Enterata All articles
Data Architecture

Proof at the Restore Point: Rehearsing Recovery Across UK Systems

Enterata
Proof at the Restore Point: Rehearsing Recovery Across UK Systems

Photo: Mangos, CC BY-SA 3.0, via Wikimedia Commons

Backup dashboards answer a narrow question: did the job run? They do not answer the question that matters during an incident: can the organisation bring a service back, within a stated time, with the right access and in the right sequence? For UK teams approaching the 2026 year-end, that gap is no longer a technical curiosity. Recovery claims appear in board assurance packs, cyber-insurance renewals and supplier due diligence. A green job history is not proof. A dated rehearsal record is.

Separate the claim from the proof

A recovery claim is a statement about capability. We can restore this system within four hours. We can recover this data to a point no older than fifteen minutes. We can rebuild this configuration from source. The proof is a record showing that someone tested the claim, under controlled conditions, and observed the result. Without that record, the claim is a belief held by the team that owns the system.

The distinction matters because backup tools are optimised for completion, not for restoration. They report that data was written to a target. They rarely report whether the data can be read by the application, whether the access model still works, whether dependencies are present, or whether the restore would create a duplicate identity, a broken integration or a regulatory problem.

Build a matrix, not a policy

A restore policy describes intent. A restore matrix describes what will be rehearsed, by whom, and what evidence will remain. The matrix should cover four asset classes that behave very differently in a UK estate.

| Asset class | Rehearsal focus | Evidence to retain | | Databases | Point-in-time restore to an isolated environment; application smoke test; row counts and referential checks | Ticket reference, source instance, backup set, restore point, target environment, operator, start and finish times, validation results, exceptions | | SaaS exports | Export completeness, attachment and metadata recovery, re-import into a non-production tenant, retention-window limits | Export manifest, record counts, sample reconciliation, vendor ticket, tenant used, data-handling approval, gaps found | | Configuration | Infrastructure as code, DNS, certificates, secrets references, integration mappings, feature flags | Repository commit, build artefact, environment diagram, deployment log, post-restore configuration comparison | | Identity data | Directory objects, groups, privileged roles, service accounts, MFA registration state, break-glass access | Restore scope, objects recovered, role-mapping test, access review sign-off, exceptions and temporary controls |

The matrix should be scoped by criticality, dependency and change velocity. Do not attempt a full estate rehearsal in one weekend. Select representative systems, including at least one that has changed significantly during 2026, and rehearse the restore path that would actually be used in an incident.

Databases: prove the application, not only the file

A database restore that produces a running instance is only the first step. The service is recovered when the application can use the data. That means testing logins, key queries, batch jobs, integration endpoints and report generation. Row counts alone can mislead if the restored data is structurally intact but semantically wrong.

UK teams should also rehearse the permissions path. A restore performed by a platform administrator may succeed while the application service account cannot read the recovered schema. The drill should record who performed the restore, which credentials were used, and whether the normal operational team could repeat the process without informal help.

SaaS exports: read the retention clause before the incident

Software-as-a-service recovery is often treated as the vendor's problem. In practice, the vendor may provide availability, backup and export as separate commitments. An export that omits attachments, audit history or deleted records may not meet the organisation's recovery requirement. A restore that depends on a retention window of thirty days may fail a legal or operational obligation that runs for years.

A useful rehearsal takes a sample export, imports it into a non-production tenant, and checks whether the business can work with the recovered records. It should also test the export request process: who can ask for it, how long it takes, what format is provided, and whether the data can be handled lawfully in the target environment.

Configuration and identity: the dependencies that decide the outcome

Configuration restore is where many recovery plans quietly fail. Infrastructure as code may rebuild compute, but DNS records, certificates, secrets references and third-party mappings may live elsewhere. A restored environment that cannot authenticate to a payment provider or a messaging service is not recovered.

Identity data deserves its own rehearsal because it sits underneath almost every other service. Restoring a directory may also restore stale accounts, disabled joiners or privileged roles that were removed for good reason. The drill should prove that access can be recovered without reopening risks that the organisation deliberately closed. Break-glass accounts should be tested, but their use should be logged and reviewed.

What evidence makes a claim defensible

For each rehearsal, retain a small, consistent evidence pack. It should include the scope, the asset and version tested, the restore point, the target environment, the operator, the start and finish times, the expected recovery time, the actual recovery time, the validation checks performed, the exceptions found and the sign-off. Where a test failed, retain the failure record. A failed drill with a remediation action is stronger evidence than a claimed success with no artefacts.

Auditors, insurers and board committees do not need every log line. They need a traceable line from claim to test to result to action. Set a retention period with legal and insurance advice, and keep the evidence outside the system being restored.

A practical next step

Before the 2026 close, choose three systems: one database-backed service, one SaaS service with a business-critical export, and one identity-dependent platform. Rehearse the restore, record the evidence, and note the exceptions. The goal is not a perfect test. The goal is to replace an untested claim with a known result, while there is still time to act on it.

All Articles

Related Articles

The Queue That Remembers Nothing: Auditing Undelivered Events Before the 2026 Close

The Queue That Remembers Nothing: Auditing Undelivered Events Before the 2026 Close

Clones Without Custody: A Data Audit for UK Lower Environments

Clones Without Custody: A Data Audit for UK Lower Environments

What the Blanks Are Telling You: A Field-Level Audit of Missing UK Customer Records

What the Blanks Are Telling You: A Field-Level Audit of Missing UK Customer Records