Old state
Can the prior material version still be identified?
CORRECTIONS & REVISION
Building an Archive That Can Admit It Was Wrong
How fixity, provenance, version history, correction, supersession, and propagation let an archive preserve error without becoming trapped by it.
QUICK ANSWER
In one sentence
How fixity, provenance, version history, correction, supersession, and propagation let an archive preserve error without becoming trapped by it.
THE CONTROLLING CLAIM
AN ARCHIVE THAT CANNOT RECORD ITS OWN ERROR IS A MONUMENT, NOT MEMORY.
The file opens. The checksum matches. The page is exactly what it was yesterday. And the sentence inside it is still wrong. That is the problem: preservation can protect an error perfectly. Memory becomes trustworthy only when it can preserve the old bytes, identify the corrected state, and show the road between them.
TRUTH BOUNDARY
NIST, Library of Congress, PREMIS, ICMJE, and Crossref sources describe hashing, preservation metadata, versioning, and correction practices in their own technical or publishing contexts. The Ministry adapts those concepts to a public web archive; no cited source certifies this site or makes the Ministry test a universal archival standard.
01 — FIXITY
The bytes can be exact while the claim is false.
A cryptographic hash can tell you whether the bytes you have are the bytes a recorded digest describes. That is powerful. It can expose silent corruption, accidental modification, or substitution. It cannot tell you whether the author understood the evidence, whether the document was misdated, whether the interpretation was fair, or whether the claim survived later research.
So keep the hash. But stop asking it to do philosophy.
02 — PROVENANCE
Where did this artifact come from, and what happened to it?
Preservation metadata exists because an artifact is more than a filename. PREMIS models preserved objects alongside events, agents, and rights. The Ministry uses the same pressure in plain language: source, date, revision, relationship, correction, and custody should not disappear just because a new page looks cleaner.
A preserved record without provenance can be intact and still be unusable.
03 — VERSION
A new state should not erase the existence of the old one.
Publishing guidance from ICMJE and Crossref separates meaningful correction from silent replacement. The practical principle is simple: if a consequential factual statement changes, the archive should preserve enough information for another reader to identify the prior state, the corrected state, the reason, and the date.
The objective is not to preserve every comma forever. The objective is to preserve material change.
04 — PROPAGATION
The hardest error is the one that already traveled.
One source can feed an article. The article can feed a summary. The summary can feed metadata, a service guide, a downloadable file, or another article. Fixing the first page while leaving the descendants untouched creates a cleaner origin and a dirty system.
Correction therefore needs dependency review: find where the claim traveled, identify whether the dependence was essential or merely contextual, and record what was checked.
05 — ARCHIVE CORRECTION TEST
Can the record survive the discovery that it was wrong?
Current state
Can the reader find the corrected or superseding state?
Change
Can a reader tell exactly what changed?
Reason
Is the evidence or reason for change inspectable?
Dependencies
Can downstream pages using the claim be found?
Propagation
Was the correction carried into material descendants?
Disagreement
Can a remaining dispute stay visible without corrupting the factual correction?
Reconstruction
Can a future maintainer reproduce the sequence without the original author?
COMPETING INTERPRETATIONS
Keep the argument capable of surviving disagreement.
- Not every typo, spacing change, or copy edit deserves a formal version event; an archive needs a materiality threshold or it becomes noise.
- Preserving all prior public states can conflict with privacy, legal, or immediate physical-harm obligations; preservation is not an absolute command to keep harmful personal information public.
- A retraction warning may be necessary when pervasive error makes an older work unreliable, while ordinary corrections should remain proportionate.
UNKNOWNS
Do not make certainty out of the missing piece.
- The Ministry has not yet defined a universal materiality threshold for every editorial change type.
- A future hosting environment may impose retention or privacy constraints that require repository preservation without continued public exposure.
- Source withdrawal does not automatically tell us which dependent claims fail; that remains a claim-by-claim dependency evaluation.
SOURCE TRAIL
Follow the record.
- MFTAC-SRC-0021 NIST — Secure Hash Standard, FIPS 180-4
- MFTAC-SRC-0023 Library of Congress — Data Integrity Management
- MFTAC-SRC-0024 Library of Congress / PREMIS — Preservation Metadata Data Dictionary v3.0
- MFTAC-SRC-0047 ICMJE — Corrections and Version Control
- MFTAC-SRC-0048 Crossref — Version control, corrections, and retractions
- MFTAC-SRC-0049 Crossref — Crossmark
RELATED READING
CORRECTION RECEIPT
Publication status: published 2026-08-30. No post-publication correction is recorded in this release. A publication-status statement is not itself a correction. If a consequential claim changes, the Correction Ledger must preserve what changed, why it changed, which Source IDs were affected, and which dependent pages were reviewed.
THE ARCHIVE DOES NOT NEED TO PRETEND IT WAS ALWAYS RIGHT. IT NEEDS TO REMEMBER EXACTLY HOW IT CHANGED.