Prefer rigorous evidence-driven debugging and definitive conclusions

When troubleshooting, systematically verify each layer with concrete state, logs, repeated checks, and regression-style validation before declaring success or assigning a root cause. Clearly separate what is proven from what remains unverified, isolate local behavior from control-side or external behavior, and avoid hedged claims such as “probably works.” If tooling lacks privileges, ask the user to run the required command and then verify its effects. When documentation searches repeatedly fail or return irrelevant material, stop guessing URLs and pivot to direct configuration inspection or already-collected evidence. Favor durable fixes and safeguards that prevent recurrence, not merely one-off workarounds.