An archive can preserve its files longer than an organization preserves the knowledge needed to interpret them. If required meanings remain in the memories of former operators, application retirement can leave a delayed dependency: records appear available until a later question requires context that was never retained.
This is a conditional effect of separating artifact preservation from knowledge preservation. It is not an estimate of archive failure frequency or a prediction that every historical dataset will become unreadable.
Intact bytes answer a narrow question
BagIt validation establishes package completeness and manifest checksums while treating payload contents as opaque bytes. A valid package can therefore remain unchanged without explaining a field, status code, or relationship.
Suppose a historical record carries a status of C. Its meaning might depend on the application, period, or process stage. A future reader can parse the character correctly while being unable to determine whether the case was completed, cancelled, or classified another way.
That constructed example shows why file integrity and interpretability need distinct acceptance evidence. Neither property substitutes for the other.
Meaning can depend on several artifacts
Schemas, code lists, relationship mappings, format descriptions, provenance, and retained business-rule explanations can be necessary to answer a historical question. Attachments and their association with records can be as important as the primary table.
The archive should identify which version of those descriptions applies to which population. Replacing an old code list with its current version can create an apparently clear but historically incorrect interpretation.
US National Archives finding-aid requirements treat descriptive documentation as necessary for reference and retrieval in their accessioning context. The broader design implication here is to test discoverability and meaning alongside possession of the records.
The retirement team can conceal the gap
During extraction, experienced staff can answer questions informally. Their presence can make the archive seem self-explanatory even when the retrieval procedure relies on their unrecorded knowledge.
A rehearsal led by a second authorized person can expose that dependence. The person should locate a record, reconstruct its required relationships, and explain the answer using retained material rather than asking the original operator to fill gaps.
GAO’s competency-gap approach supports assessing required tasks against available skills. For an archive, those tasks include interpretation and controlled retrieval as well as technical restoration.
Provenance does not supply missing artifacts
Metadata describing a build or extraction can establish lineage while leaving required inputs unavailable. SLSA v1.1 provenance, for example, treats resolved dependency information as best effort even at its cited higher build level.
If the archive depends on running a retained reader or report, its source, runtime, configuration, and dependencies need actual preservation and execution evidence. A provenance document cannot recreate an absent installer or explain an undocumented calculation.
An alternative is to preserve sufficient independent descriptions and exports to answer the required questions without that executable. The chosen arrangement needs demonstration rather than an assumption about format longevity.
Test interpretation across a handover
Select historical questions with different dates, statuses, attachments, and exceptional outcomes. Record whether a new custodian can answer them, what supporting material was used, and which interpretations remain uncertain.
The hypothesis weakens when those questions remain answerable through maintained descriptions and repeatable access after staff change. It strengthens when success repeatedly depends on the same person’s recollection.
Retention periods and access obligations need their actual context. Whatever period is required, the archive needs continuing ownership of integrity, recovery, and interpretation. Application shutdown removes execution duties while leaving these information duties in place.