Patent evidence and implications for the plan

Applied invention

The official USPTO extract covers utility grants and pregrant publication events through December 2025. In the 2015–2025 event window, records with at least one Arkansas inventor contain 4,653 utility grants and 6,025 published application documents. Explicit US application serials, checked against the source crosswalk, identify 7,288 distinct applications in that union. Of those, 3,376 appear in both series.

These are event-window observations. They are not a filing cohort, unique invention families, a grant-conversion rate or commercial results. Withdrawn grants are excluded. No application link in this cohort is missing or conflicts with the source crosswalk.

Participation and addresses

Among 4,653 grants, 4,033 have an identified assignee organization. The largest source assignee record appears on 1,142: 28.3% of the organization-assigned group and 24.5% of all grants.

Source assignee IDs can split a real organization; one normalized university name maps to two IDs in this cohort. The 755 IDs are therefore not 755 verified independent organizations. Corporate families have not been merged.

Excluding the largest record changes the 2019 count from 647 to 329. This shows sensitivity to concentration. It does not establish a cause, a corporate-family trend or broad statewide invention growth or decline.

For the same 4,653 Arkansas-inventor grants, 2,377 have at least one Arkansas assignee address; 1,636 have only known non-Arkansas assignee addresses; 37 have some unknown assignee geography without a recorded Arkansas address; and 603 have no assignee rows. These categories partition the cohort. They do not establish headquarters, current ownership, loss of value or local benefits.

Inventor-fractional counts retain all named inventors in the denominator, including unknown locations. Locations are document-specific.

Historical AI

For grant events in 2015–2023, all 3,858 eligible utility grants with Arkansas inventors match the corrected ECOPATAI/2023_r1 file. The three any-AI model flags (predict50_any_ai, predict86_any_ai and predict93_any_ai) identify 1,146, 912 and 838 grants respectively. Their shares are 29.7%, 23.6% and 21.7%.

These are alternative thresholds. They are not additive groups, and they are not observations of current AI use. The file was released 3 February 2026 and classifies historical documents through 2023.

The model and the Census business-use survey measure different populations and periods, so their rates cannot be subtracted to estimate an adoption gap. The pregrant overlay has ten eligible documents with no match in the file; these are left unclassified.

Technical subjects and classification

Analyst title screening found subjects including silicon-carbide power modules, field soil observation, food-safety tracking, polymer composites and retail checkout systems. These selected examples illustrate subjects in the records. They are not representative prevalence estimates, proof of deployment or validation of patent claims, and they do not identify a willing partner or usable data.

The CPC code baskets for the seven areas overlap and are candidate sets for review. Of 395 grants matched by the general agricultural A01 prefix, 209 include the plant-breeding A01H prefix, so calling all of them precision agriculture would be incorrect. In 95 grants, current H10 classification appears alongside issue-time H01L; changing classification versions can move the same documents between baskets.

Differences between broad and narrow definitions, or between current and issue-time classification, do not measure industry size, growth or the ranking of an area. Health and service improvements can matter without patent protection. These counts do not measure family or earliest-priority trends, comparable citations, current ownership or commercialization.

Use in the plan

These observations strengthen the case for applied partnerships and shared AI methods, and they make participation and access more specific questions to examine. The plan uses patents as context. Supported projects still need an agreed operating test, evidence of use and a check of who benefits.

This evidence does not change the seven growth areas, the existing research figures, proposed owners, dates or targets. No partner agreement is inferred.

Sources and reproducibility

Sources are the USPTO Open Data Portal products PVGPATDIS, PVPGPUBDIS, PVANNUAL and ECOPATAI/2023_r1, acquired 27 October 2026, with official dictionaries, release metadata and source hashes. The original analysis was followed by a sensitivity analysis, which hashes its inputs and outputs and records seven passed accounting checks.

Title and CPC screening was carried out by the analysis team. It has not had external subject-matter review. Inventor identifiers and individual records are not published.

Official product catalogs: grants, pregrant publications and the historical AI dataset.