How would you troubleshoot a problem you have never seen before?

Clarify the expected behavior, actual behavior, scope, and time of onset. Gather exact errors and identify recent changes. Form a small number of hypotheses and test the cheapest, safest ones first, changing one factor at a time where possible. Record observations so failed attempts are still useful. Compare a working case with a failing one and narrow the boundary where behavior differs. Escalate with a concise evidence summary when needed. Random configuration changes can hide the cause and create additional failures.