Write down the decision boundary
Name the problem, the affected architecture items, the environment or project scope and the assumption under which the decision is valid. A choice made for one integration should not become an unexplained policy for every integration in the organisation.
Record real alternatives
A useful architecture decision record explains why the selected option was preferred, what was rejected and which trade-offs were accepted. Include constraints such as operational support, security, latency or delivery cost where they materially affected the outcome.
Connect decision and design
Link the record to the applications, interfaces, technologies, principles, standards and patterns it affects. That lets a future change review find the rationale from the model, and lets a decision review see what would be impacted by a reversal.
Keep approval and evidence distinguishable
An approved decision is not the same as implemented control or independently verified operation. A review can accept a design with conditions; later test and operational evidence can close those conditions. Keeping those stages separate prevents an approval label from implying more assurance than exists.
Prefer a short record that stays current
A concise, consistently indexed decision with durable links is more valuable than a lengthy document nobody can find. Capture the minimum facts at decision time, then append evidence and supersession information as the architecture evolves.
Strong architecture practice depends on precise models, explicit ownership and decisions that remain useful as systems change. Examples are fictional and are presented as architectural guidance.