Preservation
Draft, November 2026.
Purpose and scope
The xd Crossword Corpus exists to keep published crosswords readable, findable
and studiable for as long as anyone wants them, which we take to mean a
century or more. This document says what we do to keep the corpus safe, what
we promise, and where the limits are.
The corpus is run by two volunteers, funded from their own pockets, with no
institution behind it. What we can promise is scaled to that. We do not claim
to be a certified repository. We do claim that the choices below make the
corpus unusually easy to keep, and that we are honest about where it is at
risk.
What preservation means here
Two things, in order of importance:
- The bits survive. Every puzzle, and the original file it was made
from, exists in more than one copy in more than one place, and we can tell
when a copy has been damaged.
- The content stays readable. A puzzle is stored as plain UTF-8 text in
the
.xd format, whose specification is public. A .xd file can be read
with no software at all, and the whole corpus can be processed with a few
lines of code in any language. There is no database, no binary format and
no proprietary dependency between a reader and a puzzle.
How the corpus is kept
- Version control. The corpus is a git repository. Every change to every
file is recorded with who made it, when and why. History is never rewritten.
- Copies. The repository is hosted on a commercial git service, mirrored
to a second service, and cloned on the maintainers’ machines and by other
contributors. Any complete clone is the whole corpus.
- Originals. The file each puzzle was converted from is kept, unchanged,
in a checksummed archive in cloud storage, and a receipts table links every
puzzle to its original and its source. Fixes are made to the
.xd file,
never to the original.
- Checks. Every contribution is checked by continuous integration before
it is merged: format validity, identifiers, shelving, dates and receipts.
A scheduled job checks that every registered source archive is still
present and unchanged.
- Format. The
.xd format is specified publicly and versioned
conservatively. When the format changes, existing files remain valid or are
migrated in a recorded, reviewable change.
What we commit to
- Every puzzle accepted into the corpus is kept indefinitely, with its
provenance, unless removed under the rule below.
- The original source files are kept as received.
- Changes to puzzles are recorded in history and, where they alter what was
published, explained in the puzzle’s notes (see EDITORIAL.md).
- The corpus can be handed over whole at any time: a clone of the repository
and a copy of the source archives are everything there is.
- We assess our own preservation capability once a year, publish the result in
assessments/, and act on what it finds.
Limits and known risks
- Two people. Continuity depends on the maintainers. See
GOVERNANCE.md, “Continuity”, for what happens if one or
both stop.
- Source archives have one copy. As of November 2026 the original source
files live in a single cloud bucket. Some early source archives, from before
2025, were lost when an earlier bucket was retired, although the puzzles
converted from them are intact. A second copy is the first item on our
preservation plan.
- No tested restore. We have never rebuilt the corpus from a mirror as an
exercise. We intend to, yearly.
- Hosting is personal. The services the corpus lives on are paid for and
administered by the maintainers as individuals.
Removing puzzles
Puzzles are removed from the corpus only deliberately: at the request of a
rights holder, because they turn out to be outside the collection scope, or
because they are duplicates. A removal is a recorded change with its reason in
the commit, and the puzzle’s identifier is never reused. Removed puzzles stay in
history; a removal that must also purge history is treated as an exceptional
decision and logged in DECISIONS.md.
Review
Reviewed once a year with the preservation assessment. Last reviewed:
November 2026 (first version).