Publish Cultural Riddles dataset from the source repository

LICENSE

@@ -0,0 +1,31 @@
+CulturalRiddles data terms
+
+These terms apply to the benchmark data and released annotations, responses,
+verdicts, and aggregate data in this reproduction bundle. They do not apply to
+the repository's own code, which is licensed under Apache-2.0 in ../LICENSE,
+or to separately credited third-party assets.
+
+The project's existing data policy combines the attribution terms of Creative
+Commons Attribution 4.0 International with the additional evaluation-only
+restriction below. The combined policy is custom; it is not unmodified
+CC-BY-4.0 and must not be labeled as unmodified CC-BY-4.0.
+
+Attribution terms
+
+Creative Commons Attribution 4.0 International:
+https://creativecommons.org/licenses/by/4.0/legalcode
+
+Credit CulturalRiddles and its contributors, link to the dataset and these
+terms, and indicate any changes. Dataset:
+https://huggingface.co/datasets/CohereLabsCommunity/CulturalRiddles
+
+Additional evaluation-only restriction
+
+This data must not be used to train, fine-tune, or distill models, nor as
+seeds for synthetic data generation. It is an evaluation benchmark. Using
+it as training data defeats its purpose and contaminates future measurement.
+
+This file records the existing project policy; it does not remove or narrow
+rights already granted under an earlier release. The additional restriction
+is not a condition in the standard Creative Commons license. No claim is made
+here that adding this notice changes an existing Creative Commons grant.

README.md

@@ -1,5 +1,7 @@
 ---
-license: cc-by-4.0
+license: other
+license_name: culturalriddles-evaluation-only
+license_link: LICENSE
 pretty_name: Cultural Riddles
 language:
 - aeb
@@ -57,8 +59,6 @@ task_categories:
 - question-answering
 multilinguality:
 - multilingual
-size_categories:
-- 1K<n<10K
 annotations_creators:
 - expert-generated
 language_creators:
@@ -73,111 +73,165 @@ tags:
 - multicultural
 - community-curated
 - low-resource-languages
+- human-evaluation
 configs:
-- config_name: default
+- config_name: riddles
+  default: true
   data_files:
   - split: test
-    path: data/test-*
+    path: data/riddles.parquet
+- config_name: model_answers
+  data_files:
+  - split: test
+    path: data/model_answers.parquet
+- config_name: human_annotations
+  data_files:
+  - split: test
+    path: data/human_annotations.parquet
+- config_name: models
+  data_files:
+  - split: test
+    path: data/models.parquet
 ---
 
 # Cultural Riddles
 
-**Explore the results:** [Cultural Riddles dashboard](https://huggingface.co/spaces/CohereLabsCommunity/CulturalRiddles)
-
-A community-built benchmark of original cultural riddles spanning **62 communities across 51 languages and 45 countries**, written by 166 native speakers. Each riddle describes something without ever naming it, so answering it requires genuine cultural knowledge rather than translation or web search. A native speaker would recognise the answer instantly; a model only succeeds if it carries that culture's knowledge.
+**Explore the results:** [dashboard](https://huggingface.co/spaces/CohereLabsCommunity/CulturalRiddles) · **Play:** [try the riddles yourself](https://culturalriddles.ai/) · **Code:** [github.com/Cohere-Labs-Community/CulturalRiddles](https://github.com/Cohere-Labs-Community/CulturalRiddles)
 
-## Dataset summary
+A community-built benchmark of original cultural riddles from **62 communities across 51 languages and 45 countries**, written by 167 native speakers. Each riddle describes something without naming it, so answering it takes cultural knowledge, not translation or web search. A native speaker recognises the answer at once; a model succeeds only if it carries that culture's knowledge.
 
-Most multilingual benchmarks are written in English and translated outward, which tests language but not culture. This benchmark inverts that. Native speakers from each community write their own riddles from scratch, using a shared English framework only as a starting point, not as a source to translate. The riddles are intentionally indirect: they use local nicknames, synonyms, and oblique description so that the answer cannot be reached by recognising a keyword. The goal is depth of cultural understanding, measured where today's models tend to be fluent but shallow.
+The release has four tables:
 
-This release contains **6191 riddles** across 62 communities, of which 1234 come from community-chosen wildcard topics. It is intended for evaluation only.
+| Table | Rows | What it holds |
+|---|---:|---|
+| `riddles` (default) | 6,191 | The riddles and their reference answers. |
+| `model_answers` | 170,520 | Every answer from 28 model configurations on the 6,090 benchmarked riddles, with its final score and who scored it. |
+| `human_annotations` | 74,383 | Every native-speaker label on a unique answer, including independent second labels. |
+| `models` | 28 | The model configurations, their sizes, and their recorded reasoning settings. |
 
 ## Why riddles
 
-A riddle names nothing. "This red-crossed sphere's juice is not sweet for your stomach" means nothing to a search engine, but every Hungarian recognises Unicum, a famous bitters. A model that only knows the formal name misses it entirely. Because each riddle is original, unpublished, and culturally specific, it resists both memorisation and reverse-engineering from training data, which gives the benchmark a longer useful life than fact-lookup tasks.
+A riddle names nothing. "This red-crossed sphere's juice is not sweet for your stomach" means nothing to a search engine, but every Hungarian recognises Unicum, a famous bitters. A model that only knows the formal name misses it. Because each riddle is original, unpublished, and culturally specific, it resists memorisation and lookup.
 
 ## Languages and communities
 
-The base unit is the present-day national-linguistic community, for example Spanish from Argentina or Bengali from India. 62 communities are represented across 51 languages. Languages outnumber communities because several communities share a language: Argentine, Cuban, and Peruvian riddles are all in Spanish but draw on three different cultures. That separation is the point of the benchmark.
+The unit is the present-day national-linguistic community, for example Spanish in Argentina or Bengali in India. Languages and communities differ because several communities share a language: Argentine, Cuban, and Peruvian riddles are all in Spanish but draw on three cultures. 1,234 riddles come from topics each community chose itself. See [the list of communities](https://github.com/Cohere-Labs-Community/CulturalRiddles/blob/main/docs/communities.md).
 
-The benchmark grows over time. Each community carries a `release_version` marking the collection round it entered in (`v0` is the first round, `v1` the next, and so on), so results can be compared across releases.
+## Loading the data
 
-## Dataset structure
+```python
+from datasets import load_dataset
 
-### Data fields
+riddles = load_dataset("CohereLabsCommunity/CulturalRiddles", split="test")
+answers = load_dataset("CohereLabsCommunity/CulturalRiddles", "model_answers", split="test")
+labels = load_dataset("CohereLabsCommunity/CulturalRiddles", "human_annotations", split="test")
+models = load_dataset("CohereLabsCommunity/CulturalRiddles", "models", split="test")
+
+tamil = riddles.filter(lambda r: r["community"] == "tamil_india")
+```
+
+## Tables
+
+### `riddles`
 
 | Field | Type | Description |
 |---|---|---|
-| `riddle_id` | string | Stable unique id, `{community}_{topic}_{number}`. |
+| `riddle_id` | string | Stable unique ID, `{community}_{topic}_{number}`. |
+| `community` | string | Community key, usually `{language}_{country}` (for example `tamil_india`). |
 | `language` | string | ISO 639-3 code of the riddle's language. |
-| `community` | string | Cultural community, `{culture}_{country}` (e.g. `tamil_india`). |
-| `release_version` | string | Collection round the community was added in (`v0`, `v1`, ...). |
-| `topic` | string | Topic category the riddle belongs to. |
-| `is_wildcard` | bool | True if the riddle is from a community-chosen wildcard topic. |
+| `topic` | string | Topic the riddle belongs to. |
+| `is_wildcard` | bool | True for a topic the community chose itself. |
 | `riddle_original` | string | The riddle in its original language. |
-| `riddle_english` | string | English translation of the riddle. Reference only; may be null. |
-| `reference_answer_original` | string | Gold answer in the original language. |
-| `reference_answer_english` | string | English translation of the answer. May be null. |
+| `riddle_english` | string | English translation, for human reference only. May be null. |
 | `riddle_hint` | string | Optional hint. May be null. |
-| `riddle_note` | string | Optional note from the author. Riddles flagged in the content review also carry a content note here. May be null. |
-
-### Splits
+| `riddle_note` | string | Optional note from the writer; riddles flagged in the content review also carry a content note here. May be null. |
+| `reference_answer_original` | string | Reference answer in the original language. |
+| `reference_answer_english` | string | English translation of the answer. May be null. |
 
-A single `test` split of 6191 rows. `community`, `language`, and `topic` are columns to filter on, not separate splits.
+### `model_answers`
 
-## Loading the dataset
+Each configuration answered each benchmarked riddle once (6,090 riddles from 61 communities). Scores come from rules for clear cases (32,171 answers), from native speakers (100,769), and from two LLM judges that agree (34,546); 3,034 answers have no score. Accuracy is the share of scored answers with score 3.
 
-```python
-from datasets import load_dataset
-
-ds = load_dataset("CohereLabsCommunity/CulturalRiddles", split="test")
+| Field | Type | Description |
+|---|---|---|
+| `riddle_id` | string | Joins to `riddles`. |
+| `community` | string | Community key. |
+| `topic` | string | Topic of the riddle. |
+| `model` | string | Model configuration key; joins to `models`. |
+| `model_display` | string | Readable configuration name. |
+| `model_answer` | string | The answer the model gave. |
+| `answer_norm` | string | `model_answer` stripped and lowercased; joins to `human_annotations`. |
+| `score` | int | 0 incorrect, 1 admits not knowing, 2 partially correct, 3 correct. Null when no score was reached. |
+| `scorer` | string | `rule`, `human` (native speaker), `llm_judges` (two agreeing judges), or `none`. |
+| `routing` | string | How the answer was routed: rule-based categories, human evaluation, or pending and invalid human labels. |
+| `error_type` | string | Answer type code, for example `0.3` hallucinated or `3.1` different valid answer. |
+| `error_type_label` | string | Readable answer type. |
+| `annotator_id` | string | Anonymous native-speaker ID, such as `annotator_k2p9`. Stable across tables. |
+| `annotator_note` | string | The annotator's note, as written. May be null. |
+| `judge_score_deepseek` | int | Score from the DeepSeek judge, where one ran. May be null. |
+| `judge_score_gemini` | int | Score from the Gemini judge, where one ran. May be null. |
+
+### `human_annotations`
+
+Native speakers labeled each unique (riddle, answer) pair once, so one label can cover several models. 70,006 rows are first labels and 4,377 are independent second labels used to measure agreement. 109 annotators contributed; their IDs are anonymous.
 
-# filter to one community
-tamil = ds.filter(lambda r: r["community"] == "tamil_india")
+| Field | Type | Description |
+|---|---|---|
+| `riddle_id` | string | Joins to `riddles`. |
+| `community` | string | Community key. |
+| `topic` | string | Topic of the riddle. |
+| `answer` | string | The answer the annotator labeled. |
+| `answer_norm` | string | `answer` stripped and lowercased; one label covers every model that gave it. |
+| `annotation` | string | `first` label, or `second` label from an independent annotator. |
+| `annotator_id` | string | Anonymous native-speaker ID; several are separated by `;`. |
+| `score` | int | 0 to 3, as in `model_answers`. |
+| `error_type` | string | Answer type code. |
+| `error_type_label` | string | Readable answer type. |
+| `annotator_note` | string | The annotator's note, as written. May be null. |
+| `benchmark_status` | string | `valid` or `invalid` for first labels; invalid labels are not used in scores. |
+| `iaa_row` | float | Set for rows in the inter-annotator agreement sample. |
+| `self_review` | bool | True when an annotator labeled the same pair again. |
+
+### `models`
 
-# filter to one language
-egyptian_arabic = ds.filter(lambda r: r["language"] == "arz")
-```
+| Field | Type | Description |
+|---|---|---|
+| `model` | string | Configuration key used in `model_answers`. |
+| `model_display` | string | Readable configuration name. |
+| `default_reasoning` | string | The provider's documented default reasoning setting. |
+| `actual_reasoning_effort` | string | The setting recorded for the run. `unspecified` means the request did not set it, not that reasoning was off. |
+| `total_parameter_b` | float | Published total parameters, in billions. Null when not published. |
+| `active_parameter_count_b` | float | Published active parameters for mixture-of-experts models, in billions. |
+| `access` | string | `proprietary`, `open-weight`, or `open-source`. |
+| `info_url` | string | Model card or documentation. |
+| `cutoff_date` | string | Published training-data cutoff, where known. |
 
 ## Curation
 
-Every riddle is original human writing. Contributors are native speakers who write for their own community, following a shared format of topics and questions per topic. AI tools may be used for brainstorming or to test how hard a riddle is for a model, but the riddles themselves are written by people, not generated. English translations are provided for human reference and documentation only, never to present the riddle to a model. All riddles are free of personally identifiable information.
+Every riddle is original human writing by native speakers writing for their own community. AI tools may be used for brainstorming or to test difficulty, but the riddles are written by people. English translations are for human reference only, never shown to a model. The data contains no names or contact details of writers or annotators.
 
 ## Intended use and terms
 
-Released under **CC-BY-4.0**: you may share and adapt the data, including commercially, with attribution.
+The data is released under the terms in [LICENSE](LICENSE): Creative Commons Attribution 4.0 attribution terms combined with an evaluation-only restriction.
 
-**This data must not be used to train, fine-tune, or distill models, nor as seeds for synthetic data generation.** It is an evaluation benchmark. Using it as training data defeats its purpose and contaminates future measurement. Please respect this restriction in addition to the license.
+**This data must not be used to train, fine-tune, or distill models, nor as seeds for synthetic data generation.** It is an evaluation benchmark; training on it contaminates future measurement. This applies to the model answers and labels as much as to the riddles.
 
 ## Content note
 
 Although we have filtered the dataset for harmful content, some riddles may still contain sensitive themes or potentially problematic ideas. These are presented solely as cultural and linguistic puzzles, not as endorsements of the views expressed or guidance for real-world behavior.
 
-Riddles flagged in a content review carry a short content note in their `riddle_note` field.
+Riddles flagged in a content review carry a short content note in their `riddle_note` field. Model answers are published as generated and were not filtered: many are hallucinated, and some may be offensive.
 
 ## Contributors
 
 This benchmark was proposed by **Károly Boczka** and **Julia Kreutzer**, and built by native-speaker contributors across the Cohere Labs Community.
 
-Per-community contributor credits will be added once every contributor has confirmed how they would like to be credited.
-
 ## Updates
 
-This card is regenerated from the dataset on each release, so the numbers above
-always reflect the current contents.
-
-- Current release spans 62 communities across collection
-  rounds: `v0` (44 communities), `v1` (4 communities), `v2` (13 communities), `v3` (1 communities).
-- The Spanish/Cuba riddle `spanish_cuba_history_016` was corrected at source
-  (the earlier version paired the riddle with a mismatched answer). The dataset
-  carries the corrected riddle. Model results in the paper leave this riddle out,
-  because the models only saw the earlier version. See
-  `submissions/PUBLISHING.md` in the source repository for how corrections are
-  handled.
-- After a content review, 8 riddles were removed and 5 were corrected (3
-  reference answers and 2 riddle texts). One Mongondow riddle that was written
-  in Indonesian was also removed. Each change is listed in
-  `data_loading/dataset_corrections.csv` in the source repository.
+This card is generated from the [source repository](https://github.com/Cohere-Labs-Community/CulturalRiddles) on each release.
+
+- This release adds the `model_answers`, `human_annotations`, and `models` tables. The riddles moved to `data/riddles.parquet` and no longer carry the `release_version` column.
+- After a content review, some riddles were removed or corrected. Every change is listed in `data/corrections.csv` in the source repository.
 
 ## Citation
 

data/{test-00000-of-00001.parquet → human_annotations.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:31e6a548c06bef0973a876ee990046dac4099210242aa69384b5845a08970cd5
-size 1244453
+oid sha256:30d2964e254a332d821db5904a829c33d5ab95c6063605bd5a79f5d3f6169c7e
+size 2325909

data/model_answers.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26c825e7b8c513623946089fdddf8c6a2818f0246b6feb11c50838a0871ee7b7
+size 5573006

data/models.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:257aa6008b273cd5ecf1892003413a7e1173cf6f145ddc15f852711eaf86a847
+size 8007

data/riddles.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a54bbb9e8c6d270543ecafe7e4f4a90fe45b5c6d3417a40e6ebeaa499fde74de
+size 1243815