docs(v2): refresh dataset card and viewer configs

README.md

@@ -1,40 +1,116 @@
 ---
+license: odbl
+language:
+  - en
+tags:
+  - openstreetmap
+  - wikidata
+  - wikipedia
+  - wikivoyage
+  - geospatial
+configs:
+  - config_name: polygons
+    data_files:
+      - split: polygons
+        path: polygons/*.parquet
+  - config_name: polygon_document_links
+    data_files:
+      - split: polygon_document_links
+        path: polygon_document_links/*.parquet
+  - config_name: wikipedia_documents
+    data_files:
+      - split: wikipedia_documents
+        path: wikipedia/documents/*.parquet
+  - config_name: wikipedia_sections
+    data_files:
+      - split: wikipedia_sections
+        path: wikipedia/sections/*.parquet
+  - config_name: wikivoyage_documents
+    data_files:
+      - split: wikivoyage_documents
+        path: wikivoyage/documents/*.parquet
+  - config_name: wikivoyage_sections
+    data_files:
+      - split: wikivoyage_sections
+        path: wikivoyage/sections/*.parquet
+  - config_name: wikidata_facts
+    data_files:
+      - split: wikidata_facts
+        path: wikidata/facts/*.parquet
 dataset_info:
-  config_name: default
   version: wikipedia-tags-v2
+  regions: 386
+  polygons: 1259424
+  documents: 2346473
 ---
-# OSM Polygon Wikidata + Wikipedia
 
-Version 2 keeps every V1 polygon and adds polygons discovered through valid multilingual OSM `wikipedia=*` tags, including polygons without a Wikidata QID.
+# OSM Polygon Wikidata + Wikipedia, V2
 
-The build reuses finalized V1 documents and sidecars. It fetches only direct Wikipedia pages that are not already present in V1, so reruns are resumable and do not refetch the existing corpus.
+![V2 dataset overview](assets/dataset_hero.png)
+
+V2 keeps the V1 Wikidata-derived corpus and adds valid multilingual OSM `wikipedia=*` references, including polygons without a Wikidata QID. The code is maintained in the [GitHub repository](https://github.com/NoeFlandre/osm-polygon-wikidata-only).
+
+The public V2 Trackio snapshot is [`final-dataset-snapshot-v2`](https://huggingface.co/spaces/NoeFlandre/osm-polygon-wikidata-and-wikipedia-trackio).
 
 ## Snapshot
 
 - **Hugging Face dataset:** [NoeFlandre/osm-polygon-wikidata-and-wikipedia](https://huggingface.co/datasets/NoeFlandre/osm-polygon-wikidata-and-wikipedia)
 - **Regions:** 386
 - **Polygons:** 1,259,424
+- **Unique Wikidata entities:** 1,119,287
 - **Wikipedia documents:** 2,332,053
-- **Wikipedia sections:** 12,492,162
 - **Wikivoyage documents:** 14,420
+- **Wikipedia + Wikivoyage documents:** 2,346,473
+- **Wikipedia sections:** 12,492,162
 - **Wikivoyage sections:** 302,200
 - **Wikidata facts:** 3,901,092
 - **Polygon-document links:** 2,541,553
 - **Wikipedia-tag-only polygons:** 33,682
+- **Document words:** 845,066,382
+- **Languages represented:** 353
+- **Polygon/link-table storage:** 0.1 GB
+- **Total Parquet storage:** 21.4 GB
+
+## V2 compared with V1
+
+The following deltas are computed from local V1 and V2 artifact identities, not entered manually:
+
+- **Additional polygons discovered through V2 Wikipedia tags:** 75,314
+- **Additional Wikipedia document identities:** 40,013
+
+## Coverage maps
+
+### All V2 dataset polygons
+
+![All V2 dataset polygons](assets/coverage_map.png)
+
+Every point is one retained V2 polygon, including polygons without text.
+
+### Polygons with Wikipedia or Wikivoyage text
+
+![V2 polygons with text](assets/geographic_text_presence.png)
+
+Each point is one polygon linked to at least one non-empty Wikipedia or Wikivoyage document. A polygon is counted once even when several documents qualify.
+
+### H3 density of polygons with text
+
+![V2 geographic text density](assets/geographic_text_density.png)
+
+Each H3 cell shows the raw count of unique V2 polygons with non-empty Wikipedia or Wikivoyage text. Colour uses a logarithmic scale; it is not a proportion.
 
-## Discovery and provenance
+## Deduplication and provenance
 
-Each polygon row records `discovery_sources` (`wikidata`, `wikipedia_tag`, or both), normalized multilingual references, and structured rejections. The `polygon_document_links` table stores both V1 Wikidata-sitelink relationships and V2 direct-tag relationships in one schema, with `link_sources` identifying how each relationship was found.
+V2 deduplicates documents by `document_id` and polygon-document links by `(polygon_id, project, document_id)` within each region. Byte-identical repeats collapse deterministically; conflicting rows fail closed. The `discovery_sources` and `link_sources` fields distinguish Wikidata-derived and direct Wikipedia-tag relationships.
 
 ## Repository layout
 
-- `polygons/` — one row for every retained OSM polygon.
-- `wikipedia/documents/` — canonical 32-column Wikipedia document rows; direct-only rows may have a null `wikidata`.
-- `polygon_document_links/` — unified polygon-to-document relationships for Wikipedia and Wikivoyage.
-- `wikipedia/sections/` — section-level Wikipedia text using the exact V1 22-column section schema.
-- `wikivoyage/documents/`, `wikivoyage/sections/`, `wikidata/facts/` — reused V1 sidecars.
-- `manifests/processed_pbfs.json` — versioned, per-region row counts and file hashes.
+- `polygons/<stem>.parquet` — one row per retained OSM polygon.
+- `wikipedia/documents/<stem>.parquet` and `wikipedia/sections/<stem>.parquet` — multilingual Wikipedia documents and their sections.
+  Wikipedia sections retain the exact V1 22-column section schema for lossless reuse.
+- `wikivoyage/documents/<stem>.parquet` and `wikivoyage/sections/<stem>.parquet` — Wikivoyage documents and sections reused from V1 where available.
+- `polygon_document_links/<stem>.parquet` — unified Wikipedia and Wikivoyage polygon links.
+- `wikidata/facts/<stem>.parquet` — structured Wikidata facts.
 
 ## Reproducibility
 
-The V2 build is selected explicitly with `sync-dir --dataset-version v2`. V1 commands and the V1 artifact tree remain unchanged.
+Run V2 explicitly with `sync-dir --dataset-version v2`. V1 remains a separate contract and is not modified by V2 processing.