Dataset Viewer
The dataset viewer is not available for this dataset.
Cannot get the config names for the dataset.
Error code:   ConfigNamesError
Exception:    FileNotFoundError
Message:      Couldn't find any data file at /src/services/worker/zhugetd/chainle. Couldn't find 'zhugetd/chainle' on the Hugging Face Hub either: FileNotFoundError: Unable to find 'hf://datasets/zhugetd/chainle@cbbcab15871df1f13bedb72615574c4d3d5cc5a8/data/rounds.jsonl' with any supported extension ['.csv', '.tsv', '.json', '.jsonl', '.ndjson', '.parquet', '.geoparquet', '.gpq', '.arrow', '.txt', '.conll', '.conllu', '.tar', '.xml', '.hdf5', '.h5', '.eval', '.toml', '.md', '.lance', '.tsfile', '.vortex', '.fa', '.fasta', '.fna', '.ffn', '.faa', '.frn', '.afa', '.gb', '.gbk', '.genbank', '.fq', '.fastq', '.pdb', '.ent', '.cif', '.mmcif', '.PDB', '.ENT', '.blp', '.bmp', '.dib', '.bufr', '.cur', '.pcx', '.dcx', '.dds', '.ps', '.eps', '.fit', '.fits', '.fli', '.flc', '.ftc', '.ftu', '.gbr', '.gif', '.grib', '.png', '.apng', '.jp2', '.j2k', '.jpc', '.jpf', '.jpx', '.j2c', '.icns', '.ico', '.im', '.iim', '.tif', '.tiff', '.jfif', '.jpe', '.jpg', '.jpeg', '.mpg', '.mpeg', '.msp', '.pcd', '.pxr', '.pbm', '.pgm', '.ppm', '.pnm', '.psd', '.bw', '.rgb', '.rgba', '.sgi', '.ras', '.tga', '.icb', '.vda', '.vst', '.webp', '.wmf', '.emf', '.xbm', '.xpm', '.BLP', '.BMP', '.DIB', '.BUFR', '.CUR', '.PCX', '.DCX', '.DDS', '.PS', '.EPS', '.FIT', '.FITS', '.FLI', '.FLC', '.FTC', '.FTU', '.GBR', '.GIF', '.GRIB', '.PNG', '.APNG', '.JP2', '.J2K', '.JPC', '.JPF', '.JPX', '.J2C', '.ICNS', '.ICO', '.IM', '.IIM', '.TIF', '.TIFF', '.JFIF', '.JPE', '.JPG', '.JPEG', '.MPG', '.MPEG', '.MSP', '.PCD', '.PXR', '.PBM', '.PGM', '.PPM', '.PNM', '.PSD', '.BW', '.RGB', '.RGBA', '.SGI', '.RAS', '.TGA', '.ICB', '.VDA', '.VST', '.WEBP', '.WMF', '.EMF', '.XBM', '.XPM', '.aiff', '.au', '.avr', '.caf', '.flac', '.htk', '.svx', '.mat4', '.mat5', '.mpc2k', '.ogg', '.paf', '.pvf', '.raw', '.rf64', '.sd2', '.sds', '.ircam', '.voc', '.w64', '.wav', '.nist', '.wavex', '.wve', '.xi', '.mp3', '.opus', '.3gp', '.3g2', '.avi', '.asf', '.flv', '.mp4', '.mov', '.m4v', '.mkv', '.webm', '.f4v', '.wmv', '.wma', '.ogm', '.mxf', '.nut', '.AIFF', '.AU', '.AVR', '.CAF', '.FLAC', '.HTK', '.SVX', '.MAT4', '.MAT5', '.MPC2K', '.OGG', '.PAF', '.PVF', '.RAW', '.RF64', '.SD2', '.SDS', '.IRCAM', '.VOC', '.W64', '.WAV', '.NIST', '.WAVEX', '.WVE', '.XI', '.MP3', '.OPUS', '.3GP', '.3G2', '.AVI', '.ASF', '.FLV', '.MP4', '.MOV', '.M4V', '.MKV', '.WEBM', '.F4V', '.WMV', '.WMA', '.OGM', '.MXF', '.NUT', '.glb', '.ply', '.stl', '.GLB', '.PLY', '.STL', '.pdf', '.PDF', '.nii', '.NII', '.zip', '.idx', '.manifest', '.txn']
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/dataset/config_names.py", line 67, in compute_config_names_response
                  config_names = get_dataset_config_names(
                      path=dataset,
                      token=hf_token,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 161, in get_dataset_config_names
                  dataset_module = dataset_module_factory(
                      path,
                  ...<4 lines>...
                      **download_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/load.py", line 1178, in dataset_module_factory
                  raise FileNotFoundError(
                  ...<2 lines>...
                  ) from None
              FileNotFoundError: Couldn't find any data file at /src/services/worker/zhugetd/chainle. Couldn't find 'zhugetd/chainle' on the Hugging Face Hub either: FileNotFoundError: Unable to find 'hf://datasets/zhugetd/chainle@cbbcab15871df1f13bedb72615574c4d3d5cc5a8/data/rounds.jsonl' with any supported extension ['.csv', '.tsv', '.json', '.jsonl', '.ndjson', '.parquet', '.geoparquet', '.gpq', '.arrow', '.txt', '.conll', '.conllu', '.tar', '.xml', '.hdf5', '.h5', '.eval', '.toml', '.md', '.lance', '.tsfile', '.vortex', '.fa', '.fasta', '.fna', '.ffn', '.faa', '.frn', '.afa', '.gb', '.gbk', '.genbank', '.fq', '.fastq', '.pdb', '.ent', '.cif', '.mmcif', '.PDB', '.ENT', '.blp', '.bmp', '.dib', '.bufr', '.cur', '.pcx', '.dcx', '.dds', '.ps', '.eps', '.fit', '.fits', '.fli', '.flc', '.ftc', '.ftu', '.gbr', '.gif', '.grib', '.png', '.apng', '.jp2', '.j2k', '.jpc', '.jpf', '.jpx', '.j2c', '.icns', '.ico', '.im', '.iim', '.tif', '.tiff', '.jfif', '.jpe', '.jpg', '.jpeg', '.mpg', '.mpeg', '.msp', '.pcd', '.pxr', '.pbm', '.pgm', '.ppm', '.pnm', '.psd', '.bw', '.rgb', '.rgba', '.sgi', '.ras', '.tga', '.icb', '.vda', '.vst', '.webp', '.wmf', '.emf', '.xbm', '.xpm', '.BLP', '.BMP', '.DIB', '.BUFR', '.CUR', '.PCX', '.DCX', '.DDS', '.PS', '.EPS', '.FIT', '.FITS', '.FLI', '.FLC', '.FTC', '.FTU', '.GBR', '.GIF', '.GRIB', '.PNG', '.APNG', '.JP2', '.J2K', '.JPC', '.JPF', '.JPX', '.J2C', '.ICNS', '.ICO', '.IM', '.IIM', '.TIF', '.TIFF', '.JFIF', '.JPE', '.JPG', '.JPEG', '.MPG', '.MPEG', '.MSP', '.PCD', '.PXR', '.PBM', '.PGM', '.PPM', '.PNM', '.PSD', '.BW', '.RGB', '.RGBA', '.SGI', '.RAS', '.TGA', '.ICB', '.VDA', '.VST', '.WEBP', '.WMF', '.EMF', '.XBM', '.XPM', '.aiff', '.au', '.avr', '.caf', '.flac', '.htk', '.svx', '.mat4', '.mat5', '.mpc2k', '.ogg', '.paf', '.pvf', '.raw', '.rf64', '.sd2', '.sds', '.ircam', '.voc', '.w64', '.wav', '.nist', '.wavex', '.wve', '.xi', '.mp3', '.opus', '.3gp', '.3g2', '.avi', '.asf', '.flv', '.mp4', '.mov', '.m4v', '.mkv', '.webm', '.f4v', '.wmv', '.wma', '.ogm', '.mxf', '.nut', '.AIFF', '.AU', '.AVR', '.CAF', '.FLAC', '.HTK', '.SVX', '.MAT4', '.MAT5', '.MPC2K', '.OGG', '.PAF', '.PVF', '.RAW', '.RF64', '.SD2', '.SDS', '.IRCAM', '.VOC', '.W64', '.WAV', '.NIST', '.WAVEX', '.WVE', '.XI', '.MP3', '.OPUS', '.3GP', '.3G2', '.AVI', '.ASF', '.FLV', '.MP4', '.MOV', '.M4V', '.MKV', '.WEBM', '.F4V', '.WMV', '.WMA', '.OGM', '.MXF', '.NUT', '.glb', '.ply', '.stl', '.GLB', '.PLY', '.STL', '.pdf', '.PDF', '.nii', '.NII', '.zip', '.idx', '.manifest', '.txn']

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Chainle: Word Connections and Reproducible Scoring

Chainle asks you to find one English word that connects the displayed clues. This dataset turns the independently implemented Chainle game into inspectable examples: real puzzle rounds, legal vocabulary, computed scores, and input checks. Use it to study how a word connection game evaluates balanced answers, reproduce a scoring decision, or build a small educational exercise. The release contains machine labels, not collected player responses.

What problem does this dataset solve?

A word association game can reward a plausible connection without having one uniquely correct answer. That makes Chainle evaluation difficult: a reference word alone cannot explain why another answer earns partial credit. Chainle supplies deterministic scores and separate connections to each clue, giving developers evidence beyond a binary answer key.

The dataset documents Chainle at playchainle.org, an independent implementation with its own model and puzzle bank. It is not affiliated with chainle.io. Chainle scores here describe this version only; they do not promise agreement with another website, a dictionary definition, or human judgments.

For players, the useful question is how to connect every clue. For builders, it is how to reproduce that decision. Chainle supports both through transparent examples, while keeping daily play available at the website. Search intent informed this organization; no search volume, popularity, or ranking claim is implied.

How the five rounds work

A Chainle session has five rounds with two, two, two, three, and one clue respectively. Submit one valid English word per round. Each round awards an integer from zero to 1,000, so the maximum session total is 5,000. Invalid entries do not consume a valid submission.

Chainle trims surrounding whitespace and lowercases input before checking it. Internal spaces, punctuation, and digits are rejected. Copying a clue or using certain close forms is prohibited; spelling variants and root relationships receive additional checks. A word appearing in ordinary English is not sufficient: Chainle also requires membership in its playable vocabulary.

In the three clue round, Chainle adjusts exactly one weakest connection before combining the connections. With two clues, an extremely weak connection can pull down an otherwise strong answer. The final Chainle single clue round still rejects the clue itself and specified related forms. These differences matter when interpreting a word association game score.

Daily Chainle changes at midnight UTC. The site also offers an archive and unlimited practice. Unlimited describes continued play, not an infinite set of distinct questions: the finite bank can repeat. Clues and submissions use English. Retired Chinese URLs redirect to the current English site; this release does not provide a Chinese vocabulary task.

Choosing better connecting words

Read all clues before answering Chainle. List a few concrete candidates and ask how each candidate relates to every clue. A familiar bridge is often more useful than an obscure word strongly related to just one clue. Try the daily word connection game before opening answer records if you want to avoid spoilers.

Treat Chainle feedback as evidence about its model. A surprising low score does not prove your association is meaningless. Likewise, a high score does not establish a universal language fact. Compare per clue similarities and reaches, then consider whether the model favors one sense of an ambiguous word.

Chainle displays up to eight links for each connection. These Chainle links are rounded visual feedback, so eight visible links do not guarantee 1,000 points. When comparing answers, use the stored integer points and full precision reaches. This distinction prevents a common mistake when testing a word connection game interface.

Configurations and fields

This Chainle snapshot contains four JSONL configurations. Each line is one UTF-8 JSON object, with stable IDs and native numeric, Boolean, list, and nullable string values. The data dictionary lists every field, its type, and its interpretation.

Configuration Rows Split Purpose
rounds 240 test Clues, theme, reference answer, review status
scoring 4,803 test Computed candidate scores and connections
validation 1,920 test Input normalization and rejection examples
lexicon 24,110 train Playable words, frequencies, lexical relations

The Chainle rounds configuration uses published round IDs. Chainle scoring and validation IDs append a word or case name to that ID; lexicon IDs begin with word--. Shared version fields pin the model, and scored records also identify the rules version. Answer lists are examples, not exhaustive inventories of acceptable Chainle guesses.

The Chainle scoring configuration includes published top candidates and eight deterministically selected playable words per round, deduplicated within each round. Selection labels distinguish reference answers, published candidates, and hash samples. This design provides useful high scoring comparisons and ordinary vocabulary coverage without manufacturing player histories.

All Chainle benchmark configurations use a single test split. The lexicon train split is a reference inventory, not supervised training evidence. Chainle clues, answers, and related words overlap across configurations. Do not train on one configuration and report another as an independent held out benchmark without constructing and documenting your own leakage controls.

A real record and loading examples

One Chainle round presents trade and uses market as its reference answer. The authoritative scorer gives that answer 1,000 points. Multiple answers can achieve full credit; the reference is not a claim of uniqueness. The corresponding rounds record includes these fields:

{
  "id": "r-993d4161b48fa672",
  "anchors": ["trade"],
  "anchor_count": 1,
  "theme": "society",
  "reference_word": "market",
  "reference_points": 1000,
  "review_status": "machine_validated"
}

The Chainle example is a field subset; complete records contain provenance fields too. To load Chainle locally, install the optional Python dependencies, then run this from the dataset directory:

python -m pip install -r requirements.txt
from datasets import load_dataset

scores = load_dataset(
    "json", data_files={"test": "data/scoring.jsonl"}, split="test"
)
print(scores[0]["anchors"], scores[0]["points"])

No Hugging Face repository has been created for this Chainle release. The files are already local: no account, token, or download is necessary for this example. After an owner publishes a repository, its README configuration names can be passed to load_dataset with that actual repository ID. Do not treat a placeholder identifier as an existing download.

Applications and evaluation

Use Chainle records for scorer regression checks, input handling tests, vocabulary exploration, and explanations of multi clue tradeoffs. In a classroom, hide reference answers, ask learners to propose connections, and compare their reasoning with model feedback. A word association game can prompt discussion without turning automated scores into grades for language ability.

For a Chainle scoring predictor, report absolute point error by clue count and candidate selection group. For input checking, report agreement with valid flags and exact error codes. Candidate rankings are competition ranks among eligible vocabulary entries, with spelling variant deduplication; they are not player percentiles. A Chainle top candidate sample cannot estimate ordinary player performance.

Sources and reproducible generation

Chainle derives semantic data from ConceptNet Numberbatch 19.08, frequency information from wordfreq 3.1.1, and lexical relations from WordNet 3.0. The Chainle export uses the project's already published derived resources, not scraped answers from another game. No external comparison dataset contributes records.

Generation calls the existing Chainle scorer against versioned little endian tables. Each quantized similarity decodes as the squared ratio of its stored integer to 65,535. Reaches use reference normalization, clipping, and a power transform; a harmonic mean produces points, with family and variant handling supplied by the actual implementation. This Chainle dataset preserves those results rather than substituting generic cosine similarity.

From the project root, run node datasets/chainle-word-connections/scripts/generate.mjs, then the adjacent validate.mjs. Chainle source paths, byte lengths, and SHA-256 values appear in source-manifest.json. Regeneration requires this project and its pinned assets; loading exported records does not. Package checksums detect changed deliverables, and the validation report records actual checks.

Limitations, license, and attribution

Chainle rounds are machine validated and have not received a complete human fairness or difficulty review. Embeddings and frequency sources can encode historical, cultural, and demographic biases. Vocabulary may contain sensitive or offensive terms and is not certified for children. This small bank does not cover English comprehensively or establish a general reasoning benchmark.

Chainle exports no accounts, cookies, payments, submissions, or participant measurements. Theme labels describe the existing bank; they are not a universal ontology. The Chainle package excludes daily schedules and raw embeddings, but its answers can still spoil puzzles that reuse these rounds.

The Chainle derived data is licensed CC BY-SA 4.0. Credit Chainle, ConceptNet Numberbatch contributors Robyn Speer, Joshua Chin, and Catherine Havasi, wordfreq author Robyn Speer and its credited data providers, and Princeton University's WordNet. Preserve the supplied notices and share alike obligations. Upstream licenses remain applicable; the project software and third party media are not relicensed by this card. See LICENSE.md and source details.

FAQ and further reading

Is this an official export from chainle.io? No. Chainle here refers to the independent playchainle.org implementation and the precise resource version in the manifest.

Can I use this to beat today's Chainle game? Answers may overlap a daily puzzle, but no date to answer schedule is included. Use the archive for practice and avoid dataset answers before playing if you prefer an unspoiled challenge.

Does a perfect score mean everyone would choose that word? No. A Chainle game score expresses a deterministic rule applied to model data, not agreement from human participants.

Where should I start? Play Chainle, inspect one round and its candidate records, then read the reproduction guide. For broader context, compare the documented limits with Hugging Face's dataset card guidance.

Downloads last month
47