Event-Dated US Trademark Prosecution Corpus: Dated Events, Five-Year-Proof Outcomes, Theme Model and Lead/Atypicality Scores

Derived data for An Event-Dated Corpus of US Trademark Prosecution and a Two-Sided Measure of Vocabulary Position (SSRN 7520758) and a companion paper on early entry, built from the complete USPTO trademark record. proof_outcomes.parquet: one row per registered serial: registration date, date of the first Section 8 / Section 71 cancellation, registration age at that date, and whether it falls in the five-year-proof window (4.0-8.5 years). Usable as a product-survival outcome without the text measure. scores_T50.zip: per-filing past- and future-facing divergence (K-, K+) and lead (topic_dkl) under the production scoring (50 themes, per-filing 1,826-day windows, filings 1995-2019), one parquet per Nice class. Atypicality = (K- + K+)/2. theme_model_T50.zip: the fitted theme model (scikit-learn LDA, joblib) with its 62,168-term vocabulary and metadata. case_events.parquet: 242 million dated prosecution events across 13.99 million USPTO case files (serial, code, type, date, sequence). case_extras.parquet: counsel of record, statutory filing basis, post-registration declaration flags, abandonment and status dates. event_code_dict.parquet: USPTO event codes with their modal descriptions. owner_links.zip: normalized-name links from USPTO owners to SEC registrants (CIK) and to Form D issuers. MANIFEST.txt: file sizes and SHA-256 digests. Built from public USPTO, SEC and PatentsView sources by scripts/build_release_tables.py in github.com/ericsilver/tm-vocabulary, where every table rebuilds from source. The underlying government data are public domain; PatentsView-derived material carries PatentsView's CC-BY-4.0 attribution requirement. An interactive online appendix is at aporia.institute/tm-vocabulary.