Skip to content

Systems and records

What reading instructions and software dependencies should old archives keep?

By SourceX Editorial · Updated

9 sections

Short answer

An old archive should travel with its reading instructions: a manifest that names every file format with a registry identifier, the software and version that last opened it, the code tables, schemas, fonts and keys it depends on, and a reproducibility test that a person outside the original team can run. Record a valid example and a failing one, so the next reader knows what correct looks like.

System export

Archive software dependency

This page covers what reading instructions and software dependencies should old archives keep.Illustrative

Key takeaways

  • An archive is usable only with its reading instructions: schemas, code tables, format identifiers and the software that last opened each format.
  • Write one manifest row per format, not per file, and include known losses from the export.
  • Identify formats with a signature tool against the PRONOM registry; never trust file extensions.
  • A valid checksum package proves the bytes are intact, not that anyone can interpret them; run an outsider reproducibility test as well.
  • Capture the knowledge that disappears first, such as code meanings, settings and key custodians, before the bulk export finishes.

What reading instructions and software dependencies should accompany an old archive?

Every archive needs two things besides the files: a description of how to interpret the bits, and a list of everything outside the files that a reader needs to do that. The preservation standards call the first representation information. The OAIS reference model illustrates it with a string of bits plus a separate description saying those bits are temperature readings in degrees Celsius; without the description the bits are just bits.

For a business archive the same idea is concrete. A CSV of work orders is unreadable in practice if nobody knows that status 7 means canceled by customer, that dates are day-month-year, or that the notes column was truncated at 255 characters by the old export. A folder of attachments is unreadable if half of them are in a vendor's proprietary drawing format and the license for the only viewer lapses next month.

So the deliverable is a manifest, not a memo. It names each format with an identifier, records the software that last opened it, lists the lookup tables, schemas, fonts and keys it depends on, and states a test that proves someone outside the team can open and interpret a sample. The rest of this article gives that manifest field by field, with a worked record and a failing one.

Which files stop being readable once the old software goes away?

The files most at risk are the ones whose meaning lives partly outside the file. The Library of Congress evaluates formats on several sustainability factors, and two matter most here: external dependencies, meaning whether the format needs particular software or hardware to be usable, and self-documentation, meaning whether the file carries its own descriptive and technical metadata. A file that scores badly on both is the first to become unreadable.

In a typical small or mid-sized company archive, these are the usual suspects. Each one deserves its own manifest entry rather than a line in a general note.

  • Native database files from a desktop or on-premises application, where tables are only meaningful through the application's screens and reports.
  • Exports with coded values (status 1 to 9, reason codes, region numbers) whose lookup tables live only in the application's settings.
  • Attachments in vendor formats: drawings, scanned forms in a document management system's container, call recordings in a phone system's codec.
  • Compressed or encrypted blobs stored inside database columns, which need a library or key to unpack.
  • Report definitions and templates that turned raw tables into the documents people actually used.
  • Files whose text depends on a code page or font, such as older exports written in a regional encoding rather than UTF-8.

The format dependency and reproducibility manifest, field by field

Use one manifest row per distinct format in the archive, not per file. A help desk archive might have six rows: ticket CSV, user CSV, a lookup table export, PDF attachments, PNG screenshots and a proprietary call recording format. Each row answers the same ten questions.

The format dependency and reproducibility manifest, field by field
FieldWhat it recordsExample value (illustrative)Check before you sign off
format_idRegistry identifier for the format and versionPRONOM fmt/52 (Rich Text Format 1.7)Identifier comes from a tool run, not from the extension
file_count_and_bytesHow many files of this format and total size1,204 files, 3.1 GBMatches the payload manifest totals
last_known_readerSoftware, version and operating system that last opened the files, rated minimum, recommended or known to workVendor viewer 8.2 on Windows 10, known to workSomeone opened a sample with it and recorded the date and rating
structure_infoSchema, field list, delimiter, encoding, date and number formatsUTF-8, comma, ISO dates, Table Schema fileA machine-readable schema file sits next to the data
semantic_infoCode tables and definitions that turn values into meaningstatus_codes.csv, 9 rowsEvery coded value in the data appears in the table
other_dependenciesFonts, stylesheets, codecs, libraries, keys, passwords held elsewhereCodec named in readme; key in company vault entryEach dependency is either packaged or its custodian is named
migration_copyAn open-format copy made for long-term reading, if anyCSV and PDF copies alongside originalsCopy compared to original on a sample, with differences noted
known_lossesWhat the export or migration dropped or changedNotes truncated at 255 charactersLosses written down, not discovered later
reproducibility_testA short task an outsider must be able to completeOpen ticket 4410, show status in words and attachmentRun by someone outside the original team
custodian_and_dateWho signed off the row and whenIT lead, 2026-10-18Named role, date, and next review date

How do you identify formats instead of trusting file extensions?

Run a signature-based identification tool over the whole archive and store its output with the manifest. Extensions lie: a .dat file can be a text export, a database page or a vendor container, and an attachment saved without an extension tells you nothing at all.

PRONOM, the technical registry run by The National Archives in the UK, assigns each format a persistent identifier (a PUID such as fmt/52 for Rich Text Format 1.7) and describes the format and the software associated with it. Siegfried is an open-source command-line tool that identifies files using PRONOM signatures, with options to write CSV or JSON results and to look inside zip and other container files. Its output gives you the format_id column for free and flags the files it cannot identify, which are exactly the ones that need a human to look at them.

Treat unidentified files as findings, not noise. Sort them by count and size, open a sample of each group, and either give them a manifest row with a local description or record them as unreadable with the reason. A manifest that silently omits the unknowns gives a false sense of completeness.

What does a reproducibility check prove, and how do you run it?

A reproducibility check proves that a person who never used the old system can get from the preserved files to a correct reading of a real record. File counts and checksums prove the bytes arrived intact; they do not prove anyone can understand them. You need both.

For the bytes, package the archive in a way that can be checked mechanically. The BagIt packaging specification defines a complete bag as one where every file listed in every manifest is present, and a valid bag as a complete bag where every checksum in every manifest has been verified against its file. For tabular exports, a Table Schema file (from the Frictionless Data specifications) declares each field's name, type and format, including date patterns and which strings count as missing values, so a validator can test the data against it rather than relying on someone's memory.

For the meaning, run the steps below on a random sample and record the results in the manifest's reproducibility_test field.

  • Pick five records at random from each main table and one attachment of each format.
  • Give the manifest and the package to someone outside the original team, with no access to the old system.
  • Ask them to state, for each record, its date, status in words, owner and linked attachments, using only the preserved files.
  • Compare their answers with the live system, or with screenshots taken from it before shutdown.
  • Validate the bag and run the schema validator on every table; both must pass with zero errors.
  • Record any wrong answer as a missing dependency, fix the manifest, and repeat until a fresh reader gets every sample right.

Worked record and a failing record (illustrative)

Here is one manifest row for a fictional field service company retiring a desktop job management system. The values are illustrative and describe no real company's records.

The passing row: format_id is a CSV export identified by the tool as comma-separated text; file_count_and_bytes is 14 tables, 2.4 GB; last_known_reader is the vendor application version 11 plus a spreadsheet for spot checks; structure_info points to a Table Schema file declaring job_date as a date in day-month-year order and treating blank and NA as missing; semantic_info points to job_status.csv and priority.csv exported from the settings screen; other_dependencies records that technician signatures were stored as images in a separate folder keyed by job number; known_losses records that the notes field was cut at 255 characters by the export; and the reproducibility test was passed by a bookkeeper who had never used the system, on 30 sampled jobs.

The failing row, for the same archive a week earlier, looked complete but was not. It had the same 14 tables and the bag validated, yet the bookkeeper misread 9 of 30 sampled jobs. Status 4 meant on hold in the old system but was read as complete, because job_status.csv had not been exported; and dates such as 03/04 were read in month-day order because the schema file had not declared the pattern. Nothing was missing from the bytes. Everything that went wrong was a missing reading instruction.

That contrast is the point of the exercise: a valid bag is necessary, and on its own it says nothing about whether the records can be read correctly.

Which reading knowledge disappears first if preservation waits?

The first things to go are rarely the files. They are the people, licenses and settings that explain the files, and they tend to vanish on dates set by contracts or departures rather than by anyone's archive plan.

Rank these by their deadline and capture the earliest ones first, before the bulk export is even finished.

Which reading knowledge disappears first if preservation waits?
DependencyWhy it disappearsWhat to capture nowTypical owner
Staff who know the codesDepartures in a wind-down or after a saleCode meanings, workarounds, fields used for something other than their labelOperations lead
Application license or vendor portalSubscription end or vendor support endingScreenshots of key screens, settings exports, report definitionsIT or finance
Lookup and settings tablesOften not part of the standard exportEvery code table, picklist and custom field definitionSystem administrator
Encryption keys and passwordsHeld by one person or one machineKey location and custodian in the manifest, never the key itselfIT security
Viewer or codec for attachmentsInstaller no longer downloadable, or license lapsesInstaller, version number and an open-format copy of a sampleIT
Operating environmentOld machines are wiped or recycledOperating system and version the reader ran on, as testedIT

Should you migrate each format, keep the old software, or plan for emulation?

Decide per format, and prefer migration to an open format whenever it can be done without losing meaning. For datasets, the Library of Congress Recommended Formats Statement lists line-oriented text such as CSV and platform-independent database formats such as SQLite among its preferred formats, and it generally favors character-based, platform-independent formats over native or binary ones as long as the data stays complete and precise. The SQLite developers state that they intend to support the format through the year 2050, which makes a single SQLite file a practical container when an archive has many related tables.

Keep the original files next to any migrated copy. Migration can lose things the original carried, such as formulas, layout, embedded objects or precision, and the manifest's known_losses field only helps if the original is still there to check against.

Keeping the old software, or capturing its environment for later emulation, is the fallback for formats that cannot be migrated without losing what matters. The PREMIS preservation metadata dictionary models this as environment information, including software dependencies, and suggests describing an environment as minimum, recommended or known to work, which is why the manifest asks for one of those three words and a test date.

Should you migrate each format, keep the old software, or plan for emulation?
SituationPreferred approachWhat the manifest must addWatch for
Tables in a vendor databaseExport to CSV or SQLite with a schema fileCode tables and known lossesTruncated text, lost time zones, dropped custom fields
Office documents and PDFsKeep originals; add PDF copies of key documentsReader version and any missing fontsEmbedded objects and tracked changes lost in copies
Vendor drawing or design filesKeep originals plus exported images or PDFsViewer installer, version and license statusLayers and measurements lost in image copies
Call recordings in a proprietary codecConvert a copy to an open audio formatCodec name and conversion tool versionMetadata such as call ID dropped by conversion
Encrypted or compressed blobs in columnsUnpack to files before shutdownLibrary or key custodian, and the unpack scriptBlobs that fail to unpack, counted and listed

If the preserved records later support an AI data license

A buyer evaluating records for AI training asks the same questions a future archivist would: what the fields mean, what the export dropped and whether the history is consistent. A manifest written for preservation answers most of them in advance, and it is far cheaper to write while the old system and its staff are still available.

When a company retiring an old system wants to license the archive, SourceX can take it from qualification and rights review through preparation and delivery. Preservation and readability are only the starting point: whether the company may license the records is a separate question about contracts and privacy. Whether the company qualifies is answered by a short assessment, before any archive file is shared.

Frequently asked questions

What is representation information in digital preservation?

It is the information needed to interpret a file's bits as meaningful content, such as a schema, a code table, a format specification or the software that renders the file. The OAIS reference model treats a data object together with its representation information as the unit that is preserved. In a business archive it usually means schemas, lookup tables and reader software.

Is a checksum manifest enough to prove an archive is usable?

No. A checksum manifest proves the files are present and unchanged, which is necessary but not sufficient. It says nothing about whether someone can interpret coded values, dates or attachments. Pair it with a reproducibility test run by someone outside the original team.

How do I find out what format an unknown file is?

Run a signature-based identification tool such as Siegfried, which matches files against the PRONOM registry and reports a format identifier or flags the file as unknown. Do not rely on file extensions. Unidentified files need a person to open a sample and record what they are.

Should I keep the original files after converting them to CSV or PDF?

Yes. Conversions can drop formulas, layout, precision, embedded objects or metadata, and the only way to check or recover that later is from the original. Keep both and record any known differences in the manifest.

Who should sign off the archive manifest?

The system owner or IT lead should sign each manifest row, with a date and a next review date. Someone who never used the old system should run the reproducibility test, because the original team will read coded values correctly from memory and miss the gaps.

Sources

  • The OAIS reference model defines representation information as what maps a data object to meaningful concepts, illustrated by bits plus a description that they are temperatures in degrees Celsius. Source
  • The Library of Congress sustainability factors include external dependencies (need for outside software or hardware) and self-documentation (embedded descriptive and technical metadata). Source
  • The Recommended Formats Statement lists line-oriented formats such as CSV and platform-independent formats such as SQLite as preferred for datasets. Source
  • PREMIS models environment information, including software dependencies, and describes environments as minimum, recommended or known to work. Source
  • PRONOM is a technical registry of file formats and software that assigns persistent identifiers such as fmt/52 for Rich Text Format 1.7. Source
  • Siegfried is a signature-based file format identification tool implementing PRONOM signatures, with CSV and JSON output and container scanning. Source
  • The BagIt specification defines a complete bag (every listed file present) and a valid bag (every manifest checksum verified). Source
  • The Frictionless Table Schema declares field names, types, date patterns and which strings are treated as missing values. Source
  • The SQLite developers state their intent to support SQLite through the year 2050. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies