Practical guides · Practical guide

GxP backup and archiving: restore a system and retrieve its records

A backup job reports success. Could your team resume work after an outage? Could it still explain a historical report several years later? A backup and an archive serve different purposes and need different evidence. Using the same report, this guide gives QA, IT and process owners two worked test briefs, one for current system recovery and one for later access to historical records.

Table of contents

At a glance

What is the difference between a GxP backup and an archive?

A backup lets you restore a defined system and data state after loss or disruption. An archive keeps records retrievable and intelligible with their context throughout the retention period. Test each operation. Restore the intended version, then ask an authorised user to locate and interpret a historical record. A successful backup job tells you that the job ran. It takes a restore and a retrieval test to establish what you can recover.

  • A completed backup, a restored system and a retrieved historical record answer different questions.
  • The record may include attachments, metadata, history, signature links and technical dependencies.
  • Set retention periods and test intervals for the actual data and processes concerned.
  • A file format alone cannot establish whether the complete report will remain intelligible.

Two evidence objectives for the same record

For restoration, ask whether operations can resume from a known backup. For archive retrieval, ask whether a past event can still be reconstructed from its retained record. A file may exist without explaining what happened. Equally, a readable historical report will not restart a failed application.

Annex 11 covers storage and backup in section 7, business continuity in section 16 and archive access in section 17. Sections 6.17.1–6.17.2 of the UK MHRA data integrity guidance explain the different purposes further. This article concerns GMP for human medicinal products. The MHRA document provides UK regulatory guidance rather than EU legislation.

We developed the two test assignments for systems in routine operation. Decommissioning is not a prerequisite. An active application can retain earlier report versions in a separate archive. Identify protection, access and ownership for both datasets. One product can serve more than one purpose if each purpose has appropriate controls and evidence.

Sources: [2] [4] [1] [5]

Describe the full record before testing

The example uses manufacturing report HB-017. Its current version, V3, is held in the working application. Following a traceable correction, the earlier V1 dated 12 June 2024 is preserved in the archive. The identifiers, roles and times are fictional test data.

For V1, the package comprises the report showing 12.50 kg, the reference to work instruction VA-04 V2 then in force, two attachments A-1 and A-2, and reviewer TEST-QA-01's review entry with time and meaning. It also includes the metadata needed to interpret the record and its change history. This entry is a review of the report, not statutory batch certification.

The restoration test covers V3, its attachments, database relationships, required configuration and roles. The archive test specifically requests V1 rather than the latest view of HB-017. Use the actual data inventory to identify the source data and relationships needed for a complete record. A report title alone is insufficient.

Sources: [3] [4]

Define the test environment, reference state and responsibilities

Establish a controlled reference before testing. In case A, backup set S-10 represents the state at 10:00. No further business records are created before the simulated failure at 10:15. This limited case therefore does not test recovery of intervening transactions. Assess that gap separately for real operations.

For case B, a controlled contents list identifies every component of V1. Take expected values from the previously verified source package, not the archive display under test. Use a suitable isolated environment. Prevent the restore from sending real messages, posting transactions or triggering approvals in connected production systems.

Assign responsibility for the restore, review by the process team and decisions on findings. Define how to handle missing keys, licences or permissions. A test with unrestricted administrator rights cannot establish that an authorised business user can retrieve the report. Document relevant differences between the test and production environments.

Sources: [2] [3]

Test A: restore the current state

This worked assignment combines technical restoration with reconciliation by a business user. The application remains outside production. The 90-minute recovery target is an assumption for this example, not a GMP limit. Define the real target from process needs and continuity arrangements.

The clock starts at the declared simulated failure at 10:15 and ends when operational usability of the agreed scope is documented. Starting the database service does not stop the clock. Excluding necessary steps would change what the test measures.

Test A: completed example restoration plan
CheckInstruction and executionAcceptance and evidence
Objective and scope Restore HB-017 V3 from S-10 in REC-A and simulate an interruption. V3 and its required context are usable within at most 90 minutes from 10:15.
Inputs Provide S-10, the database, attachment store, matching configuration K-3, roles and restoration instructions. The 10:00 backup is identified and all required components are linked in the inventory.
Execution Restore components in the defined order. Keep production interfaces disabled in REC-A. Record the steps, versions used and technical messages.
Completeness Compare HB-017 V3, associated attachments and relevant metadata with the reference inventory. No missing components or unexplained discrepancies within scope.
Meaning Reconcile values, units, relationships and status with the current reference. Content and meaning are preserved. V3 is not presented as V1 or as an unreviewed draft.
Usability Sign in as an authorised business user. Perform a permitted downstream activity on a marked test copy. Access and the required workflow operate as intended. Historical approvals are not triggered again.
Decision IT and the business reviewers record elapsed time, findings and the scope restored. Assess results against the criteria. Do not claim complete success while a gap remains. Verify archive retrieval separately.

Sources: [2] [4]

What a successful restore demonstrates

If the team established usability at 11:20, 65 minutes would have elapsed. That meets the assumed time target. It would still leave the test incomplete if an attachment were missing or the wrong role version had been restored. Time, completeness and usability are separate criteria.

This assignment tests a known state with no intervening business transactions. It does not establish that 15 minutes of data loss is acceptable. For an active process, assess the backup state, pending transactions and reconstruction needed. A technically acceptable recovery point does not authorise the loss of GMP records.

A reasoned selection of cases can demonstrate particular relationships and functions. It does not cover every data type or failure mode. Combine appropriate technical completeness checks with tests of actual use. Consider distributed storage, encryption, interfaces and possible failures when choosing additional cases.

Sources: [2] [4]

Test B: retrieve the historical report and its context

Case B concerns the same report, HB-017, but requests V1 from 12 June 2024. It tests intended historical access in ARC-B. The environment must expose the preserved record reliably. A working current application or a restored V3 cannot answer the historical question.

An authorised archive reader searches without administrator rights. TEST-QA-01 must remain identifiable as the original reviewer even if the original account is inactive. The test requires neither reactivating that account nor approving the old report again.

Test B: completed example archive retrieval plan
CheckInstruction and executionAcceptance and evidence
Objective and inputs Request HB-017 V1 dated 12 June 2024 from the controlled archive inventory. The reader can find V1 specifically. The current V3 is not returned as a historical substitute.
Access Use the archive reader role in ARC-B. Search by report ID and version. Authorised retrieval works without reactivating the original reviewer's account.
Components Open the report, attachments A-1/A-2, reference to VA-04 V2 and required metadata individually. Every component required by the inventory is accessible and clearly linked.
Historical context Reconcile 12.50 kg, the unit, instruction in force at the time, review entry and relevant history. Meaning, reviewer identity and timing remain clear. No recalculation using current master data.
Protection As an archive reader, attempt to edit the retained report. Trace the designated correction path. Ordinary read access permits no uncontrolled change. Existing history is preserved.
Missing context Break the link to A-2 in a separate test copy. The defined completeness check detects the gap. Do not confirm the package as complete.
Closure Business reviewers and IT document retrieval, checks of context, permissions and findings. Confirm the criteria met or identify individual gaps. Suitability is limited to the tested state and scope. It does not automatically extend to every future environment.

Sources: [2] [3] [4]

Further reading:

A readable report can still lose its historical meaning

If HB-017 V1 loses its attachments or shows only the current instruction, reviewers may be unable to reconstruct the original assessment. Check meaning and relationships. A matching filename or hash cannot substitute for that content check.

Choose a preservation method that retains the information needed from an electronic record. A PDF can be suitable for certain completed content. Dynamic data, relevant metadata and analysis capabilities may require more. Determine essential properties from the purpose of the record. There is no general PDF requirement here.

Prevent historical reports from being recalculated silently using current formulae, limits or master data. A present-day view could differ from the version actually reviewed. Make clear whether the reader is seeing a retained original, a suitable copy or a traceably generated representation.

Sources: [4] [2] [3]

Set retention and protection for the actual records and dependencies

Determine retention from the record type, activity and applicable requirements. Chapter 4 distinguishes batch documentation from other documentation. Avoid applying a blanket ten-year rule to all IT data. Record the basis, triggering event and dependencies needed for longer in the inventory.

Archives also need protection against loss and uncontrolled change, including recovery arrangements. Identify shared dependencies that could prevent both operational recovery and historical access, such as the same storage, a lost key or an expired service account.

Include software, licences, keys, roles and technical documentation. If an old file requires a particular reader, maintain a controlled means of access for the required period. When the format or reader changes, demonstrate that the necessary content and context survive.

Sources: [3] [2] [4]

Connect operating responsibilities, cloud services and retesting

Ask the provider what its restoration test actually covers. Evidence from a platform test may be valuable without covering your dataset, keys or business access. Agree participation, available support and the results you need. A generic cloud assurance cannot establish historical access.

In small and medium-sized businesses, incomplete inventories often drive the work. Attachments may sit outside the database, old reports may use other templates or access may depend on one individual. Prepare a specific package and known reference state. IT can assess restoration. The process owner must be able to judge whether the report still means the same thing.

Repeat or extend checks according to risk and relevant changes. A new backup mechanism, encryption scheme or viewer can affect different functions. There is no universal requirement here for a full restore every year. Record which evidence remains current, which cases need repeating and who will close them.

Sources: [2] [4]

Can your team demonstrate both restoration and historical access?

Check each objective and then check their common dependencies. This worksheet helps you establish what an existing record demonstrates. Plan the tests for the application you actually use.

  • Define a separate test assignment for restoration and historical retrieval.
  • Identify the required records, context and attachments.
  • Take reference data from a controlled starting state.
  • Prevent unintended effects on production systems.
  • Check completeness and usability as well as technical restoration.
  • Justify time targets and recovery state for the process. Keep unresolved gaps visible.
  • Retrieve the specified earlier version and its relationships in the archive test.
  • Cover access, change permissions and technical dependencies.
  • Assign the retention basis and ownership to the actual records.
  • Make follow-up decisions for findings, scope limitations and relevant changes.

Sources: [2] [3] [4]

Project experience and collaboration

Test system recovery and later data retrieval separately

We give IT and process owners two tasks. Restore a failed system version, then explain a historical business transaction. The teams show which backups, archives and readers they use for each. A successful restart may answer the first question while leaving the second unresolved. Depending on the gap, we extend the restore test or plan a separate retrieval test with records, metadata and dependencies. We document ownership and retention needs for each purpose.

Frequently asked questions about Backup and archiving

Does a successful daily backup demonstrate archiving?

No. It shows that the backup ran according to its technical criteria. Test both actual restoration and long-term historical access with context. Rolling backups can overwrite earlier states. Check their settings against the required retention.

Sources: [2] [4]

Do we need separate backup and archive systems?

Not necessarily. One platform can serve both purposes if the controls and evidence cover each. Assess protection, retention, access and restoration separately. Check whether a shared key loss or erroneous deletion rule could compromise both purposes without detection.

Sources: [2] [4]

Must every restore test include a manual review of every record?

Choose a justified combination of technical controls and business cases. Data types, relationships and failure modes may need different checks. State what the selection demonstrates. Opening one report cannot establish that every record is present. Manual review of every record is not the only possible method.

Sources: [2]

How often should restoration and archive access be tested?

Set scope and frequency from risk, process needs and relevant changes. Annex 11 addresses checks of restorability, periodic monitoring and the ability to access archives after relevant changes. It sets no universal annual interval here. A new backup method or viewer may need targeted testing before the next scheduled check.

Sources: [2] [4]

Is a PDF always sufficient for long-term retention?

No. Determine whether it preserves the complete record in a usable form. It may suit some completed records. Metadata, dynamic data and relationships may require additional components. In this example, a PDF without the attachments and historical review reference is incomplete.

Sources: [4] [3]

Which metadata must be retained with a report?

Identify the information needed to understand and trace it. This may include version, relationships, units, creation and review times, identities, relevant history and attachments. Base the test on the record inventory and demonstrate the relationships on retrieval. Assess the need for each technical item.

Sources: [4] [3]

May the original reviewer's account be inactive for archive access?

It need not be reactivated simply for historical retrieval. Preserve the identity and meaning of the original review and provide access to currently authorised readers. If only an unresolvable ID remains, context may have been lost. Test historical attribution and current access separately.

Sources: [4] [2]

Does the backup retention setting determine the GMP retention period?

No. Identify the applicable requirement for each record type. A default backup setting cannot establish it. Record the trigger, period and dependent documents in the inventory. Avoid a blanket “ten years for everything” when records have different retention bases and context needs.

Sources: [3] [4]

Can the cloud provider perform both tests?

It can deliver agreed technical tests. Assess whether the results cover your records, configuration and business context. With suitable provider evidence, focus the customer's checks on remaining questions. Agree deliverables, support and access conditions. The extent of additional testing depends on the actual scope.

Sources: [2] [4]

What if restoration succeeds but historical retrieval fails?

Record the outcomes separately. Restoration may meet its defined criteria while historical access remains deficient. Investigate missing attachments, unresolved versions or an unsuitable viewer. Assess the scope and effect of each gap, then identify the next action. An unqualified overall pass would obscure the open issue.

Sources: [2] [4]

Sources and review date

The primary sources and MHRA status page were checked on 6 October 2026. The scope is GMP for human medicinal products, with supplementary UK guidance from the MHRA. The notice giving OECD guidance precedence concerns UK GLP. We selected the test assignments, dates, data and version identifiers for the worked examples.

  1. European Commission, EudraLex Volume 4: GMP for human medicinal products
  2. EU GMP Annex 11: Computerised Systems, sections 7, 16 and 17
  3. EU GMP Chapter 4: Documentation, records and retention
  4. MHRA, GxP Data Integrity Guidance and Definitions, revision 1, March 2018, sections 6.17–6.20
  5. MHRA, Guidance on GxP data integrity: publication status and UK GLP notice

Sources checked: .

Daniel Herrmann, Daniel Herrmann Consulting

Daniel Herrmann Consulting

Discuss how to demonstrate restoration and historical access

We help QA, IT and process owners define the data, controls and tests they need for backup, restoration and access to historical records.

  • GxP system decommissioning →

    Retire legacy systems while preserving data, context and auditability. We plan and manage the decommissioning so that data, metadata and audit trails remain readable and retrievable after shutdown.

Send a project enquiry Customer testimonials →