Update LocalNewsQA gold split (2026-05-16)

README.md

@@ -1,61 +1,32 @@
 ---
-dataset_info:
-  features:
-  - name: question
-    dtype: string
-  - name: options
-    list: string
-  - name: correct_answer
-    dtype: string
-  - name: distractors
-    list: string
-  - name: country
-    dtype: string
-  - name: continent
-    dtype: string
-  - name: topic
-    dtype: string
-  - name: year
-    dtype: int64
-  - name: ambiguity_flag
-    dtype: bool
-  - name: evidence_hint
-    dtype: string
-  - name: split_name
-    dtype: string
-  - name: split_type
-    dtype: string
-  - name: split_family
-    dtype: string
-  - name: target_country
-    dtype: string
-  - name: contrast_country
-    dtype: string
-  - name: target_answer
-    dtype: string
-  - name: contrast_answer
-    dtype: string
-  - name: generation_shard_year_range
-    dtype: string
-  - name: generation_shard_focus
-    dtype: string
-  - name: generation_shard_angle
-    dtype: string
-  - name: generator_model
-    dtype: string
-  - name: generation_custom_id
-    dtype: string
-  - name: generation_item_index
-    dtype: int64
-  splits:
-  - name: train
-    num_bytes: 25570399
-    num_examples: 35874
-  download_size: 4892061
-  dataset_size: 25570399
 configs:
 - config_name: default
   data_files:
   - split: train
-    path: data/train-*
+    path: data/train-00000-of-00001.parquet
 ---
+
+# LocalNewsQA
+
+LocalNewsQA is an English-news benchmark for localized knowledge disambiguation. This release contains the final gold split used in the ARR May 2026 submission experiments.
+
+## Release
+
+- Version: `localnewsqa_gold_20260516`
+- Total rows: 18,700
+- Explicit rows: 17,000 (1,000 per country)
+- Ambiguous rows: 1,700 (100 per country)
+- Countries: 17
+- Topic families: 16
+
+The explicit split names the locale in the question. The ambiguous split omits the locale from the question and includes target/contrast metadata with different supported answers. Rows include cached evidence titles/URLs/excerpts when available and split-specific validation status fields.
+
+## Columns
+
+Core fields: `id`, `question`, `options`, `correct_answer`, `distractors`, `country`, `continent`, `topic`, `year`, `split_type`, `ambiguity_flag`.
+
+Locale fields: `target_country`, `contrast_country`, `target_answer`, `contrast_answer`.
+
+Evidence fields: `evidence_hint`, `target_evidence_title`, `target_evidence_url`, `target_evidence_excerpt`, `contrast_evidence_title`, `contrast_evidence_url`, `contrast_evidence_excerpt`, `target_evidence_cached`, `contrast_evidence_cached`.
+
+Validation fields: `relation_support_status`, `curation_status`, `source_row_id`, `release_version`.

data/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e7fd0e359e1cf443028801eec9e5949656b0716c3faf465f70e885f93ad1a1d
-size 4892061
+oid sha256:203c8d21858d6e7aae48156b39ac0badd62947738944c44ff097abca83788a99
+size 3560925

release_summary.json

@@ -0,0 +1,11 @@
+{
+  "rows": 18700,
+  "explicit_rows": 17000,
+  "ambiguous_rows": 1700,
+  "countries": 17,
+  "topics": 16,
+  "explicit_per_country_min": 1000,
+  "explicit_per_country_max": 1000,
+  "ambiguous_per_country_min": 100,
+  "ambiguous_per_country_max": 100
+}