My workflow is divide-and-conquer, and I never trust a single sweep of the whole run. First I establish a baseline: two carriers at the normal transmit power, third-order PIM, with the feeder terminated in a known low-PIM load. That number tells me whether the test set, cables and adapters are clean before I blame the site. If the baseline is bad, I fix the test setup first, because a dirty adapter will send you hunting a rusty bolt that does not exist.
Then I work from the top down, one segment at a time. I test the antenna alone, then the main feeder, then each jumper and connector in turn, always with low-PIM adapters and connectors torqued to the specified value and cleaned before mating. A connector that improves 10 dB after cleaning and re-torquing is a suspect, but I do not stop there: I replace it and re-measure, because a re-torque that only holds for an hour is not a repair. I also log the PIM before and after each change, so the fix is a number, not a feeling.
The tap test is the most revealing step for a genuine rusty-bolt source. While watching the PIM level I lightly tap each connector, flex each jumper, and move mounting hardware. A corroded or loose joint jumps 10 to 20 dB with mechanical disturbance and settles back, while a solid joint stays quiet. I look beyond connectors too: rusted antenna mounting bolts, loose ground-bar hardware, cabinet door seams, and any ferromagnetic material in the RF path, such as nickel-plated or steel screws, all belong on the suspect list, and I replace them with non-magnetic hardware where possible.
Two patterns help narrow it down. PIM that appears only in wind or vibration points to a loose mechanical joint; PIM that worsens after rain points to water ingress and corrosion inside a connector. After the repair I re-run the full-power, two-carrier test and watch stability over several minutes, then repeat it after the site has been through a temperature cycle. A fix that only survives one sweep is not a fix.