Check research file integrity without confusing it with truth
A file-integrity check asks whether the stored bytes remain the same. It does not prove who authored the file or whether its statements are true. Combine origin records, untouched originals, checksums when useful, and an actual restore test. Keep each check’s scope explicit.
The Library of Congress format resources provide background on preserving usable digital material. W3C provenance provides background on recording derivation. The integrity card below is a practical local-archive method, not a claim of forensic certification.
Preserve the original acquisition object
Save the source URL, publication identity, acquisition time, filename, and format. Keep the downloaded original separate from any annotated or converted copy. A file that has been highlighted and resaved may differ from the acquisition object even when the visible text seems unchanged.
If you calculate a checksum, record the algorithm and the file it applies to. Recalculate using the same method when checking a copied or restored file. Matching values support byte consistency under that method; they do not establish that the initial download came from the intended official source. That requires a separate provenance check.
Hypothetical integrity card
Assume fictional Elm Harbor’s annual report is saved in a local research archive. The example describes a proposed process, not a test already performed.
| Field | What to record |
|---|---|
| Source identity | Issuer, title, period, version, official route |
| Acquisition | Date and original local filename |
| Original checksum | Algorithm and calculated value from actual file |
| Derived copy | Annotation or extraction linked to original |
| Restore check | Location, date, readability, checksum comparison |
Do not invent a checksum to make the example look complete. In a real record, calculate it from the actual file. If no checksum was taken at acquisition, say so. A later checksum can establish a new baseline but cannot retroactively prove earlier integrity.
Treat a mismatch as a question
When hashes differ, identify which objects are being compared. An original PDF and an annotated PDF should be different. A copied original is expected to match. A publisher may also replace a public file or change metadata. Determine the relationship before calling the mismatch corruption or manipulation.
Inspect material differences relevant to the research question. If text or a table changed, check for a correction or replacement record. If only a working copy changed, preserve its derivation. Keep the prior original until the relationship is understood; overwriting it removes the evidence needed to investigate.
Verify recovery, not just backup existence
Choose a source, restore it to a fresh location, open it, and compare the relevant identity and passage. If a checksum baseline exists, compare it too. A backup directory visible in a file manager is not proof that the source can be read or that its links still work.
Test the manifest and relative paths as well as the binary file. Research evidence depends on knowing which note uses which version. A perfectly preserved PDF with a missing identity row can still be difficult to use. Integrity and findability support different parts of the archive.
Limits and common mistakes
This method does not guarantee long-term readability, prove authorship, or validate factual accuracy. File formats, software, and access routes can change. Maintain readable formats and useful identity metadata rather than relying on a checksum alone.
Common errors include hashing only an annotated copy, treating matching bytes as independent source verification, and overwriting originals after a mismatch. Another is promising a restore capability that has never been tested. Record the checks actually performed and leave untested recovery paths clearly labeled.
A final record check
Record which object a baseline protects: original download, annotated copy, extraction, or export archive. These objects can legitimately have different bytes. A later audit should compare like with like. If the original acquisition object was not preserved, label the surviving copy accurately rather than upgrading it to an untouched original because it is now the oldest file available.
FAQ
Is a checksum necessary for every small research note?
Not always. Use it where preserving an exact source object matters. At minimum, retain origin, version identity, and a readable original.
Can two different hashes still refer to the same disclosure?
Yes. Format or metadata changes can produce different files. Inspect document identity and relevant content before deciding whether the disclosure itself changed.