acquisitions. Employee-based standards generally use the average number of employees for each pay period over the preceding 24 calendar months. Size includes affiliates: control, or the power to control, can matter even when not exercised. A subsidiary’s standalone revenue or headcount is insufficient. See receipts rules, employee rules, and affiliation rules.
How we look up a company
- Resolve the provider’s legal identity, aliases, website, and Unique Entity ID (UEI) from Marketplace exports. A brand name alone is not a reliable match.
- Query the public SBA Small Business Search. Accept an exact UEI, or normalized legal name plus matching website domain. Ambiguous candidates remain unaccepted unless a documented manual review resolves the identity.
- Inspect active registration status and NAICS-specific small-business representations. The selected scope is the primary cloud-related NAICS when present, otherwise all registered cloud-related codes. Merely finding a profile, a SAM registration, or a business serving small customers is insufficient.
- Cross-check available SEC annual financial data, official company claims, dated program histories, and known ownership changes. Every resolved provider receives registry and website screening attempts; some sources are unavailable or inconclusive.
- Preserve unknown or conflicting results. Absence from SBA search is never proof that a business is large.
SAM.gov is the federal entity registration system. Public API access requires an API key; this run did not use authenticated SAM API access. SBA search provides public representations and program information, but is not a complete or independently adjudicated census of small businesses. GSA’s Entity API documentation explains its access model.
An ordinary small-business representation differs from an SBA program certification such as 8(a), HUBZone, or WOSB. Neither should be confused with FedRAMP certification, which addresses a cloud offering’s security requirements.
Two reporting groups
Every chart, analysis table, and report uses the same two exhaustive groups:
| Group | Included underlying evidence |
|---|---|
| Documented or likely small | Matched small-business representations, or explicitly explained likely-small inferences |
| Documented not small, likely not small, or unknown | Evidence of other-than-small size, or unresolved, missing, ambiguous, and conflicting evidence |
Small-group percentage = (documented small + likely small) ÷ all eligible records. Second-group percentage = (documented not small + likely not small + unknown) ÷ all eligible records. Together they sum to 100% before rounding; an empty group has no percentage. Unknowns are never dropped from the denominator and are not legal findings of large size.
Each bar is a 100% composition, not a count scale. The label at right always uses the small-group numerator and the full row denominator. Both charts use the same colors and column alignment. Counts and percentages in tables come from the same calculated fields.
The original evidence tiers remain inside individual assessments and the JSON for audit, alongside sources, confidence, and reasons. They are not separate reporting groups. Legacy documentedPct and expandedPct JSON fields remain for backward compatibility; use smallGroupPct and remainingGroupPct for the two-group analysis. Medium confidence reflects a strong or probable identity match with unresolved legal context; low confidence reflects tentative self-description or insufficient evidence. High confidence is reserved for strong identity/control evidence and decisive financial scale; it describes confidence in a research inference, not a legal determination.
The standard automated screen considers the selected cloud/software NAICS codes. A documented manual review may identify a different industry that fits the actual service. Perplexity is an example: its exact-UEI SBA record reports small under 519290, Web Search Portals and All Other Information Services, whose standard is 1,000 employees, but does not list primary 518210 as small. The Marketplace describes web search and a Search API. We therefore count it as likely small in that service-specific context, without claiming it is small for every procurement. Sources: SBA record, Marketplace description, and size-standard table.
Bland AI’s self-description during a September 15, 2026 FedRAMP community session supports a low-confidence likely-small inference. Wisedocs has a matching legal-name SBA candidate and service agreement, but incomplete UEI/domain corroboration keeps it in the likely tier. Perplexity uses the service-specific context described above. Files.com is now unknown: its company disclosure identifies Action Verb LLC and majority ownership by Riverwood affiliates, while relevant receipts and affiliate scope remain unresolved. A small-team description does not resolve those questions. Current certified non-pilot 20x therefore has 3 of 8 providers (37.5%) in the small group. Historical testimony assessments remain separately preserved; current evidence is not backdated.
For SEC screening, annual revenue periods must be nonoverlapping, end within the preceding five years, and have been filed before the assessment date. Their sum is divided by five even when some years are missing. A result above $47 million—the largest selected current cloud/software receipts threshold—supports likely other-than-small only. GAAP revenue differs from SBA receipts, historical standards vary, and affiliation exceptions may apply. Current parent mappings do not establish historical ownership; the mapping audit is retained. SEC data documentation.
Conflicting registration and financial evidence is unknown pending review. An explicit reviewed decision may retain the conflict while finding likely-not-small when authoritative control and financial evidence is decisive. Known acquisitions can also invalidate a standalone representation. Ownership research is incomplete: “documented small” does not mean all possible affiliation concerns were independently checked.
Deeper evidence review
The October 30 research now records 218 distinct individually reviewed providers across successive passes, including a 50-record stratified audit in the first pass. There were 620 bounded official-site discovery attempts alongside existing registry and SEC screening. The remaining records retain screening-level evidence; this is not a claim of exhaustive investigation of every company. Research coverage overlaps and must not be added together.
The research ledger (CSV), full ledger (JSON), and before/after changes expose source links, confidence, reasons, review depth, failed page attempts, and remaining questions. The prior classifications are preserved in the repository. Across both deep-review passes, current listed unknowns changed from 399 to 253; the small group changed from 182 to 199 of 662. These changes reflect better evidence, not changes in Marketplace access.
The second pass is incomplete: 109 of the original 352 current unknown providers have been reviewed; 243 await this pass, and 10 remain unresolved after research. 1 historical-only provider was also reviewed. The second-pass progress file separates these states. “Awaiting this pass” can include earlier screening or an earlier inconclusive individual review; it does not mean no source has ever been checked.
We prioritize exact UEI or corroborated legal identity; current government representations; official control/acquisition documents; and actual annual financial disclosures. GSA records document size for the displayed contract NAICS, with an underlying representation date often unavailable. They do not establish universal eligibility or historical status. A government contract alone, or a payment amount, is not proof of small-business status.
A completed acquisition can change the affiliate group. Private-equity investment alone does not prove that every portfolio company is an affiliate. Actual annual revenue can provide a conservative scale bound: a verified $300 million completed-year figure contributes $60 million to a five-year average even if the other four years were zero. Accounting revenue remains a proxy for SBA receipts, and tax adjustments, currency conversion, the proper fiscal window, newer-company rules, and exceptions matter. ARR, bookings, funding, valuation, and headcount alone are not substituted for receipts.
The second pass uses a reasonable-likelihood standard, rather than requiring a complete legal size worksheet for every inference. Sustained, extreme commercial scale can justify likely-not-small even when only older actual revenue and current operating milestones are public. ARR can corroborate that scale, but is never converted into statutory receipts. A recent affirmative small representation that expired can support a qualified likely-small judgment when identity and continuing operations are corroborated. Older or indirect evidence lowers confidence; valuation and funding alone remain insufficient.
Industry context matters. Ceribell’s EEG and Butterfly’s ultrasound offerings combine medical devices with cloud functions. Their filings and, for Ceribell, an active SBA representation support likely-small judgments under the electromedical manufacturing context (334510, 1,250 employees). Neither finding claims eligibility for a separate cloud-only procurement or a verified 24-month affiliate-adjusted employee average. Sources appear in the provider ledger. Likewise, confirmed nonprofit operators fall in the second group because SBA generally requires a for-profit concern; this does not assert that those organizations are large.
Private corporate tax returns generally are not public. SEC filings, foreign corporate accounts, and nonprofit Form 990 filings may supply public financial evidence, but do not automatically resolve the correct entity, affiliation, or standard. Nonprofits also raise the separate for-profit eligibility requirement. Public self-descriptions can support a likely-small inference with explicit limitations; unsupported estimates remain leads.
Historical size and certification dates
Annual and rolling-window comparisons use the most recent size assessment available in the snapshot. The certification year remains the year of first observed certification. This describes the present assessed size of past cohorts, not their size when entering FedRAMP. Historical size evidence is retained in the downloads for audit. We never backdate today’s SBA size representation or today’s Marketplace flag. A dated SBA program interval spanning a certification date supports only a likely-small assessment because the procurement NAICS and intervening ownership remain uncertain. Dated company or reporting evidence is separately curated with an explicit applicable interval. Financial facts must have been available by the assessment date.
The all-years summary uses the same most-recent size evidence as the annual chart. The annual population selector distinguishes all ever-certified packages (including delisted history) from packages that are currently certified. All-years provider totals are deduplicated; summing annual provider rows can count one business more than once. With current evidence and the currently-certified population, all-years totals reconcile exactly with the current participation section.
The published view uses current size of historical cohorts. It answers “How are the companies from that cohort classified today?” and must not be described as size at entry.
Certification means the first recorded transition to FedRAMP Certified for a package ID. Each package contributes once to annual first-certification counts. A distinct provider can occur in multiple annual cohorts if separate products were first certified in different years. Within a cohort, its earliest product certification determines its historical size assessment. “First-time providers” are providers with their earliest observed package certification in that cohort. These records cannot prove there was no earlier participation outside the available data.
Reconciling the Marketplace files
The source is the FedRAMP Marketplace data repository. We retain an immutable commit snapshot, retrieval time, and SHA-256 hashes. Three sources are joined:
- The legacy
data.jsonsupplies provider names, products, UEIs, and original small-business flags. fedramp-frc-cso-pkg.jsonsupplies newer package/provider details. Where the exported retro package ID differs from a submitted ID, the export’s retro ID is the join key and the discrepancy is logged.fedramp-status-changelog.jsonsupplies dated status, pathway, and class transitions, including initial implementation and delisting.
Seven historical legacy snapshots from April–September 2026 recover additional identities. Repository history starts April 14, 2026; earlier status transitions are retrospective data, not contemporaneous snapshots. Transition dates and recording dates are preserved separately. Source files may reflect different update times.
Latest transition date, then recording date and event identifier, determines current status. Current certified includes certified offerings in remediation. Initial implementation is a preparation stage, not certification; see FedRAMP provider implementation guidance. Current comparisons use the latest pathway, while certification cohorts use the pathway recorded at that certification. “Rev5” is the source’s label even for historical events predating Revision 5; it should be interpreted as the traditional pathway series, not evidence that those earlier authorizations used Rev5 controls.
Pilot and non-pilot 20x are the source’s separate 20x Pilot and 20x labels. “Ever pilot” is separately retained; a current non-pilot record can have pilot history. No unrecorded pilot membership is inferred.
Providers are distinct resolved legal/provider identities. Conservative alias and UEI matching merges duplicate identities; subsidiaries are not automatically collapsed into parent groups. Packages are distinct FedRAMP package IDs, not necessarily unique software products. A provider with several packages counts once within a provider group, but may appear in more than one pathway. Therefore pathway provider counts are not additive.
Government-operated services are excluded from commercial participation rates and retained in the data. Unidentified packages remain in package denominators with unknown size, but cannot be counted as distinct providers. Original Marketplace small-business flags are published separately as source metadata, not reporting groups; they are not substituted for the evidence classification.
Comparing the past two years
The rolling comparison covers the 24 months through the snapshot date and the immediately preceding 24 months. Dates use inclusive starts and exclusive ends, recorded in the CSV. Annual tables flag the current year as partial. Class-level, pathway-level, package-level, and provider-level tables are downloadable to make different denominator choices inspectable.
What this study cannot establish
Public evidence is incomplete and missingness is unlikely to be random. Small firms may have no public registration; large private firms may have no public financial statements. Historical size and historical ownership are especially sparse. Present-day provider names can also differ from the owner at certification. The initial-implementation population is visible in newer records to a degree that older preparation activity was not.
This is an observational description of recorded participation, not a causal evaluation of FedRAMP reforms. It lacks the full applicant pool, unsuccessful applicants, costs, agency demand, and time spent before listing. Pilot selection, class differences, survival and delisting, acquisitions, and changes in reporting can all affect the apparent mix. A larger documented share could reflect better evidence rather than broader access.
The original USAspending bulk download for dated contracting-officer size fields failed. A later targeted award API lookup supplied recipient metadata for DSS, but no dated contracting-officer size determination; that metadata is not treated as such. GSA public contract records supplied additional NAICS-specific representations. Of 124 candidate contract pages, 55 were retrieved and 69 remained unavailable after a bounded retry. Missing responses are not evidence of size. Additional procurement records, date-specific SAM representations, and ownership histories would materially strengthen the historical analysis.
Updating or correcting the research
The repository contains raw source snapshots, sanitized public research observations, curated identity/ownership decisions, analysis scripts, tests, and this site. research/review.json separates human-reviewed evidence from generated output. Refresh source and research snapshots, inspect the changes, run analysis and checks, then publish a new dated version. See the README for exact commands and cache behavior.
To suggest a correction, open an issue with the provider, UEI if known, dated source, applicable NAICS, and any relevant parent relationship. Please do not submit private financial or personal information.
Independent research, as of October 30, 2026. This site is not an official FedRAMP, SBA, or contracting-officer determination.