The xd Crossword Corpus

Access to the corpus

Draft, November 2026. Marked [established] where it describes what we already do, [proposed] where it is new.

The puzzles in the corpus are copyrighted by their publishers and constructors. We hold them for research, under the terms below, and we do not publish them. This document says what is available, to whom, and on what terms.

What the corpus is for

The corpus exists so that crosswords can be studied: their language, their history, their construction, their editing, and the people who made them. [established] It is not a place to get puzzles to solve. Puzzles for solving are available from their publishers, and we point people there.

Forms of access

From most to least open:

1. Public results. Statistics, reports and analyses at xd.saul.pw, and search at The Cross(word) Reference, where puzzles are shown as references with their answers filled in. No request needed. Publications can ask to be left out of the site (see CONTRIBUTING.md). [established]
2. Derived data. Datasets built from the corpus that do not reproduce puzzles: clue and answer pairs, answer frequencies, grid statistics, publication and constructor metadata. Provided on request for a stated research purpose, and in some cases published openly. [proposed] Which derived datasets can be public is decided case by case and recorded in DECISIONS.md.
3. Samples. A random or targeted sample of complete puzzles, for a purpose that needs whole puzzles but not all of them, such as evaluating a solver. [established]
4. The full corpus. Membership in the private repository, for research that needs it, and for contributors who are adding puzzles. [established]

How to ask

Open an issue in this repository or contact a maintainer, saying who you are, what you want to do, and which form of access you need. We answer within a few weeks. We say yes to most research requests; the questions we ask are whether the purpose is research, whether a less open form of access would do, and whether the result could put puzzles back into circulation.

The maintainers decide together (see GOVERNANCE.md). Each grant is recorded privately: who, when, what form of access, and for what purpose. [proposed]

Terms of use

By accepting access to the corpus or to derived data you agree:

• To use it for research and not for solving, republishing, or any product that delivers puzzles to solvers.
• Not to give it, or any substantial part of it, to anyone else. People who need it can ask us.
• To cite the corpus in anything you publish that uses it (see README.md).
• To tell us what you find. We like to hear about it, and it helps us show the corpus is worth keeping.
• To delete your copy if we ask, for example after a rights-holder complaint.

Access is personal. A grant to one person or group does not extend to their collaborators, employer or successors. [established]

Publishers and constructors

If you publish or constructed puzzles that are in the corpus, we are glad to hear from you. We will tell you what we hold from your publication, correct any errors you find, and record the constructor’s name as they wish it recorded (see EDITORIAL.md, “Headers”).

If you want puzzles removed, say which and why, and we will remove them. The removal is recorded (see PRESERVATION.md, “Removing puzzles”), and anyone with a copy is asked to delete them. [proposed]

Contributors

Editors, constructors and publishers who want to give us puzzles should read CONTRIBUTING.md; nothing there requires access to the repository. Volunteers who convert and submit puzzles themselves get access in order to do so, and the same terms apply to the rest of the corpus. The import guide is in the private corpus repository.

Review

This policy is reviewed once a year with the project’s preservation assessment (see assessments/). Changes are logged in DECISIONS.md.