Five checksum checks for a preservation copy
Published 2026-11-15 · Compsyscon preservation editorial desk · Preservation-method guide
A checksum comparison can help detect a byte change. It cannot, by itself, tell you that an old installer works, that a file came from the claimed publisher, or that you have permission to redistribute it. Keep the comparison result separate from those other questions.
Use the workstation storage inventory before deciding what to preserve. If a storage device is damaged, contains the only copy, or requires unfamiliar recovery equipment, stop at the inventory and arrange appropriate help. This article does not give disk-writing commands or test old hardware.
1. Preserve the reference before comparing
The Digital Preservation Coalition's fixity guidance describes checksums as part of checking transfers and monitoring stored digital material. A useful record includes the file identifier, its byte count, the algorithm, the reference digest, when and how that reference was obtained, and the comparison result.
Do not replace an earlier reference just because a later check disagrees with it. Preserve both observations and investigate. If you calculate a fresh reference from the changed file and then compare that file with it, the match says nothing about whether it still matches the earlier version.
2. Inspect three controlled cases
The local companion fixity-example/check_copy.py creates its own short text file in a temporary scratch folder, copies it, and hashes both files with SHA-256. It then changes a single character in the scratch copy without changing its length. No historical software, personal files, disks or network data are used.
| Case | Comparison result | Supported conclusion |
|---|---|---|
| Copy before alteration versus recorded source digest | Match | This controlled copy agrees with the recorded reference |
| Same-length copy after one character changes versus original digest | Mismatch | The file differs from that reference, despite equal byte counts |
| Changed file versus a newly calculated reference from that same file | Match | The new comparison is internally consistent; it does not restore the earlier reference |
The run also verifies directly that the first copy's bytes equal the source and that the scratch source remains unchanged. These are small constructed cases, not a test of a drive, copier, backup product or long-term storage system. Equal SHA-256 values are strong comparison evidence under trustworthy computation; they are not a mathematical guarantee against every possible collision or a substitute for provenance.
3. Keep a comparison manifest
For an authorized preservation copy, keep a record with:
- The exact relative file identifier and byte count.
- The algorithm and full reference digest, not a shortened screenshot.
- Source or acquisition record and date, with unresolved provenance explicit.
- The comparison date, tool/version and actual result or read error.
- An authorized next action if a comparison fails.
The fixture's printed report demonstrates these comparison fields, but it is not a complete preservation manifest. Filesystem permissions, timestamps and other metadata are outside its file-content hash. A changing source, unreadable file or missing reference needs its own recorded outcome; do not relabel it “match.”
4. Separate detection, repair and trust
A mismatch does not identify which storage component failed, locate the changed byte or repair it. Compare with an independently maintained trusted reference or preserved copy using an authorized recovery process. Do not overwrite the only remaining evidence simply to make the check pass.
A match to an untrusted download's accompanying checksum is also not proof of publisher identity. The source and reference need a trustworthy acquisition trail. Functionality, malware analysis and licensing are separate checks; none was performed by this demonstration.
The companion uses Python's standard-library hashlib interface. Run python3 check_copy.py only after reviewing the short script: it creates and removes its own scratch files and prints a JSON report. The readable Python script and executed result are available with this guide. No third-party packages are required.
Next, read the used-workstation reuse checklist. A verified file copy does not establish that an old machine is suitable for sensitive online work or that recovered software can run on it.
5. Run the scratch-file demonstration
Download the script, save it as check_copy.py, review it, then run python3 check_copy.py. It creates its own temporary teaching files and removes the scratch directory afterwards. It does not accept a disk, directory or personal file as input. Compare its three cases with the table above; no old hardware or preservation repository was tested.


