Start with the question
Write the decision as a choice you can act on: keep the current search or trial a dedicated engine. Name the decision owner and the hard constraints before listing candidates. Keep the current approach in the comparison so that a new tool has to justify the change.
Separate facts from assumptions
For each important claim, save the source URL and the date you checked it. Say whether it describes documented behavior, your own observation or an inference. Documentation can identify a feature to test; it does not establish relevance, reliability or cost on your workload. Leave missing evidence visible.
Explain the choice
Compare candidates against the same criteria and record the tested configuration. Write an architecture decision record with context, alternatives, reasoning and consequences. When the evidence is insufficient, propose the next test rather than selecting a winner. In the search example, the current decision is to keep the baseline while preparing a common trial.
Leave a path back
State what would change the decision and who should review it. Attach a task for missing tests and keep their observations linked to the record. When new evidence arrives, read the earlier rationale before updating the conclusion. Keep the reasons for the revision understandable to someone who was not in the original conversation.