Application continuity and priority sensitivity

This separately defined mature published-application cohort uses ODP US filing dates 2005–2019, at least one 2005–2025 utility pregrant publication with an Arkansas inventor, and the full retained pregrant source rather than the earlier 2015–2025 event window. It does not include unpublished cases, establish residence at filing, count distinct inventions or determine legal patent families or priority entitlement.

The cohort contains 6,418 source documents and 6,406 US applications. Official selective-field ODP searches returned 6,406 cohort applications; 0 remain unmatched. Retained parent-expansion inputs cover 15,894 requested applications in 166 hashed batches. All API inputs retain canonical requests, acquisition timestamps and hashes; no inventor/contact fields were requested. The event window remains frozen through December 2025; continuity assertions are a separately dated October 2026 acquisition.

Policy sensitivity

What this changes in the report

This separately defined published-case cohort compares filing-programme composition across mature US filing years without requiring a 2015–2025 grant/publication event. It complements the original 7,288-application event-window baseline; the two denominators are not interchangeable or additive. Report application and lineage counts under their own windows and geography rules rather than treating every application or grant as a separate invention.

Keep continuations/divisions separate from continuation-in-part sensitivity: a CIP can introduce new matter, so connecting a CIP does not establish identical inventive content or entitlement to every ancestral date. Sharing a provisional also does not establish equivalent claimed subject matter. National-stage relationships enter only the fifth, broadest policy. Reissue, substitution, unclassified and contradictory typed relationships remain withheld from grouping. External PCT nodes retain their original identifiers; they are not converted into invented US serials.

The 2005–2019 US-filing-year table compares the composition of observed published cases with at least six calendar years before the December-2025 source cutoff. It includes older retained publications rather than requiring a 2015–2025 event. Lineages can span filing years, so yearly group counts are not additive. Local association is observed at a retained publication, not necessarily at filing or first national publication; unpublished/nonpublication cases remain outside the frame. Earliest observed benefit/foreign-priority dates can predate the US filing window, and do not establish complete priority cohorts or valid claim-specific priority. No innovation trend, grant-conversion rate, commercialization, economic impact or new ARA ranking is inferred.

The audit quarantines 72 contradictory relationship pairs across the retrieved edge set, of which 28 fall within the selected ancestral closure; 1 ancestral pairs have a parent date later than a child date and are also withheld. 13,408 collateral/later-descendant edges are excluded from grouping so later filings cannot bridge otherwise separate older programmes. PatentsView and ODP differ on observed filing dates for 414 selected applications; these source/date concepts remain separate. ODP-internal date conflicts remain unresolved. No parent bag means no parent relation was recorded in that retrieved response, not proof of no related application. Complete observed ancestry describes only the retrieved policy graph, not a legally complete family.

Ancestral records extend outside the event window and Arkansas. Their geography is not reassigned to Arkansas merely because a selected descendant has an Arkansas inventor. Unknown API matches, external parent references, missing dates, conflicting dates and cycle-affected regions remain explicit coverage limits.

Reproduce with a new output directory containing a copy of inputs/continuity-api and the same CLI without --acquire. A completed manifest refuses overwrite; use a new run ID for changed sources or policies. The initial v1/v2 diagnostics are preserved. V3 separates date semantics, quarantines contradictory/temporally inconsistent edges and restricts grouping to the selected ancestor closure.