Skip to content

Systems and records

Shutdown archive verification: reopen a sample before access ends

By SourceX Editorial · Updated

9 sections

Short answer

To verify that a SaaS archive is readable before the account closes, draw a sample from the live system, have someone who did not run the export reopen it on a clean machine with no vendor login, and reconcile counts, attachments and checksums against the source. Capture links, codes and user maps that live only in the vendor account, and delay cancellation until every gap is fixed or explained.

System export

Shutdown archive

This page covers shutdown archive verification: reopen a sample before access ends.Illustrative

Key takeaways

  • A downloaded export proves only that a file exists; test intact, complete, readable and understandable separately.
  • Draw the verification sample from the live system so it can reveal what the export left out, then check a few items in the other direction.
  • Some exports hold links back to the service, not files, and several export downloads expire within weeks.
  • Reconcile counts by object type and year and keep a checksum manifest inside the archive.
  • Delay cancellation for any failure that cannot be fixed once the account is gone, and record the decision in a signed verification record.

What does it take to prove a SaaS archive can be read?

You prove an archive can be read by having someone who did not make the export open it on a machine with no vendor login, find a sample of records you chose in the live system, and confirm each one is present, intact and understandable. That test has to happen while the account is still open, because the live system is the only thing you can reconcile against.

Most shutdown plans stop at the first property in the table below: a file exists. A downloaded export proves very little on its own. It can be truncated, it can hold links that point back to a workspace that is about to disappear, and it can use codes nobody left at the company can decode.

A cold reopen means the archive is opened with nothing but what will survive the shutdown: the exported files, open tools such as a text editor, a spreadsheet, a database client or an email client that reads standard formats, and the readme you write. No vendor login, no browser session that is still signed in, and no help from the person who ran the export.

Treat readability as five separate claims and test each one. Each row needs its own evidence, and a failure in any row is a different kind of problem with a different fix.

What does it take to prove a SaaS archive can be read?
PropertyQuestion it answersTestEvidence to keep
ExistsDid the export finish and land in storage you control?List every export file with size and location, outside the vendor accountFile inventory with sizes and storage path
IntactAre the bytes unchanged since download?Generate a checksum per file at download and re-verify after each copyChecksum manifest and the re-verification log
CompleteIs everything that existed in the source represented?Reconcile counts per object type and year against the live systemCount reconciliation sheet with explained differences
ReadableCan a person or program open the content without the vendor?Cold reopen on a clean machine using only open toolsReopen log with tool names, versions and screenshots
UnderstandableCan someone outside the team tell what a record means?A reviewer explains sampled records using only the archive and its readmeReviewer notes and the questions they could not answer

Which reading dependencies vanish when the subscription ends?

The dependencies that break first are the ones that live in the vendor account: file links, download windows, user directories and field definitions. Several common exports carry references back to the service instead of the content itself, so they read fine today and go dark the day the account closes.

Slack is the clearest case. Its help center says JSON exports do not contain the workspace's files; they contain links back to them, and a file stops being reachable through that link if it, its channel or the workspace is deleted. An export that looks complete can therefore lose every attachment at cancellation.

Download windows are the second trap. Google says Workspace data export files are deleted 60 days after the export starts, minus the days the export took to finish, and Vault export links expire 15 days after the export begins. Atlassian says a Confluence Cloud backup link is available for 14 days, and Jira Cloud asks you to allow 48 hours between backups that include attachments. A failed reopen discovered late can leave no time to run the export again.

The Library of Congress lists external dependencies on particular hardware, operating systems or software among its format sustainability factors, alongside transparency to basic tools and self-documentation. That is a useful lens for the last category: anything you need in order to interpret a record that is not written inside the archive.

Which reading dependencies vanish when the subscription ends?
DependencyWhere it hidesWhat breaks after closureHow to capture it now
File links instead of filesChat exports, some ticket and wiki exportsAttachments become unreachableDownload the referenced files and record the link-to-file mapping
Download windowAdmin exports, eDiscovery exports, site backupsThe export itself is deletedNote the expiry date per export and download plus verify inside it
User and group directoryIdentity provider, app user listsRecords show IDs with no names or rolesExport the user list with IDs, names, roles and dates
Field and code definitionsCustom fields, picklists, status codesValues like 7 or ESC2 lose their meaningExport the schema and picklist values into a readme
Vendor viewerFormats only the product renders wellRich content shows as markup or blobsTest with open tools and keep a rendered copy where needed
People who know the systemAdmins and power users leaving in the wind-downNobody can explain workflows or codesRun the understandability review before their last day

How do you draw a shutdown archive verification sample?

Draw the sample from the live system, not from the export. A sample taken from the export can only tell you that what you exported is readable; it cannot show what the export left out. Then draw a second, smaller sample from the export and find each item in the source, so you test both directions.

Stratify instead of picking at random alone. Shutdown failures cluster in predictable places: the oldest years, the largest attachments, records owned by deleted users, private spaces the exporting admin could not see, and content with non-English names or unusual characters. A purely random draw can miss all of them.

Write down how each item was chosen, including any random seed, before you open the export. That keeps the test honest, lets someone repeat it, and stops the sample drifting toward records the team already knows are fine. There is no universal sample size; scale the number of items with the number of systems and how much the archive matters, and add items wherever a stratum fails.

Give the tester a sample sheet with one line per item: the system, the object type, the live identifier, a short description someone could recognize, and blank columns for found, opened, attachments opened and notes. The tester fills it in from the archive alone, and anything they cannot find is escalated, not explained away.

  • One stratum per system and per main object type (tickets, messages, deals, invoices, pages).
  • At least one item from the oldest and newest year in each system.
  • The records with the most attachments and the single largest attachment.
  • Records created or owned by users who have already been deactivated.
  • Private channels, restricted spaces or confidential projects, if the export claims to include them.
  • Records with edits, merges, deletions or restored versions.
  • A random draw across the rest, with the seed and the method recorded.

The archive reopen and source reconciliation checklist

Run these ten steps per system while admin access still works. Each row names what passes, so the person running it does not have to judge readability by feel.

The archive reopen and source reconciliation checklist
StepCheckPass conditionEvidence
1. DeadlineRecord the cancellation date and every export download expiryVerification can finish with time left to re-exportDated deadline sheet per system
2. InventoryList each source object type and the export file that should hold itEvery object type maps to a file or a documented exclusionSource-to-file map
3. FixityCreate checksums at download and verify after each copyAll checksums match on the storage copyManifest plus verification output
4. TotalsCompare counts per object type and year with the live systemEvery difference is explained by a named causeReconciliation sheet
5. AttachmentsCount attachments and confirm files, not links, are storedSampled attachments open from local storage offlineAttachment count and offline open log
6. Cold reopenA person not involved opens the archive on a clean machineSampled records found and opened with open tools onlyReopen log with tool versions
7. RelationshipsFollow sampled records to their thread, customer or parentLinks resolve inside the archive without the vendorTrace notes per sampled item
8. ReadmeWrite field definitions, codes, user map and export settingsReviewer can explain sampled records from the readmeReadme version and reviewer notes
9. ExceptionsList legal holds, deletion duties and data that must not be keptEach exception has an owner and an actionException register
10. Sign-offRecord results, failures and the go or no-go for cancellationSigned by the archive owner and a second reviewerSigned verification record

How do you reconcile export totals against the live system?

Reconcile at the level of object type by year, because a single grand total hides offsetting errors. If the live help desk shows a given number of tickets for each year, the export should show the same numbers, or a documented reason why not. Do the same for attachments by count and by total size.

Differences are normal; unexplained differences are the problem. Typical legitimate causes are retention rules that already deleted content, plan limits on how far back an export reaches, items in trash, and content the exporting role could not see. Slack, for example, says Free plan exports include file links only from the last 90 days and leave out data more than a year old, and a Pro export will not include year-old data that was deleted before the workspace upgraded, so a count gap there may be a plan limit rather than a broken export.

For the files themselves, use a checksum manifest. The BagIt format published as RFC 8493 is a simple, widely used way to do this: a bag carries a payload manifest that lists every payload file with a checksum, so anyone can confirm later that nothing is missing or altered. The NDSA's fixity guidance describes checksums and cryptographic hashes as the common tools for showing a file is unchanged, and notes that file system checksumming does not replace separately kept fixity data.

Keep the manifest and the reconciliation sheet inside the archive. A future reader, or a buyer's reviewer if the records are ever licensed, will ask the same questions you are answering now, and the answers are cheapest to record while the source still exists.

Worked example (illustrative): a help desk and chat archive fails its first reopen

The following case is illustrative; the company, counts and dates are invented to show the method. A 120-person services firm is closing and plans to cancel its help desk and team chat in six weeks. The operations lead has already downloaded both exports and stored them on a shared drive.

The verification lead draws 40 items from the live systems across five strata: oldest year, newest year, largest attachments, tickets owned by deactivated agents, and a random draw with the seed recorded. A finance analyst who did not run the exports performs the cold reopen on a laptop without the vendor logins.

Help desk results are mixed but fixable. Ticket counts match for every year except one, where the gap equals the tickets the retention policy deleted, which the admin confirms from the policy settings. All sampled tickets open, but status values appear as numbers, so the team exports the status and custom field definitions into the readme before the admin leaves.

Chat results fail. Messages open, but every sampled attachment is a link that requires a workspace login. The team pauses the chat cancellation, retrieves the referenced files while the workspace is live, stores them beside the export with a mapping from link to file, regenerates the checksum manifest and repeats the cold reopen. On the second run, the sampled attachments open offline.

The exception: a ticket queue under a litigation hold. The archive owner records it as excluded from any later reuse and keeps it under the hold instructions from counsel, so it is verified for readability but flagged in the exception register.

What made the difference was timing. Because the sample came from the live workspace and the reopen ran four weeks before the cancellation date, the link problem surfaced while it could still be fixed. Run the same test after closure and the only possible outcome is a note that the attachments are gone.

Which failures should delay cancellation, and which can be written down?

Delay cancellation when a failure would destroy content you cannot recreate; document and accept a failure when it only reduces convenience. The deciding question is whether the gap can still be closed after the account is gone.

Be specific about who can stop a cancellation. In many wind-downs, finance cancels subscriptions to stop spending, and the people verifying the archive are not told the date. Agree in writing that no subscription with archive content is canceled until its verification record says go, and give the archive owner the authority to hold it.

When cost forces a choice, rank systems by what would be lost. A chat workspace whose attachments are only links is a bigger risk than a help desk whose export already holds the files, even if the help desk is larger.

Which failures should delay cancellation, and which can be written down?
Failure foundRecoverable after closure?DecisionWho decides
Attachments stored as links to the live serviceNoDelay cancellation until files are retrieved and re-verifiedArchive owner
Count gap with no explanationNoDelay until the cause is found or the export is rerunArchive owner with system admin
Checksum mismatch on a stored copyYes, if the original download is intactRecopy from the verified original and re-verifyArchive owner
Codes or custom fields without definitionsOnly while admins remainCapture definitions before the admin's last daySystem admin
Content renders poorly in open toolsUsually yesDocument the limitation and keep a rendered copy of key recordsArchive owner
Records under a legal hold or deletion dutyNot a readability issueFollow counsel's instructions and log it in the exception registerCounsel

What goes in the verification record, and who signs it?

The verification record is a short, signed document that says what was tested, how, by whom, and what failed. It turns a folder of files into an archive someone else can rely on, and it is the first thing an auditor, a successor or a data reviewer will ask for.

At a glance

What goes in the verification record, and who signs it

Two signatures are better than one: the archive owner, who answers for completeness, and a second reviewer who ran or observed the cold reopen. Keep the record inside the archive next to the manifest so the two travel together.

Repeat the fixity check on a schedule after the shutdown, not only once. Storage media and copies fail quietly, and a manifest is only useful if someone compares it with the files from time to time and logs the result in the same record.

  • Archive identifier, systems covered, export dates and the tool or method used for each export.
  • Cancellation date and each export's download expiry.
  • Sampling method, strata, item list and random seed.
  • Checksum algorithm, manifest location and the last verification date.
  • Reconciliation sheet with every difference and its cause.
  • Cold reopen results, tester name, machine and tool versions.
  • Failures, the fixes applied and the result of each retest.
  • Exception register: holds, deletion duties, excluded data and owners.
  • Go or no-go for cancellation, with both signatures and the date.

If the archive might later become a licensed dataset

A readable archive is a precondition for any later reuse, but it settles nothing about rights. Being able to open records says nothing about whether contracts, privacy notices or customer terms allow them to be licensed, and that review is separate.

If a company in a wind-down wants to know whether its records are worth licensing to AI developers, SourceX starts with a short company assessment that needs no data uploads. The company approves the scope and terms and signs before anything moves, and an archive too large to transfer can remain in its own storage or travel on encrypted drives, since SourceX does not host multi-terabyte sets. The verification record described here is exactly the documentation that review would ask for first.

Frequently asked questions

How long before cancellation should archive verification start?

Start as soon as the cancellation date is known, and work backward from the shortest download window among your exports. Some export links expire within two weeks of starting, so a failed reopen found late can leave no time to rerun the export. Plan for at least one full retest cycle before the account closes.

Is a checksum enough to prove an archive is usable?

No. A checksum proves the bytes have not changed since you recorded it, which is fixity. It cannot show that the export was complete, that attachments are files rather than links, or that anyone can interpret the codes inside. Pair checksums with count reconciliation and a cold reopen.

Who should run the cold reopen test?

Someone who did not make the export and does not rely on the vendor login, such as an analyst from another team. They should use a clean machine and open tools only, so the test reflects what a future reader will face after the account is gone.

What should we do if the export window expires before verification is finished?

If the account is still open, run the export again and restart verification on the new files. If the account has closed, document what was and was not verified in the record and treat unverified parts as at risk. That is why deadlines go first on the checklist.

Does a verified archive mean the records can be licensed?

No. Verification shows the records exist, are intact and can be read. Whether they may be licensed depends on contracts, privacy notices, customer terms and any holds, which counsel reviews on the specific facts.

Sources

  • Slack's JSON exports contain links back to workspace files rather than the files, and deleted files stop being reachable through those links. Source
  • Google Workspace data export files are deleted 60 days after the export starts, minus the days the export took to complete, so they must be downloaded before that date. Source
  • Google Vault export links expire 15 days after the export begins, after which the export must be re-created. Source
  • Atlassian says a Confluence Cloud site backup download link is available for 14 days and attachments are included only if selected. Source
  • Atlassian says Jira Cloud backups that include attachments, avatars or logos need 48 hours between runs. Source
  • RFC 8493 defines BagIt, in which every bag has at least one payload manifest listing each payload file with its checksum. Source
  • The NDSA fixity guidance treats fixity as a file being unchanged, names checksums and cryptographic hashes as the common tools, and says file system checksumming does not replace separate fixity data. Source
  • The Library of Congress sustainability factors for digital formats include disclosure, adoption, transparency, self-documentation and external dependencies. Source
  • Slack says Free plan exports include links only to files from the last 90 days and exclude data more than a year old, and Pro exports will not include data older than one year that was deleted before upgrading. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies