Each example is a complete inference: clues → ranked possibilities with probabilities and their basis → how to verify → scenario reconstruction. Follow them, or take your own clues to the bench.
A fixed window plus an even interval usually points to something periodic, driven by a device or a routine.
Location and the character of the smell are the two most informative clues. Drains, cabinets, disposals and drip trays each have their own test.
Start with the cheap checks: battery usage ranking and an airplane-mode comparison, before considering software or hardware.
Do not ask "does it look like a scam" — ask "do I have to act through what it gives me?" Links, urgency, transfers, codes.
Separating "dimming" from "switching off", then checking whether it is one lamp or a whole circuit, decides almost the entire diagnosis.
Rule out observable environmental factors (smell, sound, insects, temperature), then behaviour, then health — with clear thresholds for seeing a vet.
One wired test splits the problem into line-side and radio-side, before distinguishing channel congestion from an overheating router.
Whether the damp tracks the rain immediately or lags behind decides between exterior sealing and water pooling above.
Wait for a warm day: that single zero-cost test separates a reversible cold-weather effect from irreversible battery ageing.
Inference will not choose for you. It converts "I am torn" into a short list of things you can actually investigate.