Test whether an old research archive can recover its evidence
An archive works when you can recover the source used by an old note and inspect the same evidence again. File existence alone is not enough. Test identity, readability, locator accuracy, and the relationship between the historical note and source version from a fresh location.
The Library of Congress format resources provide background on digital-format sustainability. W3C provenance provides background on source relationships. The retrieval test below is an original manual checklist, not a guarantee of permanent preservation or a claim that an actual archive has already passed.
Start from a question, not a folder
Choose a historical note with a specific claim. Identify the source ID it cites and the information cutoff it used. Try to recover the source using the manifest rather than memory. A test that starts from a known PDF on your desktop may miss broken relationships elsewhere in the archive.
Confirm the source version and period. A current download with the same title may be corrected or replaced. If the original note depends on an earlier version, the test must distinguish that object from later material. Record whether you found the exact version, a related copy, or only a later source.
Hypothetical retrieval audit
Assume fictional Oak Channel’s historical project note cites an earlier announcement. This is an invented test plan, not a report of completed testing.
| Test | Pass evidence | Failure to record |
|---|---|---|
| Identity | Correct issuer, period, source version | Similar title only |
| File access | Included file opens in available reader | File missing or unreadable |
| Locator | Intended passage found with context | Page points elsewhere |
| Relationship | Note’s claim matches source role | Later source substituted silently |
| Portability | Same route works from fresh folder | Personal absolute path required |
An archive may pass readability while failing identity. For example, a PDF opens perfectly but belongs to the following year. Report the separate outcomes rather than marking the whole test successful because a reader displayed something.
Test a small, varied sample
Choose sources with different risks: an old PDF, an HTML filing, a corrected announcement, a scanned page, and a source used in an exported note. The sample does not prove every item works. It helps reveal common failure modes such as broken relative paths, missing manifests, and locators tied to one viewer.
Record the sample and the date of the check. If checksums exist, compare them as an additional integrity test. Keep readability and provenance separate from byte consistency. A matching file can still have a broken citation, and a new format can still represent a related source requiring explicit mapping.
Repair without rewriting the past
For a broken link, locate an official route and record whether it is the same source object. For a missing version, search the known publication or filing chain. If only a later source is available, mark that limitation and do not pretend the historical evidence was recovered.
Update access paths and locator fields with a repair record. Preserve original references and cutoff relationships. A repair should make the archive more usable while retaining what the old note actually relied on. If the underlying claim cannot be reproduced, flag that claim for review rather than merely fixing the folder appearance.
Limits and common mistakes
Retrieval testing does not prove research correctness or completeness. It checks whether selected evidence remains usable. Public hosts, formats, and permissions may change, so record current results and avoid promises of permanent access.
Common mistakes include checking only recently used files, treating backup existence as recovery, and replacing missing historical material with the newest download. Another is leaving failed tests without an owner or next action. Each failure should lead to a repair task, a defined unresolved limitation, or an explicit retirement decision.
A final record check
Treat a failed retrieval as a scoped repair ticket. Identify the affected note, missing source object, and whether the problem is access, identity, readability, or location within the file. Assign the next action accordingly. Repairing one sampled item does not prove the rest of the archive passed, so retain the sample boundary in the final retrieval report.
FAQ
How often should I test an archive?
Choose a cadence appropriate to its use and changes. Test after migrations or exports and before a handoff that depends on historical evidence, rather than claiming an untested universal schedule.
Can an archive be useful with some failed retrievals?
Yes, if the failures and affected claims are visible. Keep supported material usable and isolate gaps instead of implying that all historical evidence is recoverable.