Decisions
A dated log of decisions about the corpus and its policies, newest first. One
entry per decision: what was decided, and why, in a few lines. Earlier
decisions live in the corpus repository’s commit history and merge requests.
2026-11-06
- A statement on AI. AI.md says the corpus holds only
human-made puzzles, discloses the maintainers’ own use of AI tools, and
proposes terms against using derived data or samples to generate
crosswords for publication. Reason: concern in the crossword community,
and publishing data on Hugging Face invites it to be consumed by models.
- Clue and answer data stays public.
xd-clues.zip and
xd-metadata.zip remain openly downloadable. Reason: they are already
published, they do not let a puzzle be rebuilt, and withdrawing them would
take away data researchers rely on. (Saul’s position.) The rules for
derived data are in DATASETS.md.
- Dataset venues. Public derived data is published on Hugging Face,
under one organization, and on Kaggle as an equal home: the same snapshots
with the same card text, at the same time. The audience is researchers and
hobbyist programmers alike. Hugging Face also holds the public-domain sets
and the gated samples; Zenodo holds a record with a DOI for the corpus as a
whole; GitHub releases keep the public-domain puzzle sets. Samples of
modern puzzles stay available on request under ACCESS.md, to
hobbyists and researchers alike, so a gated Hugging Face dataset is the
same process with a form. See DATASETS.md.
2026-11-05
- A public home for the project. The corpus gets a public repository,
century-arcade/xdcorpus, holding its policy documents, with the corpus
itself staying private. Reason: the policies could not be shown to
prospective contributors, users or institutions while they lived only in
the private repository. COLLECTION.md and EDITORIAL.md move here; the
operational documents (CONTRIBUTING.md, REVIEWING.md) stay with the corpus.
(Agreed by both maintainers 2026-11-05. The repository name follows the organization’s existing unhyphenated names: xdformat, xdcontrib, xdtalk.)
- The name is “corpus”, not “archive”. The corpus has been cited as the
xd Crossword Corpus since 2019; “archive” describes the scans and originals
(the Every Puzzle Project), while “corpus” describes the normalized research
collection and matches its terms of use.
- First preservation self-assessment. Assessed against the Digital
Preservation Coalition’s Rapid Assessment Model, version 3. Results in
assessments/dpc-ram-2026.md. Priorities: a
second copy of the source archives, a written access policy, a written
preservation policy, and a yearly repeat.
- Governance model. Two maintainers decide; a small group of advisors is
consulted yearly and has no vote; a successor custodian is to be named.
See GOVERNANCE.md.