Word List Sources and Methodology

This page documents what the word-list tools measure, where the current source baseline comes from, and where editorial judgment still applies. It is a reproducibility record, not a claim that every English word belongs in the game or that a listed word is likely to be an answer.

Source snapshot and license

MIT License

Copyright (c) 2020 Wordnik

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

The upstream MIT license permits use, copying, modification, distribution, sublicensing, and sale, but copies or substantial portions must retain the Wordnik copyright and permission notice. The notice is preserved with this project. A license establishes reuse rights; it does not establish a word’s frequency, familiarity, spelling preference, suitability as a puzzle answer, or acceptance by another game.

The source projection is produced deterministically: each upstream JSON-string line is parsed, then only six-character ASCII alphabetic entries are retained. Entries are lowercased, deduplicated, sorted in code-point order, and stored one per line. Checksums protect the upstream file, its license, the filtered projection, and the ordered site pools.

Accepted guesses and answers are different pools

The accepted-guess pool is the broader validation set used when a player submits a word and when a finder page lists candidates. The smaller answer pool supplies this site’s daily and unlimited solutions. A finder row marked “In the answer pool” only reports membership in that smaller local pool.

The two roles should not be confused. Accepted does not mean recommended as an opener. Answer-pool membership does not mean common in general English, approved by Wordnik, present in another game’s private list, or likely to appear next. The JSON field named common is retained for implementation compatibility; in the public interface it means answer-pool membership, not measured language frequency.

Letter-coverage score

The finder score is a deterministic comparison heuristic. For each letter, the script first calculates its presence rate: the number of answers containing that letter at least once divided by the answer-pool size. Repeated copies in one answer do not raise that letter’s presence count. A candidate’s raw coverage is the sum of the presence rates for its distinct letters.

Raw coverage is normalized between the minimum and maximum observed across the accepted-guess pool. The displayed integer is:

round(75 × normalized coverage + 15 × unique letters / 6 + 10 × vowel-balance weight)

The vowel-balance weights for zero through six A/E/I/O/U vowels are 0.10, 0.65, 1.00, 0.85, 0.50, 0.20, 0.10. Y is not counted in that component. Because normalization depends on the current pools, a pool change can alter scores even when a particular spelling stays the same.

This score does not use corpus frequency, clue meaning, position-specific tile feedback, or future-answer odds. Once a board supplies green, yellow, gray, or repeated-letter evidence, those constraints take priority over the score.

Starter-word experiment

The separate starter experiment evaluates the complete feedback partition produced by each accepted guess. Every answer is placed into a feedback bucket for that guess, then three reported metrics describe the partition.

Entropy favors a more even spread of feedback patterns. Expected remaining candidates describes the average bucket size after one guess when each answer in this site’s pool is treated equally. Worst-case bucket reports the largest unresolved group. They answer different questions, so the ranking uses a documented tie-break sequence rather than claiming one metric is universally best.

Generation and verification

The playable pools live in static/js/words.js. Generated finder files are filtered views of those pools, not separately maintained dictionaries. The data checks verify six-letter format, uniqueness, answer membership within accepted guesses, pinned-source coverage, ordered-pool checksums, generated row fields, score reproduction, and starter-experiment reproduction.

When a pool changes, the maintainer reviews the intended edit, updates the version fingerprints deliberately, regenerates every dependent file, spot-checks the results, and runs the full test and production build. Generated JSON should not be edited by hand.

Editorial review, corrections, and limits

The site maintainer is responsible for reviewing answer-pool changes and correction reports. The answer pool aims for recognizable, general-audience words; the accepted-guess pool is intentionally broader. That distinction is an editorial policy, not a dictionary completeness claim. Automated source coverage cannot determine whether a word is fair, offensive in context, too specialized, or familiar to most players, so those decisions still require review.

Report a questionable word, classification, score, or source statement through the contact page or email [email protected]. Please include the word, page URL, observed result, and the correction you propose. Reports are checked against the pinned source and generated data before a change is made; submissions do not update the game automatically.

Changelog

• 2026-08-09 — provenance baseline and experiment v1: pinned the Wordnik source commit and MIT license, recorded the 18,195-entry six-letter projection, prospectively reconciled the existing pools with ten documented replacements, published pool fingerprints, and added the versioned starter-feedback experiment. This entry does not claim the historical pools originally came from Wordnik.

Future source, pool, formula, or algorithm changes will receive a dated entry here. A date changes only when the underlying evidence or method changes, not simply to make the page appear fresh.