Preservation · Mar 19, 2026 · 5 min read

PREMIS preservation metadata, explained

PREMIS in plain terms: objects, events, agents and rights, and what a small archive actually needs to record to prove that a digital file is still intact.

A conservator in cotton gloves checking an archival ledger against a tablet

Preservation metadata answers a simple question: how do you know this file is the same one you received? Descriptive metadata tells people what an item is about. Preservation metadata tells a future custodian what happened to the file, who did it, and whether it can still be trusted.

PREMIS is the standard most institutions use for that job. Its data dictionary is long and its XML schema can look intimidating. The core idea is small, and a small archive can apply it with a spreadsheet and a few scripts.

What PREMIS is

PREMIS stands for Preservation Metadata: Implementation Strategies. It grew out of a working group sponsored by OCLC and RLG, which published the first data dictionary in 2005. The Library of Congress now hosts the standard, and an editorial committee maintains it. The current major version of the data dictionary is 3.0.

Two things about PREMIS are easy to miss. First, the data dictionary defines what information to keep, not how to store it. You can express PREMIS in its XML schema, in RDF, inside a METS document or in a database table. Second, it is not a descriptive standard. It sits alongside Dublin Core, MODS or EAD rather than replacing them.

Four entities, in plain terms

The PREMIS data model is built around a few entities. Version 3.0 also formalized the intellectual entity, which represents the work itself, such as a book or a recorded interview. Day to day, most of the effort goes into four:

  • Objects are the digital things you preserve. PREMIS distinguishes files, bitstreams inside files, and representations, which are the set of files needed to render one version of a work.
  • Events are actions that touch an object: ingest, fixity check, format identification, virus scan, migration, deletion.
  • Agents are the people, organizations and software that carry out events. A checksum tool at a specific version is an agent.
  • Rights record the permissions that allow you to preserve the object, such as copyright status, a license or a donor agreement.

The model works through links. An event points to the object it affected and the agent that performed it. A rights statement points to the objects it covers. Read together, these links form an audit trail. That trail is the actual product of preservation metadata.

quality metrics · sample collection
findability
82%
completeness
67%
consistency
74%
rights clarity
58%
link health
91%
Figure 1 · interactive illustration

What a small archive actually needs

The full data dictionary has many semantic units. Most are optional or only apply in specific cases. To prove that a file is intact, a small archive needs far fewer. The table below is the editors' minimum.

EntityRecord thisWhy it matters
ObjectIdentifier, file name, sizeLets you find the file and spot truncation
ObjectChecksum and algorithmThe core evidence of integrity
ObjectFormat name and version, with a registry IDTells you when migration may be needed
EventType, date and time, outcomeShows what was done and whether it worked
EventLinked object and linked agentConnects actions to files and to who did them
AgentName, type, software versionExplains results years later when tools change
RightsBasis and permitted actionsConfirms you may copy, store and migrate the file

For format identification, use a registry identifier such as a PRONOM PUID, which tools like DROID and Siegfried report. For checksums, SHA-256 is a sensible default today. MD5 is still common in older workflows and remains useful for detecting accidental corruption, but it should not be your only record for new material.

The events that matter most

If you record nothing else, record these events consistently:

  1. Ingest. When the file entered your custody, from whom, and with what checksum.
  2. Fixity check. Every scheduled comparison of the stored checksum against a fresh one, with the outcome.
  3. Format identification. The tool, its version and the result.
  4. Replication. When copies were made and where they live.
  5. Migration or normalization. The source file, the new file, and the tool used.

A long series of successful fixity checks is unremarkable to read. That is the point. When a check fails, the history tells you when the file was last known to be good, which copy to restore from, and whether the failure is isolated. The BagIt and fixity guide covers how to run those checks in practice.

Wooden library card catalog drawers
Photo illustration

PREMIS in shared preservation networks

Distributed preservation raises the stakes. When several institutions hold copies of the same content, each needs to show what it holds and what has happened to it. The MetaArchive project, which the MetaScholar Initiative helped run from Emory's Woodruff Library between 2004 and 2007 with Library of Congress support, built a cooperative network for at-risk digital content about the American South. It used the LOCKSS software, which repeatedly compares copies across nodes and repairs damaged ones.

A network like that already performs constant integrity checking. What it does not do by itself is explain the provenance of each file, its format history or its rights. Clear preservation metadata, kept by each contributing institution, fills that gap. The briefing on the MetaArchive final report and the overview of LOCKSS distributed preservation describe how that cooperative model worked.

Getting started without a repository system

You do not need a full preservation platform to begin. Many small archives start with three pieces:

  • A stable folder structure or BagIt bags with manifests for each accession.
  • A log of events, one row per action, with object ID, event type, timestamp, outcome and agent.
  • A short agent list that names each tool and version you use.

Use PREMIS names for your columns and controlled values for event types. The Library of Congress publishes vocabularies for event types, agent roles and other values through its linked data service. If you later move to a repository system that exports PREMIS XML, your log will map across with little effort.

A few mistakes come up again and again. Archives record a checksum at ingest and never check it again, so the history has a start and nothing after it. They log events without the tool version, which makes old results hard to interpret once software changes. They keep preservation metadata only inside one system, so a platform migration loses the trail. And they skip rights entirely, then find years later that nobody knows whether a file may be migrated or shared.

Each of these is easy to avoid once you know to look for it. Schedule fixity checks and log every run, including the successful ones. Name the agent and its version on every event. Export the log regularly to a plain format that lives next to the files, so the evidence travels with the content. Then review the log once a year. A quick read often reveals gaps that are still cheap to fix.

The aim is not perfect compliance. It is a record that a stranger can read in twenty years and trust. Start with fixity, format and events. Add rights as soon as you can. The full data dictionary and schema are available at loc.gov/standards/premis.

Questions

Is PREMIS an XML format?
PREMIS is primarily a data dictionary that defines what to record. It has an official XML schema and an RDF ontology, but you can store the same information in a database or a spreadsheet.
What is the minimum PREMIS a small archive should keep?
An identifier, checksum, format and size for each file, a log of events such as ingest and fixity checks with outcomes, the agents that performed them, and the rights basis for preservation.
Does PREMIS replace Dublin Core or MODS?
No. PREMIS covers preservation information about files and actions. Descriptive standards such as Dublin Core, MODS or EAD describe the content and stay alongside it.