Documentation

What Is a Preservation Workflow for Digital Photography?

Six stages, from memory card to periodic verification. What makes it a chain rather than a set of good intentions.

Preservation

A preservation workflow is the sequence of stages that takes an image from the memory card to a state where it will remain readable and understandable over time. What makes it a chain, rather than a list of tips, is that a missing stage cancels the others.

Perfect checksums on a collection with no descriptive metadata: you’ll know you are faithfully preserving images nobody can identify. Impeccable metadata with no off-site copy: the description disappears with the building.

The six stages

  1. Ingest. Copy from the card, verifying the copy is faithful before erasing the source. It’s the only moment the original still exists in duplicate.
  2. Identification. Assign a stable identifier, independent of filename and location.
  3. Characterisation. Read and preserve technical metadata, compute the checksum, detect the actual format.
  4. Description. Add context: project, place, people, subject, rights. This is the stage everyone defers, and the one nobody comes back to.
  5. Replication. Build copies following the 3-2-1 rule, one off-site, with a manifest that lets you check each batch.
  6. Verification. Periodic re-checking of checksums, with a log of discrepancies. Without this stage, the first five are only a snapshot of the initial state.

What breaks a chain in practice

  • Erasing the source before verifying. The most common gesture, and the most irreversible.
  • Deferred description. Context evaporates within weeks. What isn’t recorded at ingest won’t be.
  • An off-site copy never tested. A backup you’ve never attempted to restore is a hypothesis, not a copy.
  • Verification left to human initiative. Anything depending on remembering doesn’t happen. A scheduled campaign does.

The principle that holds it together

At every stage, one rule: fail loudly rather than succeed silently. A checksum that couldn’t be computed is not a valid file. An interrupted copy is not a copy. A batch whose count doesn’t add up is not complete.

A system that reads doubt as success will teach you about the problem years too late.

In practice

  1. Verify the copy before releasing the card.
  2. Describe at ingest, even coarsely.
  3. Produce a manifest for every archived batch.
  4. Schedule verification campaigns instead of deciding on them.
  5. Test a real restore at least once a year.

Obscura Flow implements this chain as one piece: verification before source deletion, durable identifiers, checksums at import, resumable uploads to remote storage, and fixity audits that only update their timestamp after a successful check.