Patent evidence: comparisons, connected filings and research-support links

Reviewed 28 October 2026. These findings extend the Arkansas patent baseline and leave the plan's seven growth areas unchanged. The plan cites a few of these observations; the tables here give the definitions and sensitivity tests needed to assess them. Patent records do not show partner needs, usable access or commercial results.

Main findings

Participation matters alongside volume. The apparent increase in grant events is sensitive to one large source assignee record, so who participates and who can use the capability need to be measured as well as how many documents exist.

The comparisons raise questions and have substantial scope limits. Food, logistics and retail/supply screens have above-national shares under the code variants examined. Expanded review of primary text finds both relevant applications and substantial mismatches. These results point to areas for further inquiry. They do not rank the seven areas.

Technical influence is close to the reference under alternative counting choices. Older Arkansas-connected grants have field/year citation averages near one, which does not establish an exceptional statewide citation advantage.

Specific support links are stronger evidence than broad commercialization claims. Two patent acknowledgments can be connected directly to verified NSF awards, including Arkansas EPSCoR ASSET III. They document historical support connections and sharpen questions about facilities and access.

Documents, applications and connected filing programmes

The 2015–2025 Arkansas-inventor event window contains 4,653 utility grants and 6,025 pregrant documents, representing 7,288 US applications. Selective-field USPTO searches matched all 7,288 applications. Ancestor expansion requested 12,172 applications and returned 12,144; 28 requested records were not returned. Links are explicit parent/child serials, not pairs inferred from flattened relationship rows.

These are application lineages that depend on the linking policy. They are not counts of independent inventions or legally established families. A continuation-in-part may introduce new matter, and sharing a provisional does not establish identical claims. Contradictory relationships are excluded, and later or collateral descendants cannot bridge the selected programmes. No ancestral inventor location is reassigned to Arkansas because a descendant has an Arkansas inventor. Continuity assertions were retrieved in October 2026; the document-event cohort covers events through December 2025.

A separate mature published-case cohort drops the event-window requirement. It contains 6,406 applications with ODP US filing dates in 2005–2019 and an observed 2005–2025 pregrant publication with an Arkansas inventor. Those applications are represented by 6,418 publications and touch 5,282 continuation/division lineages, or 4,818 under the broadest link policy. This is a separately defined denominator and should not be added to the 7,288-application baseline.

The year-by-year table describes observed published cases with at least six calendar years before the source cutoff. Unpublished cases are outside the frame. Publication geography does not establish residence at filing, and observed benefit dates are not claim-specific priority entitlement or complete priority cohorts.

Data: event-window lineage review · policy table · mature published-case review · mature filing-year table

National comparison and technical scope

The same release contains 3,545,902 nonwithdrawn utility grants with 2015–2025 events, of which 1,716,604 have at least one US50+DC inventor. Arkansas reconciles to 4,653 whole grants and 3,485.499 inventor-fractional grants; the domestic reference has 1,613,139.476 fractional grants.

Each regional fraction divides regional inventor sequences by all named inventor sequences, keeping foreign, unknown and territory inventors in the denominator. Whole state counts overlap. Arkansas and its six neighboring states use the same rules. No compatible population denominator was available, so this is not a per-capita performance comparison.

Technology-share ratios compare a provisional code screen's share of regional patents with its share in the domestic reference. They measure the composition of patents within each screen. They do not measure market opportunity or the full strategic area.

Food, transportation/logistics and supply/retail screens stay above one under both broad and narrow definitions and under both current and issue-time classification. Other results caution against literal interpretation. Power's narrow ratio changes from 1.01 at issue to 0.69 under current classification, and agriculture's narrow screen includes substantial non-precision activity. Equal-window changes can reverse direction between broad and narrow definitions for food and transportation. These ratios are supporting diagnostics.

The expanded technical review covers 90 primary documents across 91 area/document pairs: 84 stratified probability-selected pairs and seven separate purposeful challenges. Each area has four sampled narrow, four broad-only and four outside-broad documents. An analyst reviewed abstracts and first independent claims under stated technical rules.

The purposeful challenges found missed medical-adherence, indoor-drone-navigation, nanoparticle and food-sanitation applications. Sampled mismatches include asphalt aging within a health screen, organ cold storage and crop varieties within agricultural screens, and cancer therapy within broad food biotechnology codes.

The small outside-basket sample has low power to detect omissions. Selected-set yields are not population accuracy or recall estimates, and the review has not been checked by independent domain experts. Definitions were not changed after comparative results were seen. All seven areas and cross-cutting AI/ML, analytics and cybersecurity remain in scope.

Data: national/neighbor panel · provisional technology-share table · equal-window code-screen sensitivity · comparison and review methods · reviewed documents and reasons · selected-set yields

The panel's summed unknown-inventor credits are fractional document credits, not a missingness percentage. A source CPC row establishes that a classification was recorded. It does not validate the meaning of that code or its relevance to a strategic area.

Breadth, connections and concentration

Identified organizations appear on 4,033 of the 4,653 Arkansas-inventor grants. They carry 755 source IDs, including universities and government; these are not verified independent firms. 164 IDs appear in at least three event years, while 438 appear in one. The five largest IDs jointly occur on 1,632 unique grants. Similar names are flagged and are not merged into campuses, companies or corporate families. First appearance in this truncated grant window does not establish firm entry.

Equal five-year event windows show the sensitivity to concentration.

This exclusion is a sensitivity test. It is not a corporate-family correction, an explanation of cause or a measure of invention growth. The all-grant inventor-fractional comparison rises 5.2%; excluding the same source ID changes it to −9.3%.

149 grants name multiple source organization IDs. 1,971 have Arkansas and another US50+DC inventor location, and 370 include a foreign-country inventor. These overlapping observations identify possible routes for inquiry. They do not establish an operating partnership, current employment or usable institutional access.

Data: participation and selected recorded pathways · annual source-organization breadth · concentration sensitivity

Fixed-window technical influence

A score of one is the mean for the same grant year and first listed inventional CPC subclass at issue. This reference includes all US utility grants, including foreign-inventor grants, so it differs from the domestic regional-activity benchmark above. Relative scores are defined for 2,622 of 2,631 five-year grants and 3,413 of 3,423 three-year grants. Grants with missing fields or a zero expected mean have no score.

The top approximately one percent of documents account for about one fifth of the sum of relative scores. These descriptive means do not establish a statistically significant statewide advantage. The two rows have different eligible cohorts and cannot show change over time.

Citation timing uses the actual citing grant date, after the target grant and through its third or fifth anniversary. The source field named citation_date describes the cited document's month, so it is not used as the citation event date. Repeated citing–cited pairs count once, and applicant and examiner flags can overlap.

Excluding observed shared organization IDs produces whole-counted relative means of 1.15 and 1.17 respectively. Unknown assignee identities, corporate-family links and continuation relationships are not resolved in that test, so it cannot establish fully independent influence. Citations do not measure sales, licensing or adoption.

Data: fixed-window method and findings · citation comparison table · counting and tail sensitivity · detailed sensitivity table

Documented research-support connections

295 grants carry extracted government-interest statements and 285 carry parsed contract numbers. Where no statement was extracted, support is unobserved in this product; that does not prove the absence of public support.

US9087694, polycrystalline semiconductors. The document lists University of Arkansas trustees and Silicon Solar Solutions as source organizations and explicitly acknowledges NSF SBIR IIP-1248962. The official NSF record identifies Silicon Solar Solutions as the award recipient. This documents a specific university/company support connection.

US12139559, cellulose nanocrystals. The statement acknowledges OIA-1457888. The official NSF record identifies this as Arkansas's RII Track-1 ASSET III nanostructured-surfaces award. This provides a specific historical connection between materials research and EPSCoR.

These cases were selected because statements exist. They do not estimate prevalence or programme returns. Exact statements and award records take precedence over automated agency tags. In the semiconductor example, a parsed SBA tag follows the SBIR programme wording and does not establish SBA funding.

Selected assignment histories also refine the questions to ask. The Cree/Wolfspeed power-module case records merger and name-change instruments followed by bank security interests and releases. The food-process-control case includes security instruments as well. The latest recorded assignee labels therefore cannot be read as simple commercial transfers or local value loss. Recorded and executed dates, grant events and later 2026 case updates are kept separate. Recordation does not determine title or establish a license, sale or actual operating access.

Data: two research-support cases and official award links · selected recorded histories

Sources

The analysis retains official USPTO product catalogs, release fields, file hashes, dictionaries, selective-field API requests, acquisition dates and output indexes. New citation and government-interest files match the core release. Cohort records stop in December 2025. Live continuity and assignment data and selected NSF checks were acquired on 28 October 2026. Primary technical text is retained with hashes and canonical patent links. The analytical code and tests are retained, including the code versions used for the lineage analysis.

USPTO describes the application/continuity/document API, PatentsView methods and products, applicant/examiner citation sources and recorded security interests. The source records for NSF award 1248962 and NSF award 1457888 support the two examples above.

Patent records, publication activity and recorded support links cannot show a partner's defined need, usable data and facilities, an agreed comparison or accepted delivery roles. Those require direct responses from partners. Independent domain review could refine the technical inclusion rules.