A Filled Architecture Review Decision Record
An illustrative keep, repair or replace decision for a customer-request workflow, including data, acceptance, recovery and the first release boundary.
Fictional system, explicit decision
This filled example is a demonstration, not a client assessment, validated production architecture or estimate. A fictional service business receives customer requests through a portal. Staff copy the approved information into an older operations system. The goal is to make that handoff reviewable without replacing the whole platform.
Keep, repair or replace
| Component | Proposed decision | Reason and evidence still needed |
|---|---|---|
| Existing customer accounts | Keep initially. | They already identify users; verify access boundaries and account recovery. |
| Request capture | Repair validation and status visibility. | Missing account references cause avoidable handoff errors in the fictional example. Confirm with real records before a live decision. |
| Manual handoff | Replace one bounded step with a pending integration record. | An approved request needs one traceable destination and duplicate protection. Confirm supported interfaces. |
| Operations database | Keep behind a stable interface. | There is no evidence that replacing it is necessary for the first useful release. |
| Reporting | Defer broader redesign. | First record status and failures consistently; revisit reporting after the workflow is understood. |
First release boundary
A user can create a request and see its status. A reviewer can inspect it and resolve required information. An approved request becomes one integration job, with a stable reference and a visible failure state. No new payment, insurance decision or autonomous external action is included.
Acceptance evidence
- A complete request can be reviewed and approved by the appropriate role.
- An incomplete request remains in review until the missing information is resolved.
- The same approved reference cannot create two destination records when retried.
- A failed integration leaves a recoverable state and an accountable owner.
- The existing process remains available until transition criteria are met.
These are proposed tests for this example. They do not establish a measured result in another system.
Data and recovery
Before a real change, agree the source of truth, record identifiers, reconciliation and any schema compatibility requirements. Capture the state needed to recover an interrupted handoff. Confirm backup and restore procedures for data changes.
A feature flag can route traffic. It cannot, by itself, reverse an incompatible migration or an external action that already happened. Some transitions need a maintenance window. The assessment should expose that limit before implementation.
Decision and next step
Recommendation in this fictional example: retain the account and operations systems, repair request validation and build one reviewable handoff. First verify the interface, permissions and duplicate behaviour. If those prerequisites fail, reconsider the release boundary before committing to a build.
Try the interactive portal example or download the blank architecture worksheet.