Trim to essentials: keep Usage, cut the reference scorer and the per-task table

README.md

@@ -1,593 +1,529 @@
----
-license: mit
-language:
-- ko
-- en
-pretty_name: NOLLI
-size_categories:
-- 1K<n<10K
-task_categories:
-- question-answering
-tags:
-- reasoning
-- logic
-- puzzles
-- korean
-configs:
-- config_name: array_formula
-  data_files:
-  - split: test
-    path: array_formula/test-*
-- config_name: causal_dag
-  data_files:
-  - split: test
-    path: causal_dag/test-*
-- config_name: cipher
-  data_files:
-  - split: test
-    path: cipher/test-*
-- config_name: cryptarithmetic
-  data_files:
-  - split: test
-    path: cryptarithmetic/test-*
-- config_name: inequality
-  data_files:
-  - split: test
-    path: inequality/test-*
-- config_name: jamo
-  data_files:
-  - split: test
-    path: jamo/test-*
-- config_name: kinship
-  data_files:
-  - split: test
-    path: kinship/test-*
-- config_name: korean_units
-  data_files:
-  - split: test
-    path: korean_units/test-*
-- config_name: minesweeper
-  data_files:
-  - split: test
-    path: minesweeper/test-*
-- config_name: number_baseball
-  data_files:
-  - split: test
-    path: number_baseball/test-*
-- config_name: saju
-  data_files:
-  - split: test
-    path: saju/test-*
-- config_name: sat_puzzles
-  data_files:
-  - split: test
-    path: sat_puzzles/test-*
-- config_name: sudoku
-  data_files:
-  - split: test
-    path: sudoku/test-*
-- config_name: time
-  data_files:
-  - split: test
-    path: time/test-*
-- config_name: yacht_dice
-  data_files:
-  - split: test
-    path: yacht_dice/test-*
-dataset_info:
-- config_name: array_formula
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 13407331
-    num_examples: 600
-  download_size: 3956244
-  dataset_size: 13407331
-- config_name: causal_dag
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 10991552
-    num_examples: 600
-  download_size: 2486519
-  dataset_size: 10991552
-- config_name: cipher
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 2191175
-    num_examples: 600
-  download_size: 277124
-  dataset_size: 2191175
-- config_name: cryptarithmetic
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 2697176
-    num_examples: 600
-  download_size: 545993
-  dataset_size: 2697176
-- config_name: inequality
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 2259573
-    num_examples: 600
-  download_size: 436357
-  dataset_size: 2259573
-- config_name: jamo
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 384368
-    num_examples: 300
-  download_size: 52641
-  dataset_size: 384368
-- config_name: kinship
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: choices
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 3188738
-    num_examples: 300
-  download_size: 496080
-  dataset_size: 3188738
-- config_name: korean_units
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 1401654
-    num_examples: 300
-  download_size: 639655
-  dataset_size: 1401654
-- config_name: minesweeper
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 2570080
-    num_examples: 600
-  download_size: 424457
-  dataset_size: 2570080
-- config_name: number_baseball
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 1299609
-    num_examples: 600
-  download_size: 220324
-  dataset_size: 1299609
-- config_name: saju
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 180478
-    num_examples: 300
-  download_size: 26018
-  dataset_size: 180478
-- config_name: sat_puzzles
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 28051763
-    num_examples: 600
-  download_size: 4931284
-  dataset_size: 28051763
-- config_name: sudoku
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 1446800
-    num_examples: 600
-  download_size: 291709
-  dataset_size: 1446800
-- config_name: time
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 258002
-    num_examples: 300
-  download_size: 39819
-  dataset_size: 258002
-- config_name: yacht_dice
-  features:
-  - name: id
-    dtype: string
-  - name: task
-    dtype: string
-  - name: language
-    dtype: string
-  - name: difficulty
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: solution
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 2112133
-    num_examples: 600
-  download_size: 374101
-  dataset_size: 2112133
----
-
-# NOLLI
-
-**NOLLI (논리)** is a benchmark of rule-based logical puzzles in **Korean and English**, designed with
-difficulty calibration in mind. It covers **15 task types x 3 difficulty tiers (easy / medium / hard)
-x 100 examples** each, for a total of **7,500 examples** (EN 3,000 + KO 4,500).
-
-All examples are programmatically generated with solver-verified answers. If you need more data,
-the generators in the GitHub repository can produce additional examples for any task, language,
-and difficulty tier.
-
-📄 Paper: [https://arxiv.org/abs/2608.04397](https://arxiv.org/abs/2608.04397)\
-💻 Code: [https://github.com/HAE-RAE/NOLLI](https://github.com/HAE-RAE/NOLLI)
-
-## Tasks
-
-Tasks follow a three-level cross-lingual spectrum: **direct translations**, **script adaptations**,
-and **Korean-only** tasks.
-
-### 1. Direct Translations (8 tasks · EN + KO)
-
-Same generators and parameters in both languages, so performance gaps isolate presentation language.
-
-| task | description | en (easy/med/hard) | ko (easy/med/hard) | total |
-|---|---|---|---|---|
-| `array_formula` | Apply row/column aggregation formulas (SUM/MEAN/MAX/...) to a 2D array in sequence | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `causal_dag` | Infer event-propagation time through a causal graph with per-edge delays | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `inequality` | CSP: place numbers 1..N satisfying inequality constraints (unique solution) | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `minesweeper` | Minesweeper with minimal hints preserving a unique solution | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `number_baseball` | Infer a hidden N-digit number from Strike/Ball hints | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `sat_puzzles` | Boolean satisfiability (CNF) framed as a natural-language scenario | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `sudoku` | 9x9 Sudoku with guaranteed-unique solutions; report the digits at 4-6 spot-checked cells | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `yacht_dice` | Assign 12 dice rolls to 12 scoring categories to maximize total score; report the partial sum over spot-checked rounds | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-
-### 2. Script Adaptations (2 tasks · EN + KO)
-
-Not translation-equivalent: English uses Roman letters; Korean uses Hangul *jamo*.
-Calibrated independently per language.
-
-| task | description | en (easy/med/hard) | ko (easy/med/hard) | total |
-|---|---|---|---|---|
-| `cipher` | Decode stacked ciphers (Substitution, Vigenere, Reverse, Playfair, Transposition) | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-| `cryptarithmetic` | Letters-for-digits arithmetic (SEND+MORE=MONEY style); Korean maps *jamo* groups | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
-
-### 3. Korean-Only (5 tasks · KO)
-
-No English counterpart — built on structures specific to the Korean language and culture
-(kinship terminology, the sexagenary calendar, traditional units, Hangul orthography).
-
-| task | description | ko (easy/med/hard) | total |
-|---|---|---|---|
-| `jamo` | Decompose a Hangul syllable into 초성/중성/종성, shift 초성, recompose | 100 / 100 / 100 | 300 |
-| `kinship` | Infer a Korean kinship term from a chain of family relationships (26-way multiple choice) | 100 / 100 / 100 | 300 |
-| `korean_units` | Convert traditional Korean units (평·마지기·되·자·냥, ...) with randomized rates | 100 / 100 / 100 | 300 |
-| `saju` | Compute the four 사주(四柱) pillars from a birth date/time | 100 / 100 / 100 | 300 |
-| `time` | Korean calendar reasoning: holiday anchor + relative-day expression → date / 60갑자 일진 | 100 / 100 / 100 | 300 |
-
-## Usage
-
-There is no default config — a config name is always required.
-
-```python
-from datasets import load_dataset, get_dataset_config_names
-
-# one task
-ds = load_dataset("HAERAE-HUB/NOLLI", "sudoku", split="test")
-
-# filter by language / difficulty (both are columns)
-ko_hard = ds.filter(lambda x: x["language"] == "ko" and x["difficulty"] == "hard")
-
-# all 15 tasks
-tasks = get_dataset_config_names("HAERAE-HUB/NOLLI")
-all_ds = {t: load_dataset("HAERAE-HUB/NOLLI", t, split="test") for t in tasks}
-```
-
-Note that `kinship` carries an extra `choices` column, so the configs cannot be merged
-with `concatenate_datasets` without first aligning the schema.
-
-## Structure
-
-Each config corresponds to one task. All configs share a single `test` split and the same schema
-(the `choices` column exists only in `kinship`):
-
-| column | type | description |
-|---|---|---|
-| `id` | str | unique example id, e.g. `sudoku_ko_0000` |
-| `task` | str | task name (same as config name) |
-| `language` | str | `ko` or `en` |
-| `difficulty` | str | `easy` / `medium` / `hard` |
-| `question` | str | full prompt for the model |
-| `choices` | str | JSON-encoded answer choices — only present in the `kinship` config |
-| `answer` | str | gold answer — format varies by task, see [Answer format](#answer-format) |
-| `solution` | str | step-by-step reference solution |
-
-## Answer format
-
-The `answer` string is not free text: each task has its own gold format, and the prompt in
-`question` already tells the model which one to produce. Two things differ across tasks and
-matter when scoring.
-
-**1. Final-line protocol.** Seven tasks instruct the model to end its response with a literal
-`Answer: ...` line; the other eight ask for the bare value with no prefix. Parsing every task
-as if it used `Answer:` — or as if none did — silently drops half the benchmark.
-
-| protocol | tasks |
-|---|---|
-| ends with `Answer: <value>` | `inequality`, `jamo`, `korean_units`, `minesweeper`, `number_baseball`, `sudoku`, `yacht_dice` |
-| bare value, no prefix | `array_formula`, `causal_dag`, `cipher`, `cryptarithmetic`, `kinship`, `saju`, `sat_puzzles`, `time` |
-
-**2. Gold value shape.**
-
-| task | gold format | example |
-|---|---|---|
-| `array_formula` | integer (545/600) **or** a category string (55/600, from prompts that ask for "the exact value" rather than "only a number"); no commas, no decimals | `7395000`, `Fruit`, `실버` |
-| `causal_dag` | single integer | `19` |
-| `cipher` | ciphertext plaintext, no spaces — EN uppercase `A-Z`, KO Hangul syllables | `NOWHQXVOKMVDLLKBKCDMTVQ`, `문화시장관리지도` |
-| `cryptarithmetic` | integer (the value of the target word) | `3203254` |
-| `inequality` | flattened N x N grid, space-separated — 25 tokens (5x5) or 36 (6x6) | `3 1 2 5 4 2 5 1 ...` |
-| `jamo` | Hangul string | `딓떼` |
-| `kinship` | a single choice letter `A`-`Z` (always 26 options; `choices` maps letter to term) | `F` |
-| `korean_units` | integer **with its unit suffix**; the unit varies per item (`돈` 81, `푼` 77, `평` 75, `홉` 67) | `465373918평` |
-| `minesweeper` | 0-based `(row,col)` mine coordinates, comma-separated; 6-24 mines per item; order is not significant | `(0,2), (0,6), (2,3)` |
-| `number_baseball` | the secret number — 7 digits (200 items) or 8 digits (400) | `3264579` |
-| `saju` | 2-character 간지 | `정미` |
-| `sat_puzzles` | JSON object mapping each name to a boolean; 9, 11, or 14 keys | `{"Alice": false, "Bob": true, ...}` |
-| `sudoku` | space-separated digits at the 4-6 spot-checked 1-based `(row,col)` coordinates listed in the prompt — **not** the full grid | `3 7 3 6 5` |
-| `time` | one of three shapes depending on the item: `YYYY.M.D` date (82), 2-character 60갑자 일진 (177), or a `요일` weekday (41) | `1999.1.24`, `을유`, `목요일` |
-| `yacht_dice` | integer — the partial sum of per-round scores over the spot-checked rounds under the optimal 12-round assignment (2 rounds for easy/medium, 8 for hard); **not** the 12-round total | `42` |
-
-**Reference extraction.** A minimal exact-match scorer that covers all 15 tasks:
-
-```python
-import json, re
-
-ANSWER_LINE = {"inequality", "jamo", "korean_units", "minesweeper",
-               "number_baseball", "sudoku", "yacht_dice"}
-
-def extract(task: str, output: str) -> str | None:
-    lines = [l for l in output.strip().splitlines() if l.strip()]
-    if not lines:
-        return None
-    if task in ANSWER_LINE:
-        for line in reversed(lines):                       # last `Answer:` wins
-            m = re.match(r"\s*Answer\s*:\s*(.*)$", line)
-            if m:
-                return m.group(1).strip()
-        return None
-    return lines[-1].strip()
-
-def is_correct(task: str, pred: str | None, gold: str) -> bool:
-    if pred is None:
-        return False
-    if task == "sat_puzzles":                              # key order is not significant
-        try:
-            return json.loads(pred) == json.loads(gold)
-        except json.JSONDecodeError:
-            return False
-    if task == "minesweeper":                              # coordinate order is not significant
-        coords = lambda s: set(re.findall(r"\(\s*\d+\s*,\s*\d+\s*\)", s.replace(" ", "")))
-        return coords(pred) == coords(gold)
-    if task in {"inequality", "sudoku"}:                   # whitespace-insensitive token list
-        return pred.split() == gold.split()
-    if task in {"causal_dag", "cryptarithmetic", "yacht_dice"}:
-        try:
-            return int(pred.replace(",", "")) == int(gold)
-        except ValueError:
-            return False
-    return pred.strip() == gold.strip()                    # incl. the `korean_units` unit suffix
-```
-
-## Citation
-
-```bibtex
-@misc{choi2026nolli,
-      title={NOLLI: A Difficulty-Calibrated Puzzle Benchmark for Diagnosing the English-Korean Performance Gap},
-      author={Dasol Choi and Joonyong Park and Daegon Yu and Soo Yong Kim and Youngsook Song and Seunghyeok Hong},
-      year={2026},
-      eprint={2608.04397},
-      archivePrefix={arXiv},
-      primaryClass={cs.CL},
-      url={https://arxiv.org/abs/2608.04397},
-}
-```
+---
+license: mit
+language:
+- ko
+- en
+pretty_name: NOLLI
+size_categories:
+- 1K<n<10K
+task_categories:
+- question-answering
+tags:
+- reasoning
+- logic
+- puzzles
+- korean
+configs:
+- config_name: array_formula
+  data_files:
+  - split: test
+    path: array_formula/test-*
+- config_name: causal_dag
+  data_files:
+  - split: test
+    path: causal_dag/test-*
+- config_name: cipher
+  data_files:
+  - split: test
+    path: cipher/test-*
+- config_name: cryptarithmetic
+  data_files:
+  - split: test
+    path: cryptarithmetic/test-*
+- config_name: inequality
+  data_files:
+  - split: test
+    path: inequality/test-*
+- config_name: jamo
+  data_files:
+  - split: test
+    path: jamo/test-*
+- config_name: kinship
+  data_files:
+  - split: test
+    path: kinship/test-*
+- config_name: korean_units
+  data_files:
+  - split: test
+    path: korean_units/test-*
+- config_name: minesweeper
+  data_files:
+  - split: test
+    path: minesweeper/test-*
+- config_name: number_baseball
+  data_files:
+  - split: test
+    path: number_baseball/test-*
+- config_name: saju
+  data_files:
+  - split: test
+    path: saju/test-*
+- config_name: sat_puzzles
+  data_files:
+  - split: test
+    path: sat_puzzles/test-*
+- config_name: sudoku
+  data_files:
+  - split: test
+    path: sudoku/test-*
+- config_name: time
+  data_files:
+  - split: test
+    path: time/test-*
+- config_name: yacht_dice
+  data_files:
+  - split: test
+    path: yacht_dice/test-*
+dataset_info:
+- config_name: array_formula
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 13407331
+    num_examples: 600
+  download_size: 3956244
+  dataset_size: 13407331
+- config_name: causal_dag
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 10991552
+    num_examples: 600
+  download_size: 2486519
+  dataset_size: 10991552
+- config_name: cipher
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 2191175
+    num_examples: 600
+  download_size: 277124
+  dataset_size: 2191175
+- config_name: cryptarithmetic
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 2697176
+    num_examples: 600
+  download_size: 545993
+  dataset_size: 2697176
+- config_name: inequality
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 2259573
+    num_examples: 600
+  download_size: 436357
+  dataset_size: 2259573
+- config_name: jamo
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 384368
+    num_examples: 300
+  download_size: 52641
+  dataset_size: 384368
+- config_name: kinship
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: choices
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 3188738
+    num_examples: 300
+  download_size: 496080
+  dataset_size: 3188738
+- config_name: korean_units
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 1401654
+    num_examples: 300
+  download_size: 639655
+  dataset_size: 1401654
+- config_name: minesweeper
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 2570080
+    num_examples: 600
+  download_size: 424457
+  dataset_size: 2570080
+- config_name: number_baseball
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 1299609
+    num_examples: 600
+  download_size: 220324
+  dataset_size: 1299609
+- config_name: saju
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 180478
+    num_examples: 300
+  download_size: 26018
+  dataset_size: 180478
+- config_name: sat_puzzles
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 28051763
+    num_examples: 600
+  download_size: 4931284
+  dataset_size: 28051763
+- config_name: sudoku
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 1446800
+    num_examples: 600
+  download_size: 291709
+  dataset_size: 1446800
+- config_name: time
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 258002
+    num_examples: 300
+  download_size: 39819
+  dataset_size: 258002
+- config_name: yacht_dice
+  features:
+  - name: id
+    dtype: string
+  - name: task
+    dtype: string
+  - name: language
+    dtype: string
+  - name: difficulty
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: solution
+    dtype: string
+  splits:
+  - name: test
+    num_bytes: 2112133
+    num_examples: 600
+  download_size: 374101
+  dataset_size: 2112133
+---
+
+# NOLLI
+
+**NOLLI (논리)** is a benchmark of rule-based logical puzzles in **Korean and English**, designed with
+difficulty calibration in mind. It covers **15 task types x 3 difficulty tiers (easy / medium / hard)
+x 100 examples** each, for a total of **7,500 examples** (EN 3,000 + KO 4,500).
+
+All examples are programmatically generated with solver-verified answers. If you need more data,
+the generators in the GitHub repository can produce additional examples for any task, language,
+and difficulty tier.
+
+📄 Paper: [https://arxiv.org/abs/2608.04397](https://arxiv.org/abs/2608.04397)\
+💻 Code: [https://github.com/HAE-RAE/NOLLI](https://github.com/HAE-RAE/NOLLI)
+
+## Tasks
+
+Tasks follow a three-level cross-lingual spectrum: **direct translations**, **script adaptations**,
+and **Korean-only** tasks.
+
+### 1. Direct Translations (8 tasks · EN + KO)
+
+Same generators and parameters in both languages, so performance gaps isolate presentation language.
+
+| task | description | en (easy/med/hard) | ko (easy/med/hard) | total |
+|---|---|---|---|---|
+| `array_formula` | Apply row/column aggregation formulas (SUM/MEAN/MAX/...) to a 2D array in sequence | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `causal_dag` | Infer event-propagation time through a causal graph with per-edge delays | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `inequality` | CSP: place numbers 1..N satisfying inequality constraints (unique solution) | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `minesweeper` | Minesweeper with minimal hints preserving a unique solution | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `number_baseball` | Infer a hidden N-digit number from Strike/Ball hints | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `sat_puzzles` | Boolean satisfiability (CNF) framed as a natural-language scenario | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `sudoku` | 9x9 Sudoku with guaranteed-unique solutions; graded on the digits at 4-6 spot-checked cells, not the full grid | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `yacht_dice` | Assign 12 dice rolls to 12 scoring categories to maximize total score; graded on the partial sum over spot-checked rounds, not the 12-round total | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+
+### 2. Script Adaptations (2 tasks · EN + KO)
+
+Not translation-equivalent: English uses Roman letters; Korean uses Hangul *jamo*.
+Calibrated independently per language.
+
+| task | description | en (easy/med/hard) | ko (easy/med/hard) | total |
+|---|---|---|---|---|
+| `cipher` | Decode stacked ciphers (Substitution, Vigenere, Reverse, Playfair, Transposition) | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+| `cryptarithmetic` | Letters-for-digits arithmetic (SEND+MORE=MONEY style); Korean maps *jamo* groups | 100 / 100 / 100 | 100 / 100 / 100 | 600 |
+
+### 3. Korean-Only (5 tasks · KO)
+
+No English counterpart — built on structures specific to the Korean language and culture
+(kinship terminology, the sexagenary calendar, traditional units, Hangul orthography).
+
+| task | description | ko (easy/med/hard) | total |
+|---|---|---|---|
+| `jamo` | Decompose a Hangul syllable into 초성/중성/종성, shift 초성, recompose | 100 / 100 / 100 | 300 |
+| `kinship` | Infer a Korean kinship term from a chain of family relationships (26-way multiple choice) | 100 / 100 / 100 | 300 |
+| `korean_units` | Convert traditional Korean units (평·마지기·되·자·냥, ...) with randomized rates | 100 / 100 / 100 | 300 |
+| `saju` | Compute the four 사주(四柱) pillars from a birth date/time | 100 / 100 / 100 | 300 |
+| `time` | Korean calendar reasoning: holiday anchor + relative-day expression → date / 60갑자 일진 | 100 / 100 / 100 | 300 |
+
+## Usage
+
+```python
+from datasets import load_dataset, get_dataset_config_names
+
+# one task — `kinship` is one of the five Korean-only tasks
+ds = load_dataset("HAERAE-HUB/NOLLI", "kinship", split="test")
+
+# language and difficulty are columns; the 10 bilingual tasks carry both en and ko
+cipher = load_dataset("HAERAE-HUB/NOLLI", "cipher", split="test")
+ko_hard = cipher.filter(lambda x: x["language"] == "ko" and x["difficulty"] == "hard")
+
+# all 15 tasks
+tasks = get_dataset_config_names("HAERAE-HUB/NOLLI")
+all_ds = {t: load_dataset("HAERAE-HUB/NOLLI", t, split="test") for t in tasks}
+```
+
+Note that `kinship` carries an extra `choices` column, so the configs cannot be merged
+with `concatenate_datasets` without first aligning the schema.
+
+## Structure
+
+Each config corresponds to one task. All configs share a single `test` split and the same schema
+(the `choices` column exists only in `kinship`):
+
+| column | type | description |
+|---|---|---|
+| `id` | str | unique example id, e.g. `sudoku_ko_0000` |
+| `task` | str | task name (same as config name) |
+| `language` | str | `ko` or `en` |
+| `difficulty` | str | `easy` / `medium` / `hard` |
+| `question` | str | full prompt for the model |
+| `choices` | str | JSON-encoded answer choices — only present in the `kinship` config |
+| `answer` | str | gold answer — format varies by task, see [Answer format](#answer-format) |
+| `solution` | str | step-by-step reference solution |
+
+## Answer format
+
+`answer` is not free text — the format is task-specific, and `question` already tells the model
+which one to produce. Two things to know before scoring.
+
+**Final line.** Seven tasks require a literal `Answer: <value>` last line — `inequality`, `jamo`,
+`korean_units`, `minesweeper`, `number_baseball`, `sudoku`, `yacht_dice`. The other eight take the
+bare value with no prefix.
+
+**Comparison.** Exact string match, with three exceptions: `sat_puzzles` (a JSON object) and
+`minesweeper` (a coordinate list) are order-insensitive, and `korean_units` carries a unit suffix
+that varies per item (`465373918평`).
+
+## Citation
+
+```bibtex
+@misc{choi2026nolli,
+      title={NOLLI: A Difficulty-Calibrated Puzzle Benchmark for Diagnosing the English-Korean Performance Gap},
+      author={Dasol Choi and Joonyong Park and Daegon Yu and Soo Yong Kim and Youngsook Song and Seunghyeok Hong},
+      year={2026},
+      eprint={2608.04397},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL},
+      url={https://arxiv.org/abs/2608.04397},
+}
+```