Simplify v2 to a single full config

Expose only one full subset/config backed by data/full.jsonl. Remove dev/test split files and core_1000 standalone artifacts.

README.md

@@ -27,76 +27,58 @@ tags:
 - long-document
 - synthetic-data
 configs:
-- config_name: default
-  data_files:
-  - split: dev
-    path: data/core/dev.jsonl
-  - split: test
-    path: data/core/test.jsonl
-- config_name: core_1000
-  data_files:
-  - split: dev
-    path: data/core/dev.jsonl
-  - split: test
-    path: data/core/test.jsonl
-- config_name: full_2000
-  data_files:
-  - split: dev
-    path: data/full/dev.jsonl
-  - split: test
-    path: data/full/test.jsonl
+- config_name: full
+  data_files: data/full.jsonl
 ---
 
 # TW-Local-MT-Bench v2
 
 TW-Local-MT-Bench v2 is a 2,000-example synthetic benchmark for evaluating whether a translation or rewriting model can produce natural Taiwan Traditional Chinese (`zh-TW`) from English, Mainland Chinese, Hong Kong Traditional Chinese, Cantonese-style written Chinese, or code-mixed English/Chinese documents.
 
-The benchmark focuses on Taiwan localization rather than generic Traditional Chinese conversion. A good system must preserve source facts, names, numbers, dates, currencies, Markdown, HTML tags, placeholders, URLs, email addresses, API names, product names, and document structure while using vocabulary, punctuation, UI wording, institutional terms, and everyday phrasing that are natural for readers in Taiwan.
+The dataset now exposes a single Hugging Face subset/config: `full`. It is not split into `dev` and `test` files. The full 2,000 rows are loaded together from `data/full.jsonl`. The original row-level `split` column is kept as metadata for downstream analysis, but the Hub dataset itself is published as one complete subset.
 
-## Version Summary
+## Loading
 
-v2 merges the earlier v0 sentence-level localization set with a v1 long-document and code-mixed localization set.
+```python
+from datasets import load_dataset
 
-- v0: sentence-level locale-sensitive translation from English, zh-CN, and zh-Hant-HK into Taiwan Traditional Chinese.
-- v1: paragraph and long-document localization with code-mixing, Markdown, placeholders, cross-paragraph terminology consistency, and broader domain coverage.
-- v2: unified schema with two official views: `core_1000` and `full_2000`.
+dataset = load_dataset("voidful/zh_translation_benchmark", "full")
 
-## Configs
+# Hugging Face stores a single data file under the conventional "train" split.
+full = dataset["train"]
+print(full.num_rows)  # 2000
+print(full[0])
+```
 
-The default config is `core_1000`, intended for leaderboard-style reporting and fast model comparison.
+If you want a plain `Dataset` object:
 
 ```python
 from datasets import load_dataset
 
-# Official core set: 1,000 rows, dev 200 / test 800
-core = load_dataset("voidful/zh_translation_benchmark")
-
-# Same as default, explicit name
-core = load_dataset("voidful/zh_translation_benchmark", "core_1000")
-
-# Full analysis set: 2,000 rows, dev 399 / test 1601
-full = load_dataset("voidful/zh_translation_benchmark", "full_2000")
+full = load_dataset("voidful/zh_translation_benchmark", "full", split="train")
 ```
 
-## Dataset Views
+## Version Summary
 
-| View | Rows | Dev | Test | Recommended use |
-|---|---:|---:|---:|---|
-| `core_1000` | 1,000 | 200 | 800 | Main leaderboard, fast comparison, model regression tests |
-| `full_2000` | 2,000 | 399 | 1,601 | Error analysis, domain breakdown, stress testing |
+v2 merges the earlier v0 sentence-level localization set with a v1 long-document and code-mixed localization set.
 
-## Direction Distribution
+- v0: sentence-level locale-sensitive translation from English, zh-CN, and zh-Hant-HK into Taiwan Traditional Chinese.
+- v1: paragraph and long-document localization with code-mixing, Markdown, placeholders, cross-paragraph terminology consistency, and broader domain coverage.
+- v2: unified schema with one official `full` subset containing all 2,000 rows.
 
-### Core 1000
+## Dataset Summary
 
-| Direction | Count |
+| Property | Value |
 |---|---:|
-| English -> Taiwan Traditional Chinese | 350 |
-| mixed English/Chinese -> Taiwan Traditional Chinese | 300 |
-| zh-CN -> Taiwan Traditional Chinese | 250 |
-| zh-Hant-HK -> Taiwan Traditional Chinese | 100 |
+| Rows | 2,000 |
+| Unique dedup hashes | 2,000 |
+| Domains | 56 |
+| Domain groups | 18 |
+| Source versions | v0: 1,000; v1: 1,000 |
+| Published subset/config | `full` |
+| Published split files | none; one full JSONL file |
 
-### Full 2000
+## Direction Distribution
 
 | Direction | Count |
 |---|---:|
@@ -107,10 +89,11 @@ full = load_dataset("voidful/zh_translation_benchmark", "full_2000")
 
 ## Length Distribution
 
-| View | Sentence | Paragraph | Long Document |
-|---|---:|---:|---:|
-| `core_1000` | 500 | 300 | 200 |
-| `full_2000` | 1,500 | 300 | 200 |
+| Length bucket | Count |
+|---|---:|
+| Sentence | 1,500 |
+| Paragraph | 300 |
+| Long document | 200 |
 
 ## What This Dataset Tests
 
@@ -148,54 +131,44 @@ v2 also tests:
 
 The full set covers 56 domains across 18 domain groups. Major domains include:
 
-| Domain | Full count | Core count |
-|---|---:|---:|
-| `software_it` | 130 | 35 |
-| `government_public_service` | 72 | 34 |
-| `workplace_business` | 67 | 9 |
-| `education` | 66 | 34 |
-| `ecommerce_product_ui` | 61 | 12 |
-| `zhcn_software_it` | 60 | 20 |
-| `zhcn_ecommerce` | 50 | 21 |
-| `zhcn_workplace` | 50 | 20 |
-| `zhcn_public_health_gov` | 50 | 21 |
-| `zhcn_daily_transit_food` | 50 | 22 |
-| `zhcn_education_media` | 40 | 21 |
-| `product_ui` | 37 | 25 |
-| `cybersecurity` | 37 | 25 |
-| `ecommerce` | 37 | 25 |
-| `fintech_banking` | 37 | 25 |
-| `insurance` | 37 | 25 |
-| `healthcare_clinic` | 37 | 25 |
-| `healthcare` | 36 | 7 |
-| `marketing_social_media` | 35 | 9 |
-| `medical_device` | 35 | 23 |
-| `legal_privacy` | 35 | 23 |
-| `academic_research` | 35 | 23 |
-| `workplace_hr` | 35 | 23 |
+| Domain | Count |
+|---|---:|
+| `software_it` | 130 |
+| `government_public_service` | 72 |
+| `workplace_business` | 67 |
+| `education` | 66 |
+| `ecommerce_product_ui` | 61 |
+| `zhcn_software_it` | 60 |
+| `zhcn_ecommerce` | 50 |
+| `zhcn_workplace` | 50 |
+| `zhcn_public_health_gov` | 50 |
+| `zhcn_daily_transit_food` | 50 |
+| `zhcn_education_media` | 40 |
+| `product_ui` | 37 |
+| `cybersecurity` | 37 |
+| `ecommerce` | 37 |
+| `fintech_banking` | 37 |
+| `insurance` | 37 |
+| `healthcare_clinic` | 37 |
+| `healthcare` | 36 |
+| `marketing_social_media` | 35 |
+| `medical_device` | 35 |
+| `legal_privacy` | 35 |
+| `academic_research` | 35 |
+| `workplace_hr` | 35 |
 
 Use the `domain`, `domain_group`, `domain_zh`, and `subdomain` columns for more detailed breakdowns.
 
 ## Files
 
-The Hugging Face loader uses JSONL files under `data/`.
-
 | Path | Description |
 |---|---|
-| `data/core/dev.jsonl` | Core dev split |
-| `data/core/test.jsonl` | Core test split |
-| `data/core/dev.csv` | Core dev split in CSV |
-| `data/core/test.csv` | Core test split in CSV |
-| `data/full/dev.jsonl` | Full dev split |
-| `data/full/test.jsonl` | Full test split |
-| `data/full/dev.csv` | Full dev split in CSV |
-| `data/full/test.csv` | Full test split in CSV |
+| `data/full.jsonl` | Single 2,000-row JSONL file used by the Hugging Face loader |
+| `data/full.csv` | Single 2,000-row CSV file |
 | `full/tw_localization_benchmark_v2_2000.csv` | Complete 2,000-row CSV |
 | `full/tw_localization_benchmark_v2_2000.jsonl` | Complete 2,000-row JSONL |
-| `full/tw_localization_benchmark_v2_core1000.csv` | Official core 1,000-row CSV |
-| `full/tw_localization_benchmark_v2_core1000.jsonl` | Official core 1,000-row JSONL |
 | `full/tw_localization_benchmark_v2.xlsx` | Workbook with Overview, Examples_2000, Core_1000, Rubric, Locale_Policy, Glossary, Eval_Prompts, Auto_Checks, Model_Slots, and ChangeLog sheets |
-| `full/tw_localization_benchmark_v2_manifest.json` | Manifest with counts, checksums, schema, and core selection rule |
+| `full/tw_localization_benchmark_v2_manifest.json` | Manifest with counts, checksums, schema, and historical core selection rule |
 | `full/tw_localization_benchmark_v2_README.md` | Original v2 local package README |
 
 ## Schema
@@ -207,8 +180,8 @@ Each example contains:
 | `id` | Stable v2 ID, such as `TWLOC-V2-0001` |
 | `legacy_id` | Original v0 or v1 ID |
 | `version_source` | Source version: `v0` or `v1` |
-| `core_1000` | Whether the row belongs to the official core set |
-| `split` | `dev` or `test` |
+| `core_1000` | Whether the row belonged to the historical core set |
+| `split` | Original dev/test marker retained as row metadata |
 | `source_lang` | Normalized source language bucket |
 | `source_locale_raw` | Original source locale label |
 | `target_lang` | Target language, `zh-TW` |
@@ -289,21 +262,6 @@ Suggested 100-point human evaluation rubric:
 | Fluency and register | 10 | Natural, readable output matching customer support, legal, medical, UI, government, or casual register |
 | Format and constraint preservation | 10 | Markdown, placeholders, URLs, HTML, numbers, currencies, tables, and bullets |
 
-Suggested leaderboard reporting:
-
-- `overall_score`
-- `core_1000_score`
-- `full_2000_score`
-- `sentence_score`
-- `paragraph_score`
-- `long_doc_score`
-- `code_mix_score`
-- `zh_cn_to_tw_score`
-- `hk_to_tw_score`
-- `format_preservation_score`
-- `taiwan_localization_score`
-- `domain_breakdown`
-
 ## LLM-Assisted Judging Prompt
 
 ```text
@@ -344,10 +302,9 @@ needs_native_taiwanese_reviewer_pass_before_public_release
 
 For a public leaderboard, the recommended process is:
 
-1. Use `core_1000` as the main visible benchmark.
-2. Add a Taiwan-native reviewer pass before freezing the references.
-3. Keep an additional private hidden test set.
-4. Do not rely on a single reference-matching metric as the final ranking criterion.
+1. Add a Taiwan-native reviewer pass before freezing the references.
+2. Keep an additional private hidden test set.
+3. Do not rely on a single reference-matching metric as the final ranking criterion.
 
 ## Checksums