The minimum is Level 1 of the NDSA Levels of Digital Preservation in all five of its areas: two complete copies of every file in separate places, checksums verified on arrival, a virus check, a decision about who may change or delete files, an inventory of what you hold and where, and a record of the file formats. Next comes Level 2 where the risk is highest, usually storage and integrity: a third copy in another geographic area, and a checksum check every time files are copied or moved. Version 2.1 of the Levels, released in March 2026, keeps that order and says plainly that Level 4 need not be every institution’s aim.
This article reads the version 2.1 matrix for an organization that has just received its first delivery of scans. It covers what to check on the day files arrive, how many copies to keep and on what, a fixity schedule one part-time person can run, and the smaller steps in control, metadata and formats. Sizing in terabytes stays with the storage calculator, and the fate of the paper is covered in destroying paper originals after scanning.
What the NDSA Levels of Digital Preservation are, and what changed in 2.1
The NDSA Levels of Digital Preservation are a one-page matrix of practices published by the National Digital Stewardship Alliance. It sorts the technical work of keeping files into five functional areas (Storage, Integrity, Control, Metadata and Content) and four levels, each building on the one before.
The NDSA, founded in 2010, is a consortium of institutions committed to long-term preservation of digital information. The first Levels appeared in 2013, version 2.0 in October 2019, and version 2.1 in March 2026. All versions sit on the NDSA Levels page, with the documents in NDSA’s OSF repository.
The four level names describe the order of work:
| Level | Name in the matrix | What it means in practice |
|---|---|---|
| 1 | Know your content | You know what you have, where every copy is, and you have baseline checksums |
| 2 | Protect your content | A third copy far away, checks on every move, written rules about access |
| 3 | Monitor your content | Scheduled fixity checks, a different media type, logs of who did what |
| 4 | Sustain your content | Planned migration of storage and formats, review of logs, repair as routine |
Who the Levels are written for
The v2.1 implementation guidance says the Levels aim at practices “more substantial than the conventional digital archiving advice geared towards individuals, but less daunting and demanding than those required for certification as a trustworthy digital repository.” It names historical societies among the organizations that may find them useful.
The guidance is also clear about limits. The Levels “focus on the technological aspects of a digital preservation program” and do not reflect policies or resources. A small archive needs two things the matrix leaves out: a named person who runs the routine, and money for storage after the project that paid for the scans has ended.
What changed between version 2.0 and version 2.1
The 2026 revision changed less than its version number might suggest. The implementation guidance lists three things: a leaf icon on recommendations with environmental sustainability considerations, a new Environmental Sustainability Guide that explains them, and “minor edits to wording in Storage Levels 1 and 4.” The working definitions also moved into a standalone document.
Comparing the version 2.0 matrix (2019) with the version 2.1 matrix shows exactly which storage cells changed:
| Storage cell | Version 2.0 (2019) | Version 2.1 (March 2026) |
|---|---|---|
| Level 1, third item | “Put content into stable storage” | “Put content into dedicated preservation storage” |
| Level 4, first item | “Have at least three copies in geographic locations, each with a different disaster threat” | “Have at least three copies in separate geographic locations, each with a different disaster threat” |
| Level 4, second item | “Maximize storage diversification to avoid single points of failure” | “Diversify storage solutions to avoid single points of failure” |
| Level 4, third item | “Have a plan and execute actions to address obsolescence of storage hardware, software, and media” | “Plan for and execute actions to address storage hardware, software, and media obsolescence” |
The rest of the matrix reads as it did in 2019, apart from small spelling fixes. If a guide quotes “stable storage,” it is quoting version 2.0. The leaf icons sit only in the Storage, Integrity and Content rows, marking the places where how and how often you act has the largest environmental effect.
The bigger change is in tone. The guidance now says that “Level 4 need not be the aim of every institution,” and that sustainability goals “may require taking a sufficiency approach in which you purposefully do not pursue higher levels in all Functional Areas.” For a small archive that is permission, in writing, to stop at a level that matches the value of the files and the hours available.
The minimum: Level 1 across all five areas
Level 1 in every area is the minimum that makes a collection of files knowable and recoverable. It takes no special software, and most of it is done once, on the day files arrive, then kept current.
Here is every Level 1 item in the version 2.1 matrix, with what it means when the files are a vendor’s delivery of scans:
| Area | Level 1 text (v2.1) | What it means after a delivery |
|---|---|---|
| Storage | “Have two complete copies in separate locations” | Two full copies of every master, access file and record, in two buildings |
| Storage | “Document all storage media where content is stored” | A list of each drive, server or service, with its purchase or renewal date |
| Storage | “Put content into dedicated preservation storage” | The preservation copies live apart from the shared drive where staff work |
| Integrity | “Verify integrity information if it has been provided with the content” | Validate the vendor’s checksums before you accept the delivery |
| Integrity | “Generate integrity information if not provided with the content” | If no checksums came, make them yourself on day one |
| Integrity | “Virus check all content; isolate content for quarantine as needed” | Scan the delivery in a holding area before it reaches preservation storage |
| Control | “Determine the human and software agents that should be authorized to read, write, move, and delete content” | Decide which people and which programs may change the preservation copies |
| Metadata | “Create inventory of content, also documenting current storage locations” | A spreadsheet of packages, file counts, sizes and where each copy sits |
| Metadata | “Back up inventory and store at least one copy separately from content” | The inventory lives somewhere other than the drives it describes |
| Content | “Document file formats and other essential content characteristics including how and when these were identified” | Record that masters are TIFF, access files JPEG, and how you know |
What “separate locations” means
The v2.1 working definitions settle a common misunderstanding. Separate locations mean “separate physical locations,” and two copies “are not considered separate if the storage infrastructure for both copies is in the same physical location (e.g., the same datacenter).” Two drives in the same desk drawer, or a computer and the external drive plugged into it, do not count.
A separate geographic location, the Level 2 test, goes further. The 2026 definitions say it “could mean in another state, another time zone, or even another country,” and ideally a place with different disaster threats. The 2019 definitions began that list with “across town”; version 2.1 dropped it, so a second building in the same town now reads as Level 1, not Level 2.
What “dedicated preservation storage” means
The 2019 definitions explained “stable storage” as “Dedicated storage designated and managed specifically for digital preservation,” so the new wording spells out what the old term meant. In practice, preservation copies sit on storage used for nothing else, not in a folder on the shared drive where someone can rename or delete a master while looking for an image. A pair of external drives used only for preservation qualifies; a subfolder named “DO NOT TOUCH” does not.
Why Level 1 everywhere before Level 3 anywhere
The implementation guidance notes that the 2019 revision gave the Levels “a stronger progressive implication,” while accepting that organizations sometimes reach a higher level in one area first. A simple rule follows: finish Level 1 in all five areas before you go beyond Level 2 in any of them. A third copy in another state does little good if nobody knows which drive holds which box.
Receiving the first delivery: checksums, BagIt and a virus check
Accept a delivery only after you have verified it. That means a virus check in a holding area, a full validation of every checksum the vendor supplied, a count of files against what you ordered, and a second validation after each copy lands in preservation storage.
A checksum is a short string calculated from every bit of a file; change one bit and the checksum changes. The Digital Preservation Coalition’s handbook calls it a “digital fingerprint.” Checking that a file still matches its recorded checksum is a fixity check, and fixity, in the NDSA working definitions, is “the characteristic that indicates a digital object’s bitstream remains unchanged over time.”
Why checksums must come with the files
The 2014 NDSA report Checking Your Digital Content puts the first question fixity answers as: “Have you received the files you expected?” It calls it “ideal to encourage content providers or producers to submit fixity information along with content objects.” Checksums made when the files were created cover the trip to your storage; checksums you make after arrival cannot catch damage in transit.
So write it into the contract: the delivery arrives as BagIt bags with checksum manifests. BagIt is a simple folder layout published as RFC 8493 in October 2018. A bag holds a short bagit.txt declaration, a data folder with the files (the payload), at least one manifest listing every payload file with its checksum, and optional tag files such as bag-info.txt. The RFC says tools that create and validate bags “MUST support the SHA-256 and SHA-512 algorithms” and should use SHA-512 by default for new bags.
Complete is not the same as valid
RFC 8493 defines two states, and the difference matters on delivery day:
- A complete bag has every required element, every file listed in its manifests, and every listed tag file present.
- A valid bag is complete and “every checksum in every manifest has been successfully verified against the corresponding file.”
Many tools offer a quick check that compares only the file count and total byte count recorded in bag-info.txt as Payload-Oxum. The RFC calls this “strictly an optimization” and says tools “MUST perform the standard checksum validation process before proclaiming a bag to be valid.” In the Library of Congress’s bagit-python tool, bagit.py --validate --fast runs that quick look, while bagit.py --validate checks every checksum. Use the second one before you accept anything.
The receiving procedure, step by step
- Copy the delivery to a holding area. Use a computer or drive that is not your preservation storage.
- Virus check the copy. The matrix asks you to “isolate content for quarantine as needed.”
- Validate every bag in full. Run full validation, not the fast check, and save the report with the date.
- Count against the order. The 2014 NDSA report calls expected file count and size a “simple accessible first line of inquiry.”
- Stop on any failure. Do not repair or rename files yourself. Ask the vendor to resend the affected bag, and validate the new copy.
- Copy to each preservation location and validate there. This is the Level 2 item “Verify integrity information when moving or copying content.”
- Back up the manifests separately. This is the Level 2 item “Back up integrity information and store copy in a separate location from the content.”
- Record the event and sign off. Note the date, person, tool, result and copy locations in the inventory, then accept the delivery in writing.
The Library of Congress also published Bagger, a graphical BagIt tool. Its project page says the Library “is reviewing next steps for the future of Bagger,” so check its status before you build a routine around it.
Example (fictional). The Aster County Historical Society receives Box 3 of the Okerlund family papers and the Wren Gulch Mercantile ledger as one bag, achs-box3, with a manifest-sha256.txt file. The payload holds 410 TIFF masters, 410 JPEG access copies, page text, ALTO files, 281 catalog records and the vendor’s reports. The society’s volunteer copies the bag to a laptop used only for intake, runs a virus scan, then runs full validation: 0 failures.
She copies the bag to the society’s two preservation drives, one at the museum and one at the county library branch across town. Validation on the second drive fails for ACHS-3-L077.tif, page 77 of the day book: the laptop went to sleep during the copy. She deletes the partial copy, copies the bag again and validates: 0 failures. Every result, the failure included, goes into the inventory with the date, and the manifests go into the society’s records storage, which is separate from both drives.
What to ask for in the contract
The receiving routine works only if the delivery is built for it. Put these lines in the scope of work, alongside the vendor selection checklist:
- BagIt bags conforming to RFC 8493, with SHA-256 or SHA-512 manifests.
- A
bag-info.txtwith aBagging-Date, anExternal-Identifierthat matches your box numbers, and aPayload-Oxum. - An inventory listing every file with its size, format and checksum.
- A record of what was done to each file after capture, such as cropping or stitching.
- How long the vendor keeps its working copies after you sign off.
Storage: how many copies, where, and on what
Level 1 storage is two complete copies in separate physical places. Level 2 is three complete copies with one in a separate geographic location, and Level 3 adds a copy facing a different disaster threat and a copy on a different type of storage media.
The NDSA Levels and the 3-2-1 rule compared
Many archives start from the 3-2-1 backup rule. The US-CERT paper Data Backup Options (2012) states it this way, crediting Peter Krogh’s The DAM Book (2009): keep three copies of any important file, one primary and two backups; keep them on two different media types; store one copy offsite. It lines up with the NDSA matrix, but not exactly.
| Question | 3-2-1 rule (US-CERT, 2012) | NDSA Levels v2.1 |
|---|---|---|
| How many copies | 3 (1 primary, 2 backups) | 2 at Level 1, 3 at Level 2, 3 in separate geographic locations at Level 4 |
| How far apart | 1 copy “offsite,” outside the building | Level 1 separate physical locations; Level 2 a separate geographic location; Level 3 a different disaster threat |
| Different media | 2 media types | At least one copy on a different media type at Level 3 |
| Checking the copies | Not part of the rule | Checksums at every level from Level 1 on |
| Who may change them | Not part of the rule | Control row, from Level 1 on |
Read together, 3-2-1 roughly matches NDSA Level 2 storage with one Level 3 item added early. The bigger gap is that 3-2-1 says nothing about checking that copies are still whole.
Backup is not preservation
The DPC handbook’s storage chapter is blunt: “Backup and digital preservation are not the same thing.” Preservation storage needs “a higher level of geographic redundancy, stronger disaster recovery, longer-term planning, and most importantly active monitoring of data integrity.”
A sync service is the clearest case. The US-CERT paper warns that rolling backups “can silently propagate any corruption or malware in the primary files to the backup files.” Count only copies that would survive someone deleting or damaging the first.
Storage options for a small archive
The NDSA Environmental Sustainability Guide sorts storage by availability. Online storage is always powered; nearline storage waits in standby and wakes without human help; offline storage needs a person to plug it in or load it.
| Option | Availability | Good for | What to watch |
|---|---|---|---|
| External hard drives, kept in two buildings | Offline | A first Level 1 setup with no monthly bill | Someone must plug them in for checks; drives of the same model and batch can fail together |
| Network storage in your building | Online | The working preservation copy, with easy checks | It is one location; it needs power, updates and a replacement plan |
| Cloud storage with immediate access | Online | A copy in another region, reached from anywhere | Read the terms on regions, deletion, fixity checks and leaving the service |
| Cloud storage with delayed retrieval | Nearline | A far copy you rarely read | Retrieval can take time and may cost money, so test a restore before you need one |
| Data tape, loaded by hand | Offline | Large collections and a different media type | Drives are costly and tape systems are not all compatible with each other |
| Optical discs | Offline | Small, fixed sets of files | Low capacity, handled one disc at a time, and they do not last forever |
The US-CERT paper notes that tape systems are expensive and “not all compatible,” and that optical discs “do not last forever.” The DPC handbook adds that vendor reliability figures are “typically based on models and simulations and not from long-term observations,” and that correlated failures, such as a manufacturing fault affecting every drive in a server, “can be far more significant” than the failures those figures cover.
Questions to ask a cloud provider
The 2014 NDSA report says you may rely on a provider’s own checks, but you need to understand “how often and how detailed fixity checks are done.” Before you sign, get written answers to these:
- In which regions is the data physically stored?
- Does the service check fixity itself, how often, and can you see the results?
- What does retrieval cost, and how long does it take, for each storage class?
- How do you get all your files out if you leave or the price changes?
For the Level 3 copy, the working definitions give an example: content on the U.S. east coast exposed to hurricanes and flooding should have a copy somewhere without that threat. For a Front Range archive, ask whether the far copy faces the same wildfire, hail or flood exposure as your building.
A fixity schedule sized to staff time
Check fixity whenever files are copied or moved, then on a fixed interval you can keep up for years. The DPC handbook’s general guideline is about every six months for hard drive systems and once a year for data tapes.
How often to check
The NDSA’s 2014 report gives monthly, quarterly or yearly as examples and notes that “the more often you check, the more likely you are to detect and repair errors.” It also lists the costs: checks add wear to media and compete for computer time. The 2026 guide adds that “the more frequently you verify your data, the greater the environmental impact.”
For a small archive, the right interval is one that still happens after the grant ends and the volunteer who set it up moves away. A schedule that slips is worse than a slower one that holds.
| Copy | When to check | Notes |
|---|---|---|
| Every copy | After every copy or move | Level 2; validate the bag where it lands |
| Network storage in the building | About every six months | DPC general guideline for hard drive systems |
| Offline drives | When they come out for their scheduled check, at least yearly | The check is also the moment to confirm the drive still mounts |
| Data tape | About once a year | DPC general guideline for tapes |
| Cloud copy | Per the provider’s own checks, plus a yearly sample you download and verify | Record what the provider checks and how you know |
| Any copy | After a power cut, a move, a drive error or a reported failure | The Level 4 item “Verify integrity information in response to specific events or activities” |
Which checksum algorithm
The choice matters less than running the checks at all. The 2014 NDSA report says SHA-256 “is recommended for applications where security is important,” and that “using any fixity instrument is much better than none at all.” The DPC handbook says that if checksums are used to detect accidental damage, such as a storage failure, “then MD5 is sufficient and has the advantage of being well supported in tools and is quick to calculate.”
Two rules settle most cases. Keep the algorithm your vendor delivered, since recomputing everything costs time and energy for little gain. If you choose for yourself, the 2026 guide suggests “the least energy-intensive algorithm that meets the needs of your use case”; it cites an AVPreserve comparison in which SHA-256 took about 38 percent more CPU runtime than MD5, while noting that reading data from disk or a network often matters more.
Sizing the routine to the hours you have
A full fixity check is mostly unattended machine time. Staff time goes to starting it, reading the report and logging the result. Measure your first full check and plan from that number.
Example (fictional). By its third year of scanning, the Aster County Historical Society expects about 30,000 images. Suppose its masters average 45 MB and its access copies 1.5 MB, which comes to about 1.4 TB per copy. Suppose also that the drive kept connected at the museum reads at about 150 MB per second; the society will measure its own rate on the first run.
At that hypothetical rate, one full check of the museum copy takes about two and a half hours of unattended running, started on a Friday afternoon. The volunteer spends perhaps half an hour on the report and the log. Twice a year for the museum drive and once a year for the drive at the library branch comes to three full checks, and the cloud copy gets a yearly sample of 200 files downloaded and verified.
The 2026 guide offers one more saving: if you are testing a storage medium for degradation, you “could calculate checksums on a small sample of files for each media object instead of doing so for all files.” A sample is a reasonable test of an offline drive between full checks. It is not a substitute for checking every file after a move.
What to do when a check fails
A failed check is the system working. The 2014 NDSA report puts it plainly: “If you have multiple copies of digital objects, you can discern which copy is correct and then replace the corrupted digital files or objects.” Follow the same steps every time:
- Do not overwrite anything yet. Note the file, the copy, the date and the error.
- Check the same file on the other copies. A copy that matches its recorded checksum is your good copy.
- Replace the damaged file and verify the replacement.
- Look for a pattern. The 2014 report warns that failures “at high rates” can indicate media failure; several on one drive mean replacing the drive.
- Record what happened. Level 3 asks you to “Document integrity information verification processes and outcomes.”
Control: deciding who can change or delete files
Control is about who, and what, may read, write, move or delete preservation copies. Level 1 asks you to decide; Level 2 asks you to write the decision down and apply it.
The matrix speaks of “human and software agents,” a term from PREMIS, the preservation metadata standard, where agents are “persons, organizations, or software applications.” Software matters as much as people. A sync client or a cloud lifecycle rule can delete files as thoroughly as a person, and faster.
A control plan that fits on one page
| Who or what | Preservation copies | Access copies | Why |
|---|---|---|---|
| Collections manager (named) | Read, write, move, delete | Read, write | Runs intake, checks and repairs |
| One backup person (named) | Read; write only in an emergency | Read, write | Covers absences without a shared login |
| Volunteers and researchers | No access | Read | Access copies can be made again |
| Sync or backup software | None on the offline drives | As configured | Keeps an automated rule from spreading a deletion |
| Cloud lifecycle rules | Reviewed once a year | Reviewed once a year | An expiry rule is a software agent |
Level 3 adds logs of who did what, and Level 4 a periodic review of them. Most small archives can stop at Level 2 here for some time, using the inventory as a simple log of every intake, copy, check and repair.
Metadata: an inventory you can keep in a spreadsheet
The Level 1 metadata item is an inventory of what you hold that also records where every copy is stored, backed up somewhere other than the drives it describes. A spreadsheet is enough to start.
The working definitions describe an inventory as “a broad description or list of digital content to be preserved,” which may include “file paths and names, file formats, size, media, and location of the content.” Keep it at the package level, one row per bag, and let the manifests carry the file-level detail.
| Package | Contents | Files | Size | Copy A | Copy B | Copy C | Last full check | Result |
|---|---|---|---|---|---|---|---|---|
| achs-box3 | Box 3, Okerlund papers and day book | 1,449 | 19.2 GB | Museum drive PRES-01 | Library branch drive PRES-02 | Cloud, second region | 2026-10-04 | All copies valid |
| achs-box7 | Box 7, school board minutes | 2,016 | 31.5 GB | Museum drive PRES-01 | Library branch drive PRES-02 | Cloud, second region | 2026-10-04 | 1 file repaired on PRES-02 |
The rows above are fictional. A second tab lists the media themselves (drive serial numbers, purchase dates, the cloud account and its renewal date), which covers the Level 1 storage item “Document all storage media where content is stored.”
Levels 2 to 4: from identifiable files to a record of actions
Level 2 asks you to “Store enough metadata to know what the content is.” In practice, a file should be identifiable without the database that made it: names that carry box, folder and item, which the file naming builder produces in a fixed-width pattern, and catalog records delivered in the same bag as the images. For how much to catalog per file, see item-level or collection-level description.
Level 3 asks you to choose metadata standards, and Level 4 to “Record preservation actions associated with content and when those actions occur.” The usual standard is PREMIS, which the Library of Congress calls “the international standard for metadata to support the preservation of digital objects”; version 3.0 is current. A small archive can log PREMIS-style events (ingest, fixity check, replication, repair) as spreadsheet rows long before it adopts a system that writes PREMIS XML.
Content: file formats and the vendor’s specification
The Content row starts with a list of the file formats you hold, including how and when you identified them. At Level 2 you verify those formats and “Build relationships with content creators to encourage sustainable file choices.”
Identifying formats
A file extension is a claim, not an identification. DROID, from the UK National Archives, performs “automated batch identification of file formats” by reading the files themselves. Run it on each delivery, keep the report with the manifests, and note the tool version and date: that is the “how and when” part of the Level 1 item.
For a digitized collection, the “content creator” in the matrix is whoever ran the scanner, so sustainable file choices belong in the contract’s specification. The Library of Congress Recommended Formats Statement lists TIFF, JPEG 2000, PNG and JPEG among its preferred formats for digital photographs, and FADGI star ratings explained covers capture quality.
Levels 3 and 4 ask you to watch for obsolescence and migrate when needed. The 2026 guide calls bulk conversion at ingest potentially “a high-cost/low-reward action for file formats that have a low risk of obsolescence at present.” For TIFF masters and JPEG access copies, a yearly look at whether the formats remain widely supported is enough for now.
Environmental sustainability: enough copies, not more
Version 2.1’s main message is sufficiency: keep the copies and run the checks the value of the files justifies, and no more. Two complete copies in different places remain the floor.
The Environmental Sustainability Guide says that “Having two full copies of data in different locations is essential for redundancy and safeguarding against data loss, and should be an uncompromised baseline for all preserved materials.” Above that floor, it asks you to align the number of copies with the value of the material, “to avoid over-duplication of lower-value materials.”
Practical choices the guide supports
- Match copies to value. Unique masters of originals you might deaccession deserve the third far copy first. Access copies, which can be made again from masters, may need fewer.
- Prefer nearline or offline for copies you rarely read. The guide notes that online storage “requires constant energy,” while nearline and offline storage generally use power only when active.
- Delete copies you do not need. Working copies on laptops and duplicate exports cost storage and checks without adding safety.
- Do not wait for measurement. The guide says “It is not necessary to measure or assess the environmental impact of digital preservation prior to taking action.”
A first-year preservation plan for a small archive
A small archive can reach Level 1 everywhere, and Level 2 in storage, integrity, control and metadata, within its first year of receiving scans. The plan below assumes one responsible person with a few hours a month.
On delivery day
- Copy each delivery to a holding area and virus check it.
- Validate every bag in full and compare file counts with the order.
- Copy to two preservation locations in separate buildings, and validate each copy.
- Store the manifests, reports and inventory away from the drives.
In the first three months
- Write the one-page control plan.
- Build the inventory and media list, and back both up off the drives.
- Identify formats with DROID or a similar tool, and record tool version and date.
- Choose the third, far copy and get written answers to the cloud questions.
By the end of the first year
- Add the far copy and validate it.
- Run, time and log the first scheduled full check of every copy.
- Test a restore from the far copy, including retrieval time and cost.
- Put storage costs and renewal dates in next year’s budget.
Every year after
Repeat the scheduled checks, review the control plan and lifecycle rules, confirm your formats remain supported, and replace drives at the age your plan sets. The preservation readiness checklist turns this into an annual review against the NDSA areas, and the storage calculator estimates the terabytes for each copy as the collection grows. For the wider sequence that leads to a delivery like this, from selection and capture to cataloging, see what digitizing a historical collection actually involves.
Dastok is a Colorado Springs firm that digitizes historical documents, photographs and museum collections. If you would rather receive files already packaged for this routine, our digital preservation service delivers validated BagIt bags, an inventory and a written fixity procedure, and a pilot box lets you test the handover on real material before you commit a collection.