Funded and Cleared, Still Unfindable: NIH Device Awardees After 510(k) (2026)

TL;DR

A founder who has an NIH small-business award and a 510(k) clearance often treats those two documents as the public record of the company. They are not. They are two US administrative files with two jobs. A foreign buyer, a hospital procurement team, a distributor, or a national device register uses a different identity, a different field, and often a different legal person. Funding plus clearance can be complete, and the name the rest of the world would search can still fail to join.

VayoMed Research computed four independent public lanes, accessed 22 August 2026. The NIH lane is a device-related extract of project rows coded to SBIR/STTR activity codes R41–R44: 6,196 rows, 2,140 unique normalized organization names, $2,972,573,910 in award amounts, every org_country equal to United States. That extract is already a small-business slice, not the whole of NIH RePORTER [1][2]. The FDA lane is the downloadable 510(k) file exported 22 July 2026: 175,559 records, 31,763 unique normalized applicant strings, of which 11,464 records and 4,797 unique applicants fall in 2023–2026 [1][3]. The USAspending lane is federal contracts in four medical-device NAICS codes: 9,787 awards, 974 unique normalized recipients, $37,638,513,903 in obligations. That is a buyer file. It is not the SBIR market, and it must not be narrated as one [1][4].

A normalized-name intersection of those files produces five headline counts:

• NIH organization ∩ 510(k) applicant = 336
• NIH (fiscal year 2020 or later) ∩ 510(k) decisions 2023–2026 = 60
• USAspending recipient ∩ 510(k) applicant = 263
• NIH ∩ USAspending = 17
• Triple overlap = 13

Read 336 as a deterministic exact normalized-name result, not as a census or a statistical bound. The revised matcher applies Unicode folding, uppercases names, strips punctuation and removes legal suffixes such as Inc., LLC and GmbH. It deliberately retains substantive words such as MEDICAL, MED, HEALTH and HEALTHCARE and rejects empty or generic-only keys. The result can still contain false positives and false negatives because it does not resolve subsidiaries, acquisitions or stable entity identifiers. Named scores stay private. Published figures are aggregates [1].

Those 336 strings were then tested for presence in ten national public device registers. Under this key, Israel AMAR matches 63 (18.8%), Brazil ANVISA 29 (8.6%), Singapore HSA 28 (8.3%), Indonesia 13 (3.9%), Mexico 2, Argentina 1, and UAE, Bangladesh, Ecuador and Malaysia 0. The recent-60 subset is thinner: Israel 7 (11.7%), Brazil 6 (10.0%), Singapore 5 (8.3%) and Argentina 1. Singapore’s result changed materially when both registrationHolderNameLocal and productOwnerNameLocal were retained; using only the first non-empty field had silently discarded product-owner identities. These remain name-presence results, not customs findings. NIH and the national registers do not share a stable entity identifier [1][5][6].

A 510(k) order clears a device for commercial distribution in the United States. It does not approve it, and it is not an establishment registration, a device listing, a Unique Entity Identifier, or a foreign registration [3][7]. NIH Phase I and Phase II awards fund research and development. They are not Phase III commercialization, and NIH is generally not the purchaser of the resulting product [8][9].

This paper does not rescore websites. The July 2026 commercial-readiness audit already did that work on 60 apparent first-time non-U.S. 510(k) applicants: official domains for 56 of 60, the newly cleared product term on 27, a K-number on three [10]. The adjacent August papers measured AI-recommendation panels and press-release citation neighborhoods [11][12]. This report starts from funding + clearance + procurement + foreign public registers. The missing layer, for a company that already has the first two, is a managed public-evidence system — owned site, source-grounded pages, a monthly third-party record, monitored queries — not another grant and not another clearance. VayoMed runs that system as one $24,000/year Done-for-You engine. Cadence is contracted. Coverage, rankings, leads and AI mentions are not [13][14].

1. Three US public files that do three different jobs

Takeaway: The NIH slice ($2.97 billion, 6,196 project rows), the 510(k) file (175,559 records) and the USAspending extract ($37.64 billion, 9,787 contracts) are not three views of the same market. They are a grant-coded R&D file, a marketing-authorization file, and a federal-buyer file. Treating any one of them as “who is commercially real” is a category error.

A commercial lead who has just closed an NIH award, or who has just received a substantial-equivalence letter, is looking at a document that answers a specific administrative question. The NIH file answers: did a United States small business concern receive a coded research award from an NIH Institute or Center. The 510(k) file answers: did FDA find this device substantially equivalent to a predicate for US commercial distribution. The USAspending file answers: did a federal awarding agency obligate dollars to this recipient under a medical-device NAICS code. None of those questions is “can a distributor in São Paulo, a hospital buyer in Singapore, or a retrieval system asked for ‘best X’ find a usable public identity.”

The rest of this section measures each file on its own terms, then states the job it cannot do.

1.1 The NIH lane is already an SBIR/STTR slice, not all of RePORTER

NIH RePORTER is the public search and download surface for NIH-funded projects [2]. Activity codes are the agency’s shorthand for award mechanism. R43 is SBIR Phase I. R44 is SBIR Phase II. R41 is STTR Phase I. R42 is STTR Phase II. NIH’s own activity-code table is unambiguous on the point: these four codes exist to help eligible United States small business concerns bring scientific innovations to the marketplace, with STTR adding a required nonprofit research-institution partner [15][8].

Our extract, stamped 10 June 2026, is already restricted to device-related projects inside those four codes. It is not a download of every R01, every U-series cooperative agreement, or every intramural project. A separate cross-agency SBIR award compilation covering NSF, DoD and the other participating agencies was not available for this analysis. R41–R44 are the substitute. That is a limitation, not a silent upgrade. This paper therefore cannot speak for America’s Seed Fund as a whole; it can speak for an NIH-coded, device-related small-business slice [1][9][16].

The 6,196 rows split as follows:

SBIR codes (R43 + R44) are 5,372 rows, 86.7% of the slice. STTR codes (R41 + R42) are 824 rows, 13.3%. That mix is consistent with how the two programs are used: SBIR is the default small-business R&D vehicle; STTR is the smaller, partnership-required sibling [17][18][19]. NIH SEED — the office that now brands NIH’s small-business programs — states the statutory purposes in the usual four-part list: stimulate innovation, use small business to meet federal R&D needs, encourage participation by undercapitalized firms, and increase private-sector commercialization of federally funded R&D [8][20]. Commercialization is a purpose. It is not a field in RePORTER, and it is not a 510(k).

Unique normalized organization names in the slice: 2,140. Award-amount sum: $2,972,573,910. Mean award per project row: about $479,800. Mean per unique organization, if one naively divides the sum by 2,140: about $1.39 million. That second mean is a descriptive ratio, not a typical check size. A Phase I and a Phase II for the same firm both count; a firm with one modest R43 and a firm with a stack of R44s are not the same object. SBA’s current guideline amounts, which NIH SEED republishes, sit near $323,090 for Phase I and $2,153,927 for Phase II, with Institute-level variation and a Commercialization Readiness Pilot cap published separately [8][21][22]. The $2.97 billion figure is the sum of award amounts in this extract. It is not NIH’s entire small-business budget, and it is not industry revenue.

Every organization country in the file is UNITED STATES. That is a feature of the programs, not a sampling accident. SBIR and STTR awards go to eligible US small business concerns [15][9]. A later section will join these US names to foreign registers. The join is hard in part because the starting file is domestically scoped by design.

Device-ish titles — project titles matching a device-language heuristic inside this already device-related extract — number 1,606, or 25.9% of rows. The other 74.1% can still be device-related by Institute, by abstract, or by later product even if the title does not contain the heuristic tokens. We do not treat 1,606 as the “true device count.” We treat it as a reminder that title text is a weak classifier, and that this extract is already a filtered slice rather than a keyword hunt over all of NIH [1].

Fiscal-year volume in the extract is not a start-up census. It is a count of project rows. The series thickens after 2007 and runs through a full FY2023–FY2025, with FY2026 still a stub on a 10 June 2026 stamp:

FY2007–FY2026 account for 5,296 of 6,196 rows (85.5%). FY1985–FY2006, the long tail in the same file, account for the remaining 900 (14.5%). FY2020–FY2025 — a six-year window that a commercial lead might call “recent enough to matter” — sum to 2,108 rows, 34.0% of the extract. FY2023–FY2025 sum to 1,077. None of those windows is a count of unique companies. A firm can appear in several years. A year can contain Phase I and Phase II for different products.

Two program facts belong next to this chart rather than in a glossary.

First, Phase III is not an NIH SBIR award. SBA’s policy directive and NIH SEED both describe Phase III as commercialization financed by non-SBIR sources, including private capital and, in some agencies, follow-on government procurement. NIH is generally not the final purchaser of the technology [9][8][23]. A company that has “won SBIR” has not, by that fact, appeared in USAspending as a device vendor.

Second, NIH SBIR is not NSF SBIR. NSF brands its program America’s Seed Fund powered by NSF, funds across almost all technology areas except clinical trials and Schedule I substances, and uses a Project Pitch plus a different Phase I/II envelope [16][24][25]. NIH SEED sits inside HHS, funds life-science product development, and publishes Institute-specific budget lanes and clinical-trial-optional parent announcements [20][21]. Collapsing “SBIR companies” across agencies without an award file that actually spans agencies is how a missing folder becomes a fake market. We did not do that. We used R41–R44 and said so.

NIH also offers commercialization training — I-Corps at NIH, a Commercialization Readiness Pilot — that is adjacent to, not a substitute for, a public commercial identity [26][22]. Those programs can improve a firm’s customer-discovery notes. They do not write the product page a foreign register or a buyer will retrieve.

1.2 The 510(k) lane is a marketing-authorization file

FDA’s premarket notification, the 510(k), is the pathway by which most Class II devices — and some Class I devices — demonstrate substantial equivalence to a legally marketed predicate. A substantial-equivalence order clears the device for commercial distribution. FDA’s own consumer language is that it is inappropriate to call a 510(k) device “FDA approved.” Premarket approval (PMA) is a different pathway for higher-risk devices [3][27][7][28].

The public 510(k) database and the downloadable files are how a third party inspects that clearance record [29][30]. Our export is dated 22 July 2026. It contains 175,559 records.

Decision-year volume in the current export, 2015 through the 2026 partial year:

2023–2026 together are 11,464 records, 6.5% of the all-time file in this export, and 4,797 unique normalized applicant strings. The all-time unique-applicant count is 31,763. So about 15.1% of historical applicant strings have at least one decision in the current four-year window. That is a useful recent-activity cut. It is not “the number of companies in medtech.”

Applicant country in the same export is dominated by the United States, as expected for a US marketing-authorization file, but it is not a US-only file:

Non-US applicant rows are 29,063, 16.6% of the file. The NIH extract, by contrast, is 100% US organizations. A name join between NIH orgs and 510(k) applicants is therefore a join between US-coded award recipients and a global applicant column. Some matches will be US firms that also cleared a device. Some will be string collisions. The join section reports the exact result of the documented normalizer and does not treat it as an entity census.

Device class in the same export is mostly Class II, which is the 510(k) heartland:

Class III rows in a 510(k) file are not a secret PMA census. They are a reminder that the 510(k) table is a submission-history file, not a cleaned product catalog, and that class fields can be messy at the margin. PMA remains the separate premarket-approval pathway [27].

Clearance is also not establishment registration, and it is not device listing. 21 CFR Part 807 is the registration-and-listing regulation. FDA’s consumer page is explicit: registering an establishment or listing a device does not mean FDA has approved the establishment or the listed device [31][32][7]. A firm can be cleared, registered, listed, or any combination, and those are still US administrative states. They are not a row on Singapore’s SMDR or Brazil’s consultas.

Unique Device Identification is a third US identity layer again. FDA’s UDI system identifies devices from manufacturing through distribution; GUDID is the public device-identifier database. IMDRF’s 2013 UDI guidance is the international frame those national systems sit inside [33][34][35]. UDI identifies a device. NIH org_name identifies an award recipient. 510(k) applicant identifies a submitter. Those three strings are allowed to diverge. In this industry they often do.

1.3 The USAspending lane is a buyer file, not a grant file

USAspending.gov is the official public source for federal award spending. Recipients are identified, with limited exceptions, by a Unique Entity Identifier issued through SAM.gov. DUNS has been retired for this purpose. 2 CFR Part 25 is the UEI registration rule for entities doing business with the federal government [4][36][37][38][39]. The DATA Act and FFATA are why a contract obligation of this size is supposed to be visible at all [40][41].

Our extract is contracts — not grants — in four NAICS codes that Census defines as surgical and medical instruments (339112), surgical appliances and supplies (339113), electromedical and electrotherapeutic apparatus (334510), and in-vitro diagnostic substances (325413) [42][43][44][45][46]. Stamp: 10 June 2026. 9,787 contract rows. 974 unique normalized recipient names. Obligation sum $37,638,513,903.

Say the next sentence out loud before using the number in a slide: $37.6 billion is not the SBIR market. It is federal procurement in four device NAICS codes, dominated by the Department of Veterans Affairs and the Department of Defense, awarded in large part to incumbents that can fulfil federal contracts. Mean obligation per contract row is about $3.85 million. Mean per unique recipient, if one divides $37.64 billion by 974, is about $38.6 million. That is the opposite of a Phase I envelope.

NAICS mix inside the 9,787 rows:

Awarding-agency mix:

VA plus DoD are 8,824 contracts, 90.2% of the file. HHS — the parent of NIH — is 621 contracts, 6.3%. A medical-device NAICS contract awarded by VA is a hospital-system purchase, not a commercialization grant. NIH’s own small-business page is explicit that NIH is generally not the final purchaser [8]. USAspending HHS rows in this NAICS cut are still procurement, not RePORTER.

SBA’s SBIR policy directive does describe Phase III preference when an agency procures technology developed under SBIR, but that preference is a contracting rule, not an automatic row, and it is not how this NAICS extract was built [9][23]. We did not filter USAspending on “SBIR Phase III.” We filtered on device NAICS. The 974 recipients therefore include large instrument manufacturers that dwarf the NIH small-business cohort. That is why a later join of NIH names to USAspending names returns 17, not hundreds.

The identity lesson inside this file is the UEI. Federal award data can join because SAM.gov issues a 12-character identifier and USAspending stores it [37][38][47]. NIH RePORTER project rows, in the extract we used, are joined on organization name. FDA 510(k) rows are joined on applicant name. Foreign registers, as the next sections show, are often joined on holder, registrant, or product owner. The US procurement file has a real key. The name-join we are about to run does not.

Three files, three jobs:

A company can live in one column and be absent from the other two. That is the normal state, not a scandal. The scandal, when there is one, is treating the first two columns as a substitute for a public identity a buyer can use.

2. The join: 336 exact normalized-name matches, not a census of companies

Takeaway: A conservative exact name normalizer finds 336 NIH-organization strings that also appear as 510(k) applicants, 60 of them in a recent window, 263 USAspending-recipient strings that match a 510(k) applicant, 17 NIH–USAspending matches, and 13 names in all three files. Name match is not DUNS, UEI, FEI or SRN, so these are reproducible candidate matches rather than resolved legal entities.

Record linkage is an old statistical problem. Fellegi and Sunter’s 1969 model is still the theoretical backbone: compare fields, estimate match and non-match probabilities, classify pairs as links, non-links, or clerical review [48][49]. Census Bureau research on record linkage and string comparators exists because names are a bad key [50][51]. Jaro’s work on name comparison, and the later Jaro–Winkler adjustment, exist for the same reason [52]. We did not run a Fellegi–Sunter model. We ran a deterministic normalizer and we are telling you what that choice does to the count.

The normalizer: Unicode fold; uppercase; strip punctuation to spaces; remove legal-form tokens such as INC, LTD, LLC, CORP, CORPORATION, GMBH and AG; retain substantive industry words including MEDICAL, MED, HEALTH and HEALTHCARE; reject empty and generic-only keys. What remains is joined as an exact string.

336 is 15.7% of the 2,140 unique NIH organization strings and 1.1% of the 31,763 unique 510(k) applicant strings. 60 is 17.9% of those 336. 13 is 3.9% of 336 and 0.6% of 2,140. 17 NIH–USAspending matches are 0.8% of NIH orgs. 263 is 27.0% of the 974 USAspending recipients — a higher hit rate, which is what one would expect if federal device vendors are more often also 510(k) applicants than NIH small-business awardees are.

Those percentages are still string-match rates. They are not “the share of SBIR firms that commercialized.”

The revised matcher fixes the earlier over-stripping defect: MEDICAL, MED, HEALTH and HEALTHCARE remain part of the key, and empty or generic-only results are rejected. That removes non-entity debris such as AND. It does not turn a string match into an entity resolution. A legal-entity census would require UEI, FEI, historical DUNS, company number, or clerical review of the kind Fellegi–Sunter reserves for ambiguous pairs [48][38][53]. We did not do that review for publication. The 336 matches are therefore neither an upper nor a lower bound; they are the exact output of the documented rule. Named-account scores stay private.

False negatives remain. A firm that awards under “Acme Surgical” and files 510(k)s under “Acme Surgical Devices LLC” may miss because DEVICES is substantive, while subsidiaries, doing-business-as names and post-acquisition brands will also miss. NIH org_name is the award recipient at the time of the project. 510(k) applicant is the submitter at the time of the submission. Years apart, they need not be the same string even when they are the same economic group.

The recent-60 cut exists to stop a 1980s R43 and a 2024 K-number from looking like a current commercialization story. Fiscal year 2020 or later on the NIH side, decision years 2023–2026 on the 510(k) side. 60 of 336 (17.9%) meet both. The other 276 dual-status strings are historical on at least one side. Historical dual status is still dual status in the files. It is a weak proxy for “this company is in market now.”

The USAspending joins tell a different story because the denominator is different. 263 of 974 federal device-NAICS recipients (27.0%) match a 510(k) applicant string. That is the incumbent pattern: a firm that sells instruments to VA is likely to have, or to have acquired someone who has, a 510(k). The NIH–USAspending join of 17, and the triple join of 13, are the small residue of names that survive all three string matches. We will not name them. The finding is the size of the residue, not a league table.

What a failed join does not say:

• It does not say the NIH award was wasted.
• It does not say the 510(k) is invalid.
• It does not say the company has no revenue.
• It does not say the company is not selling abroad through a distributor whose local entity holds the registration.

What it does say: the public files a founder is proud of do not automatically produce the public files a foreign buyer or a register uses. The next section measures that gap on ten export-market registers.

One more identity comparison, because the US system actually has keys it is not using in this join. SAM.gov’s Unique Entity Identifier is the federal-award key [38][47]. FDA’s establishment identifier (FEI) and the public registration-and-listing search are the facility key [53][31]. UDI/GUDID is the device key [33][34]. None of those keys is stored as a joinable field on the NIH organization name we used, and none of them is the manufacturer string on every foreign register. The infrastructure problem is not that identifiers do not exist. It is that the files a commercialization story wants to traverse do not share one.

3. Export-register presence is identity infrastructure, not a customs conclusion

Takeaway: Of 336 NIH∩510(k) strings, Israel matches 18.8%, Brazil 8.6%, Singapore 8.3% and Indonesia 3.9% under this key. Singapore is not zero once both product-owner and registration-holder fields are included. Malaysia remains zero under this exact rule. Presence and absence are identity-field findings, not evidence about legal supply or customs activity.

A national device register is a legal list of products that may be supplied in that jurisdiction, usually in the name of a local holder, registrant, or authorized representative. IMDRF exists in part because those lists do not share a single data model [35][54]. We used ten public registers as a presence-rate test of the 336 strings. This is not a tutorial on how to register in those markets, and it is not a methodology paper on register pipelines. It asks a narrower question: does the normalized NIH/510(k) name appear in any available manufacturer, holder or product-owner identity field.

Presence of the 336 dual-status strings:

The recent-60 subset:

Israel is the high-water mark at 18.8%, and even that is fewer than one in five dual-status strings. Brazil matches 8.6%, Singapore 8.3% and Indonesia 3.9%. Four registers match zero.

The honest interpretation requires looking at which field each register actually fills. The 336 names were matched against all available manufacturer, holder and product-owner identity fields. The architecture of those fields is the finding.

Singapore and Malaysia are not empty markets. HSA’s public Infosearch surface covers the Singapore Medical Device Register for Class B, C and D and a separate Class A listing. SMDR rows carry product owner and registrant, not a dedicated manufacturer field [5][55][56][57]. Including both populated identity fields produces 28 Singapore matches; the earlier zero came from selecting only the first available field. Class A devices in Singapore are a notification/listing regime distinct from SMDR registration of B/C/D [55][57]. Malaysia’s MDA register is held in the name of a local establishment. The holder is the legal person the authority talks to [6][58]. Matching a Maryland LLC’s NIH organization name to a Kuala Lumpur holder name can fail even when a device is listed.

Israel’s relatively high match rate is consistent with a register that stores manufacturer names and with a 510(k) file that already contains 1,668 Israeli applicant rows. AMAR, the Medical Equipment Division of Israel’s Ministry of Health, registers devices and issues import permits [59][60]. Brazil’s ANVISA consultas surface is similarly manufacturer-rich: 10,768 unique manufacturer strings, 15,983 rows with manufacturer country code US in this extract, against 102,659 total rows [61][62]. Indonesia’s list likewise carries manufacturer country, including 6,686 US-coded rows [63]. They match 18.8%, 8.6% and 3.9% of the 336.

Argentina’s ANMAT, Mexico’s COFEPRIS, Ecuador’s ARCSA, Bangladesh’s DGDA and the UAE list are real authorities with real public or semi-public databases [64][65][66][67][68]. In this extract, Argentina, Mexico and Ecuador do not offer a manufacturer-name set we could join (unique manufacturer strings = 0); they offer holder names. One Argentina and two Mexico matches against 336 NIH strings are what a holder-side join produces when the NIH name is not usually the local registrant. Bangladesh has both fields and still matches zero of 336, on a small register (4,422 rows). UAE matches zero on 879 manufacturer strings. Small denominators and transliteration cut both ways: a miss can be absence, or it can be “Inc.” versus a Hebrew, Arabic, Spanish or Portuguese legal form the normalizer did not recover.

Two adjacent public lists were inspected as contrast and not used as primary presence lanes: Canada’s Medical Devices Active Licence Listing and Australia’s ARTG. They are mature, English-language registers with their own grain (licence versus inclusion; medicines mixed with devices on ARTG) [69][70]. Adding them as extra presence tests would have required a separate grain decision this paper did not make. The ten-register table is the claim.

What a zero must not be used for:

It is not a finding that the unmatched names do not export to these markets. A local registrant can hold the row in its own name. A Singapore Class A device can sit on a different list. A distributor can import under a licence that never mentions the NIH string.

It is not a finding of illegal exports. Unregistered supply is a regulated-market question for the authority and the importer. A failed name join in a research extract is not that question.

It is not a finding that HSA or MDA databases are empty. This extract contains 20,860 Singapore rows and 49,721 Malaysia rows. HSA contains 28 exact normalized-name matches under the complete identity-field rule; MDA has no matches under that same rule.

The operational reading is narrower and more useful. If a US founder’s public identity is “the name on the NIH award” and “the name on the 510(k),” that identity is only weakly portable. Portable identity in export markets is the name on the holder/registrant/product-owner row, plus the product name those databases actually store, plus the English-language commercial evidence a buyer sees before they ever open Infosearch. IMDRF’s UDI work is an attempt to give devices a shared identifier. It does not give companies a shared identifier across RePORTER, 510(k), SAM and SMDR [35][54]. Until those systems share a key, presence rates will keep looking like market-access scores when they are identity-infrastructure scores.

MDA and HSA have even begun to rely on each other’s reviews for some Class B/C/D devices — a 2025–2026 regulatory-reliance pilot that MDA has said continues via the verification route for SMDR-registered products [58][71]. Reliance between authorities does not create a join key between NIH org_name and MDA holder. Harmonization of review is not harmonization of public identity.

4. What the July website audit already showed — and what this paper does not repeat

Takeaway: The July 2026 audit measured owned websites for 60 apparent first-time non-U.S. 510(k) applicants. This paper measures funding, clearance, procurement and foreign-register presence for a different cohort. The two gaps stack. They are not the same gap. We do not rebuild the 60-site scorecard, the AI Recommends panels, or the PR-distribution tables.

In July 2026 VayoMed Research asked a website question: after a 510(k) decision is public, can a buyer actually find the company and the product. The cohort was 60 non-U.S. firms that appeared to receive a first 510(k) under their current applicant name. The method was domain resolution plus a bounded HTML audit (471 pages). The results, already published:

• 56 of 60 could be connected to an official domain
• 49 of those domains returned public HTML during the audit
• 46 of the 49 reachable sites exposed a contact or inquiry route
• the newly cleared product term was found for 27 of 60
• FDA or 510(k) language was found for 32
• a K-number was found for three [10]

That paper’s funnel from 2026 K-number decisions to the 60-site sample is also already public and will not be re-derived here. Apparent first-time is a name-normalized reading of the 510(k) applicant column; it is not proof of first clearance across a corporate group [10].

The present paper’s cohort is almost the complement. NIH R41–R44 organizations are United States small businesses. The July sample was non-U.S. first-time applicants. One study asks whether a newly cleared foreign manufacturer’s website carries the product. This study asks whether a funded-and-cleared US name appears in federal procurement and in foreign registers. A company can fail both tests: missing from SMDR under its NIH string, and missing its own K-number on its homepage. A company can pass one and fail the other: registered abroad under a distributor’s holder name, with a US site that never mentions the cleared product.

We are not going to pretend those are one finding. Stacking them is the point.

Two other August 2026 Deep Research pieces sit adjacent and are also not rebuilt. The AI Recommends paper measured 40 capped category panels and 188 brand reports: the stored “best ultrasound machines” panel is 14 strings against 620 unique 510(k) applicants in three diagnostic-ultrasound codes; category-answer citations resolve to a recommended brand’s own site 12.4% of the time [11]. The press-release paper measured what six campaigns actually produced on branded queries, including a 98.5% own-site citation share in that study’s sweep [12]. Google’s published rule for AI Overviews and AI Mode remains ordinary Search eligibility: indexed, snippet-eligible, people-first content; no separate application [72][73]. Those papers answer “what do models cite” and “what does a wire campaign produce.” This paper answers “which public administrative files share a name.” Linking them is honest. Merging their tables is not.

The coordination-tax paper rebuilt an in-house growth-engine cost from BLS medians at $220,888 for a 1.5 FTE configuration, against VayoMed’s published comparison that the traditional route runs $200,000–$300,000+ per year. That comparison is an estimate, not an audited universal cost, and most companies never spend the rebuilt figure; they assign fragments of the work to people who already have jobs [74][13]. We will reuse that pricing honesty in the next section. We will not rerun the wage build.

If you have already read the July audit, the new information in this article is the NIH–510(k)–USAspending–register join, the limits of exact name matching, and the field-architecture explanation for why a product-owner field must not be discarded. If you have not, the July audit is the owned-website half of the same commercialization gap, and it is a better use of twenty minutes than a second website scorecard in this URL.

5. The missing public-evidence layer, mapped to a $24,000 engine

Takeaway: The files in this paper do not publish a product page, a citable explanation of intended use, a monthly public record, or a monitored query set. That layer is a managed system. VayoMed’s Done-for-You engine is $24,000 a year for six services, including 110–180 source-grounded articles and a monthly press release whose guarantee is cadence and distribution. It does not guarantee leads, rankings, coverage or AI mentions.

Return to the founder who has the NIH notice of award and the SE letter. The public objects that exist are: a RePORTER row a specialist can find [2]; a 510(k) summary a specialist can find [29]; possibly a SAM record if the firm registered for federal work [37]. The public objects that usually do not exist yet, or exist in a form a buyer cannot use, are the ones the July audit already scored on a different cohort: a website that names the product, states clearance language accurately (cleared, not approved), puts a K-number where a verifying buyer would look, and offers a contact path [10][3]. Add, from this paper: a foreign-register row in the local holder’s name; English-language pages a distributor can forward; a third-party URL that is not the FDA database.

Those objects are not produced by another R43. NIH Phase III is explicitly not an NIH-funded phase [8][9]. They are not produced by another 510(k), which is a US marketing-authorization event [3]. They are not produced by waiting for USAspending to show a VA contract; 17 of 2,140 NIH org strings matched that file [1]. They are produced by operating a public-evidence system.

VayoMed’s offer is that system as one annual subscription, not as an “SBIR commercialization SKU” and not as a ranking product [13][14]. The six services, used here as the map from evidence gap to workstream:

Price: $24,000 per year, 12-month term, all six services, 40+ markets as a live-site claim [13]. The comparable traditional route is estimated at $200,000–$300,000+ per year. That range is a comparison estimate, not an audited cost of every company’s in-house stack. The coordination-tax rebuild landed at $220,888 for a 1.5 FTE configuration and said the honest limit out loud: most firms never spend it [74]. $24,000 does not buy 1.5 full-time equivalents of regulatory, clinical and MLR attention. Those stay with the client. Content follows the client’s review path; it does not replace it [14].

Two promises the evidence in this paper forbids:

• We will not promise that operating the engine will create an HSA row, an ANVISA registro, a VA contract, or a ChatGPT citation. Register rows are held by local legal persons under national law. Contracts are awarded by agencies. Model answers are non-deterministic generations [11][72].
• We will not redefine VayoMed as an SBIR consultancy because this week’s paper used RePORTER. The offer is the integrated growth engine. The research is a wedge and a proof of competence [13].

What the engine can be held to is operational. A site launches. Pages ship on a review path. A release goes out each month on contracted channels. Queries are monitored on dated snapshots. Those are cadence facts. They are the public-evidence layer the three US files do not build. For a company whose name still fails to join an export register, that layer is the work that is actually in front of the commercial lead — alongside, not instead of, the regulatory work of appointing a registrant and filing in each market.

If the only need is to read the public AI-visibility reports, those reports are already on the site at no charge [11]. If the need is to operate the loop, that is the $24,000 conversation. The NIH notice and the SE letter remain necessary. They have never been sufficient as a findability strategy.

Frequently asked questions

Does an NIH SBIR award mean my device is FDA cleared?

No. NIH SBIR/STTR activity codes R43/R44 and R41/R42 fund research and development at eligible US small businesses. A 510(k) is a separate FDA premarket notification that, if successful, clears a device as substantially equivalent to a predicate [15][3][8]. In this extract, 2,140 unique normalized NIH organization strings join to 510(k) applicant strings in 336 cases under the documented exact rule. That is not a legal-entity census [1]. Clearance is also not approval [7].

Why is $37.6 billion the wrong denominator for an SBIR company?

Because it is federal contract obligations in four medical-device NAICS codes — 9,787 awards, 974 normalized recipients — not small-business grant outlays. VA awarded 6,442 of those contracts; DoD 2,382; HHS 621. Mean obligation per recipient in this cut is on the order of $39 million. NIH SEED states that NIH is generally not the final purchaser of SBIR-funded technology. Phase III commercialization is not an NIH SBIR award [1][4][8][9][43].

If my company name is missing from HSA Infosearch or MDA, does that mean we cannot sell in Singapore or Malaysia?

No. The complete identity-field rule finds 28 of 336 normalized names in HSA and none in MDA. A miss means only that NIH org_name did not exactly join to the available product-owner, registrant or holder strings. Singapore’s SMDR (Class B/C/D), its separate Class A listing and Malaysia’s register are populated databases; a local registrant can hold the row in a name that never appears in RePORTER [1][5][6][55].

Is 336 the number of NIH-funded companies that have a 510(k)?

No. 336 is the number of normalized name strings that appear in both files after punctuation and legal-suffix removal. Substantive words such as MEDICAL and HEALTH are retained, but acquisitions, subsidiaries and name changes remain unresolved. The figure is the exact result of this rule, not a legal-entity census or statistical bound. A census would need UEI, FEI or clerical review [1][48][38][53].

What should I do after 510(k) if buyers still cannot find us?

Treat clearance as the US marketing-authorization event it is, then build the public-evidence layer the authorization file does not contain: product pages that use accurate clearance language, a contact path, a K-number a verifying buyer can reconcile, a repeating third-party record, and whatever local holder identity each export market actually requires [3][10][31]. The July audit showed the website half of that job on a 60-company sample. This paper shows the register-identity half. VayoMed’s $24,000/year engine operates the digital half as cadence; it does not file your foreign registrations and it does not guarantee leads or rankings [13][14].

How is this different from the July website audit and the AI Recommends paper?

Different cohort, different files, different question. July scored owned websites for 60 non-U.S. apparent first-time 510(k) applicants [10]. The AI Recommends paper measured capped model panels and citation hosts [11]. This paper joins NIH device-related small-business project organizations to 510(k) applicants, USAspending vendors and foreign registers, and stops at aggregates. It does not rebuild those two studies.

Why don’t NIH, FDA and USAspending already share a company key?

USAspending can join on SAM.gov’s Unique Entity Identifier because federal award policy requires it [36][38]. FDA establishments have FEI numbers; devices have UDI/GUDID entries [53][33]. NIH RePORTER project records, as used here, present an organization name. Foreign registers often present a local holder. IMDRF has harmonized UDI guidance for devices, not a global company identifier across grant, clearance and registration systems [35]. Name matching is what researchers do when those keys are not on the file.

Does VayoMed guarantee that a funded-and-cleared company will show up in AI answers or in a foreign register?

No. The Done-for-You subscription is $24,000 a year for six operational services, including 110–180 source-grounded articles and one monthly press release whose guarantee is production and distribution cadence, not coverage, traffic, leads, rankings or AI mentions. Foreign-register rows are held by local legal persons under national law. 40+ markets is a live-site claim about where the engine is operated, not a promise of 40 registrations [13][14][12].

Methodology and limitations

NIH lane. Device-related NIH project extract stamped 10 June 2026. 6,196 rows with activity codes R44 (3,227), R43 (2,145), R41 (439), R42 (385). Unique normalized organization names 2,140 after the same normalizer used in the joins. Award-amount sum $2,972,573,910. org_country is UNITED STATES on every row. Device-ish titles 1,606 by title heuristic. Fiscal-year counts as tabulated in section 1. Public surfaces: NIH RePORTER and NIH activity-code definitions [2][15]. A separate cross-agency SBIR award compilation was not available; R41–R44 are the substitute. This is not all of NIH RePORTER and not NSF/DoD SBIR [16].

510(k) lane. FDA downloadable 510(k) files, export 22 July 2026. 175,559 records; 31,763 unique normalized applicants; 11,464 records and 4,797 unique applicants with decision years 2023–2026. Country, class and year cuts as tabulated. Public surfaces: FDA 510(k) landing page, 510(k) database, downloadable files [3][29][30].

USAspending lane. Medical-device NAICS contracts extract stamped 10 June 2026. 9,787 rows; 974 unique normalized recipients; $37,638,513,903 obligations. NAICS 339112 / 339113 / 334510 / 325413. Agency counts as tabulated. This is procurement, not grants, and not “the SBIR market” [4][36][43].

Joins. Deterministic exact match after Unicode fold, uppercasing, punctuation stripping and legal-suffix removal. MEDICAL, MED, HEALTH and HEALTHCARE are retained; empty and generic-only keys are rejected. The output is 336 exact normalized-name matches, not an entity census or a statistical bound. Recent dual status: NIH fiscal year ≥ 2020 and 510(k) decision year 2023–2026, yielding 60. Named matches are not published.

Registers. Ten national public device-register extracts accessed 22 August 2026 for presence rates only. Every available manufacturer, holder and product-owner identity field was included. Singapore has no dedicated manufacturer strings in this projection but has both product-owner and registration-holder fields; retaining both yields 28 matches, including 5 in the recent subset. Malaysia has no dedicated manufacturer strings and its holder field yields no exact matches. Canada MDALL and Australia ARTG were inspected as contrast and not used as primary presence lanes [69][70]. Warning-letter and MAUDE corpora were inspected and rejected as off-question. EDGAR company files were inspected as a public-company contrast and rejected as a primary lane: listed companies are not the SBIR cohort.

What we did not do. We did not resolve UEI, FEI, historical DUNS or national company numbers. We did not run probabilistic record linkage [48]. We did not audit websites for the 336 (that would have duplicated July on the wrong cohort). We did not rebuild AI Recommends panels or PR pickup tables. We did not name awardees. We did not interpret a missed register join as unregistered or illegal supply.

Offer facts. Taken from the live VayoMed site and company profile: $24,000/year Done-for-You engine; 110–180 source-grounded articles; monthly PR cadence and distribution, not coverage; 40+ markets as a live-site claim; traditional-route comparison $200,000–$300,000+/year as an estimate [13][14][74].

Compute date. 22 August 2026. NIH and USAspending extracts carry a 10 June 2026 stamp; the 510(k) export carries a 22 July 2026 stamp. FY2026 NIH rows (22) are a stub. 2026 510(k) rows (1,764) are partial through 22 July.

Public bulk surfaces used as lineage, not as a second compute. NIH Institute pages (for example NIEHS) document the same R41–R44 activity codes used in the extract [75]. SBA’s public award search is the cross-agency SBIR/STTR browser we could not substitute for the missing compilation; it is cited here as the surface a reader would use to inspect awards outside NIH [76]. NIH ExPORTER is the bulk administrative download behind RePORTER project files [77]. USAspending’s Award Data Archive is the corresponding bulk contract download [78]. FDA’s device-regulation overview sits above the 510(k), PMA, and registration-and-listing pages already cited [79]. The European Commission’s UDI page records the EU implementation of the same IMDRF identifier concept used in the identity discussion; it is not an additional presence-rate register [80]. HSA Infosearch itself splits the public medical-device search into a Singapore Medical Device Register listing and a separate Class A database — the field-architecture finding in section 3, restated at the official split [5][81][82]. Indonesia’s public alat-kesehatan consultation portal is the device-list surface behind the ministry landing already cited [63][83]. Malaysia’s establishment registration-submission guidance (MeDC@St / verification route) is the holder-side process MDA pointed to when it continued HSA reliance [58][84]. Brazil’s open-data and consultas stack is the ANVISA public query already used for manufacturer-rich presence rates [61][62][85].

Conclusion: the award and the letter are not the record a buyer uses

Four findings survive the method limits.

First, the three US files are three jobs. $2.97 billion across 6,196 NIH small-business project rows is an R&D story. 175,559 510(k) records are a US marketing-authorization story. $37.64 billion across 9,787 device-NAICS contracts is a federal-buyer story dominated by VA and DoD. Mixing the third into a slide titled “the SBIR market” is a category error.

Second, dual status is limited under a conservative exact matcher. 336 NIH∩510(k) strings, 60 recent, 13 in all three files. Publish 336 as the output of the documented normalization rule, not as a headcount of commercialized firms or as an upper/lower bound.

Third, export-register presence of those strings is identity infrastructure. The complete rule finds 18.8% in Israel, 8.6% in Brazil, 8.3% in Singapore and none in Malaysia. HSA and MDA store product owner, registrant and holder identities; NIH stores an organization name. A failed join is not a regulatory finding, and a successful 510(k) is not an SMDR row.

Fourth, the missing layer is still a public-evidence system. The July audit showed that even a reachable website often omits the product and the K-number. This paper shows that the administrative files above that website do not repair the omission. The work is owned pages, source-grounded articles, a monthly public record, and monitored queries, operated as cadence, alongside the regulatory work of local holders. That is the $24,000/year engine. Appearance in a foreign register, a VA catalogue, or a model shortlist is a possible downstream observation. It is not the deliverable.

If you have the award and the letter, you have completed two necessary US processes. The buyer’s next search will not be RePORTER. Build the identity they will actually use.

Talk to VayoMed about the Done-for-You engine — $24,000 a year, six services, cadence guaranteed, outcomes not.

---

Sources

1. VayoMed analysis of NIH RePORTER device-related SBIR/STTR project records (stamp 10 June 2026; 6,196 rows, activity codes R41–R44), FDA downloadable 510(k) files (export 22 July 2026; 175,559 records), USAspending.gov medical-device NAICS contracts (stamp 10 June 2026; 9,787 awards), and selected national public device-register extracts (presence-rate joins accessed 22 August 2026). Method in Methodology. NIH public search: NIH RePORTER.

2. National Institutes of Health, RePORTER (Research Portfolio Online Reporting Tools Expenditures and Results), accessed August 2026.

3. U.S. Food and Drug Administration, Premarket Notification 510(k) (a substantial-equivalence order “clears” the device), accessed August 2026.

4. U.S. Department of the Treasury / Bureau of the Fiscal Service, USAspending.gov (official source of federal spending data), accessed August 2026.

5. Health Sciences Authority (Singapore), Infosearch (public search across HSA-registered health products, including medical devices; SMDR versus Class A listing are different regimes), accessed August 2026.

6. Medical Device Authority, Malaysia, MDA (national medical-device regulator; public register held in the name of a local establishment), accessed August 2026.

7. U.S. Food and Drug Administration, Is It Really ‘FDA Approved’? (registration/listing is not approval; 510(k) devices are cleared), accessed August 2026.

8. NIH SEED, Understanding SBIR and STTR (phase structure; NIH generally not the final purchaser; published budget guidelines), accessed August 2026.

9. U.S. Small Business Administration, SBIR/STTR Policy Directives (Phase III is commercialization with non-SBIR funds; agency implementation rules), accessed August 2026.

10. VayoMed Research, After 510(k), Can Buyers Actually Find You? A MedTech Commercial-Readiness Audit, 21 July 2026.

11. VayoMed Research, AI Recommends Life Sciences 2026: 40 Capped Category Panels and the Citations Behind Them, 17 August 2026.

12. VayoMed Research, Medical Device Press Release Distribution Benchmarks 2026: What Six Campaigns Actually Produced, 10 August 2026.

13. VayoMed, Done-for-You Global Growth Engine for Life Sciences ($24,000/year; six services; 40+ markets as a live-site claim; PR guarantee is cadence and distribution), accessed August 2026.

14. VayoMed, Content Engine for Life Sciences (110–180 source-grounded articles per year; client review path), accessed August 2026.

15. National Institutes of Health, Activity Codes (R41, R42, R43, R44 definitions), accessed August 2026.

16. U.S. National Science Foundation, America’s Seed Fund powered by NSF (NSF SBIR/STTR; not the NIH extract used in this paper), accessed August 2026.

17. National Institute of Diabetes and Digestive and Kidney Diseases, R43/R44: Small Business Innovation Research Grant (SBIR), accessed August 2026.

18. National Institute of Diabetes and Digestive and Kidney Diseases, R41/R42: Small Business Technology Transfer Grant (STTR), accessed August 2026.

19. U.S. Small Business Administration, SBIR or STTR? Which One Is Right for Me?, accessed August 2026.

20. NIH SEED, SEED home (NIH small-business programs and commercialization support), accessed August 2026.

21. NIH SEED, SBIR and STTR Funding Opportunities, accessed August 2026.

22. NIH SEED, Commercialization Readiness Pilot (CRP) Program, accessed August 2026.

23. U.S. Small Business Administration, About SBIR and STTR (America’s Seed Fund coordinated by SBA across participating agencies), accessed August 2026.

24. U.S. National Science Foundation, About America’s Seed Fund powered by NSF, accessed August 2026.

25. U.S. National Science Foundation, What We Fund (clinical trials generally out of scope for NSF SBIR), accessed August 2026.

26. NIH SEED, I-Corps at NIH, accessed August 2026.

27. U.S. Food and Drug Administration, Premarket Approval (PMA), accessed August 2026.

28. U.S. Food and Drug Administration, 510(k) Program transcript (CDRH Learn), accessed August 2026.

29. U.S. Food and Drug Administration, 510(k) Premarket Notification database, accessed August 2026.

30. U.S. Food and Drug Administration, Downloadable 510(k) Files, accessed August 2026.

31. U.S. Food and Drug Administration, Device Registration and Listing, accessed August 2026.

32. Electronic Code of Federal Regulations, 21 CFR Part 807 — Establishment Registration and Device Listing, accessed August 2026.

33. U.S. Food and Drug Administration, UDI Basics, accessed August 2026.

34. U.S. Food and Drug Administration, Global Unique Device Identification Database (GUDID), accessed August 2026.

35. International Medical Device Regulators Forum, UDI Guidance: Unique Device Identification (UDI) of Medical Devices (IMDRF/UDI WG/N7, 18 December 2013), accessed August 2026.

36. USAspending.gov, Analyst’s Guide to Federal Spending Data (recipient identification; UEI), accessed August 2026.

37. U.S. General Services Administration, SAM.gov (System for Award Management; Unique Entity ID issuance), accessed August 2026.

38. U.S. General Services Administration, Check Entity Status (UEI is a 12-character identifier issued by SAM.gov), accessed August 2026.

39. Electronic Code of Federal Regulations, 2 CFR Part 25 — Universal Identifier and System for Award Management, accessed August 2026.

40. U.S. Congress, Digital Accountability and Transparency Act of 2014 (DATA Act), accessed August 2026.

41. U.S. Department of the Treasury, Federal Spending Transparency / DATA Act Information Model Schema, accessed August 2026.

42. U.S. Census Bureau, North American Industry Classification System (NAICS), accessed August 2026.

43. U.S. Census Bureau, NAICS 339112 — Surgical and Medical Instrument Manufacturing, accessed August 2026.

44. U.S. Census Bureau, NAICS 339113 — Surgical Appliance and Supplies Manufacturing, accessed August 2026.

45. U.S. Census Bureau, NAICS 334510 — Electromedical and Electrotherapeutic Apparatus Manufacturing, accessed August 2026.

46. U.S. Census Bureau, NAICS 325413 — In-Vitro Diagnostic Substance Manufacturing, accessed August 2026.

47. U.S. General Services Administration, GSA systems switch to Unique Entity ID in SAM on 4 April 2022 (UEI replaced DUNS for federal awards), accessed August 2026.

48. Fellegi, I. P., and Sunter, A. B., “A Theory for Record Linkage,” Journal of the American Statistical Association 64, no. 328 (1969): 1183–1210. Taylor & Francis record.

49. Vo, T. H., et al., “A simple two-step procedure using the Fellegi–Sunter model for frequency-based record linkage,” BMC Medical Research Methodology (2022). PMC9336505.

50. U.S. Census Bureau, Record Linkage & Machine Learning (Center for Statistical Research & Methodology), accessed August 2026.

51. Yancey, W. E., “Evaluating String Comparator Performance for Record Linkage,” U.S. Census Bureau Statistical Research Division working paper RRS2005-05 (2005), citing Winkler’s 1990 string-comparator extension of Fellegi–Sunter. Census working-paper page.

52. Jaro, M. A., “Advances in Record-Linkage Methodology as Applied to Matching the 1985 Census of Tampa, Florida,” Journal of the American Statistical Association 84, no. 406 (1989): 414–420. Taylor & Francis record.

53. U.S. Food and Drug Administration, Establishment Registration & Device Listing search (FEI and listing queries), accessed August 2026.

54. International Medical Device Regulators Forum, IMDRF home, accessed August 2026.

55. Health Sciences Authority (Singapore), Medical devices, accessed August 2026.

56. Health Sciences Authority (Singapore), Registration of medical devices, accessed August 2026.

57. Government of Singapore, data.gov.sg, Listing of Registered Medical Devices (SMDR snapshot; product owner and registrant fields), accessed August 2026.

58. Medical Device Authority, Malaysia, Approval issued by Health Sciences Authority (HSA) permitted by MDA for conformity assessments by way of verification process, accessed August 2026.

59. Israel Ministry of Health, Medical Equipment Division (AMAR; registry of medical supplies), accessed August 2026.

60. Israel Ministry of Health, Ministry of Health home, accessed August 2026.

61. Agência Nacional de Vigilância Sanitária (ANVISA), Medical devices, accessed August 2026.

62. ANVISA, Consultas — produtos para a saúde (public product consultations), accessed August 2026.

63. Ministry of Health, Republic of Indonesia, Kementerian Kesehatan (national health authority responsible for medical-device listing), accessed August 2026.

64. Administración Nacional de Medicamentos, Alimentos y Tecnología Médica (ANMAT), ANMAT, accessed August 2026.

65. Comisión Federal para la Protección contra Riesgos Sanitarios (COFEPRIS), COFEPRIS, accessed August 2026.

66. Agencia Nacional de Regulación, Control y Vigilancia Sanitaria (ARCSA), Ecuador, ARCSA, accessed August 2026.

67. Directorate General of Drug Administration (DGDA), Bangladesh, DGDA, accessed August 2026.

68. United Arab Emirates Ministry of Health and Prevention, MOHAP, accessed August 2026.

69. Health Canada, Medical Devices Active Licence Listing (MDALL), accessed August 2026.

70. Therapeutic Goods Administration (Australia), Australian Register of Therapeutic Goods (ARTG), accessed August 2026.

71. Health Sciences Authority (Singapore), HSA e-Services (entry point to Infosearch and related device searches), accessed August 2026.

72. Google Search Central, AI features and your website (AI Overviews, AI Mode; same technical and helpful-content requirements as Search), accessed August 2026.

73. Google Search Central, Creating helpful, reliable, people-first content, accessed August 2026.

74. VayoMed Research, What It Costs to Run Your Own Growth Engine, 27 July 2026 (BLS rebuild; $200,000–$300,000+ comparison estimate).

75. National Institute of Environmental Health Sciences, SBIR/STTR Grants (R41, R42, R43, R44), accessed August 2026.

76. U.S. Small Business Administration, SBIR/STTR award search, accessed August 2026.

77. NIH RePORTER, ExPORTER data catalog (bulk project files underlying public RePORTER searches), accessed August 2026.

78. USAspending.gov, Award Data Archive / download center, accessed August 2026.

79. U.S. Food and Drug Administration, Overview of Device Regulation, accessed August 2026.

80. European Commission, Unique Device Identifier (UDI) (EU implementation of IMDRF UDI concepts), accessed August 2026.

81. Health Sciences Authority (Singapore), Singapore Medical Device Register (SMDR) public listing (Class B, C and D; product owner and registrant), accessed August 2026.

82. Health Sciences Authority (Singapore), Class A Medical Device Database (separate from SMDR), accessed August 2026.

83. Ministry of Health, Republic of Indonesia, e-Regalkes — registrasi alat kesehatan (medical-device marketing-authorization portal), accessed August 2026.

84. Medical Device Authority, Malaysia, MDA public portal (device-status check and MeDCASt registry entry points), accessed August 2026.

85. Agência Nacional de Vigilância Sanitária (ANVISA), ANVISA portal (Portuguese public home; consultas and open data linked from here), accessed August 2026.

Original analysis: NIH R41–R44 device-related project counts, 510(k) applicant and year/country/class cuts, USAspending agency and NAICS cuts, name-normalized joins, and foreign-register presence rates computed 22 August 2026 from public administrative files. Chart data files published alongside this report preserve every displayed series. Internal working paths are not part of the reader-facing record.