Datasets:

---

v13s

/

golden-fto-layer-a

like 0

Follow

Vox 1

v1.3.1 is live — the dataset now spans three patent offices. The headline addition is the JP slice: 5,801,435 rejection-notice triples lowered from the JPO 整理標準化 (seiri-hyōjunka) ISO bulk corpus, filing years 2002–2019.

What's new

• JP slice (5,801,435 rows) — the full 2002–2019 整理標準化 lowering, ~1,160× the citation density of the US slice. Joins US (5,000) + EP (850) for 5,807,285 rows total.
• Labels only (JP) — per JPO 利用規約 第4条, JP rows publish derived labels only: prior_art[].severity_st14 + examination.legal_articles[]. The rejection-notice prose (rejection_basis_text) is withheld (null). 特許庁 (JPO) is attributed as the upstream source.
• Per-jurisdiction configs — us, ep, jp, or default (all three).

Quick start

from datasets import load_dataset
ds_jp = load_dataset("v13s/golden-fto-layer-a", "jp", split="train")
print(len(ds_jp))  # 5,801,435

Heads-up on JP severity

JP severity is derived from the 整理標準化 citation graph (severity_source="cited_type_proxy"), refined to WIPO ST.14 (severity_source="prose_extracted") on apps the enrichment pipeline has reached. Because it is citation-graph-derived rather than rejection-derived, a novelty-relevant reference can co-occur with an unamended grant — this is expected, not a data error.

Known limitations

• The JP cross-citation prior_art_index (an ~81M-ref reduce) is deferred to a later patch; US + EP indexes ship today. Query the JP cases partition directly for now.
• severity_st14 rolls in progressively as the JPO-API-gated enrichment processes more apps.

Feedback welcome — especially on the JP labels-only schema and the proxy-vs-prose severity distinction. 🙏