# Changelog

## 2.0.0 — 2026-07-23

### Dataset

- Refreshed the active dataset from the current VBPL Next.js catalog and JSON gateway.
- Published 171,556 metadata rows, 170,824 unique HTML documents, and 1,033,255 unique relationship edges.
- Added UUID and portal-prefixed document IDs; all active-config join keys are now strings.
- Removed duplicate content rows while retaining historical records no longer listed by the portal.
- Filled missing refresh metadata from the preceding published snapshot.
- Split the incompatible `legacy` config into loadable `legacy_metadata` and `legacy_content` configs.
- Re-encoded active Parquet files as bounded Arrow string row groups for Hugging Face `datasets` and Dataset Viewer compatibility.

### Crawler and tooling

- Replaced the retired VBPL crawl path with the current catalog server action and public document API.
- Consolidated metadata, relationships, and full-text HTML into one resumable spider.
- Added streaming JSONL-to-Parquet conversion, deterministic upsert, and comparison tools.
- Removed the obsolete content spider, Pandas helper, and unused Scrapy item/pipeline boilerplate.
- Added a reproducible `uv` project and lockfile.