Diagnostic workflow when a die starts losing tolerance

The part is the output of the system, which makes it the highest-information diagnostic available before any teardown. What you are looking for, in order:

Which dimension drifted. Pull the inspection sheet for the last in-tolerance part and the first out-of-tolerance part. Identify the specific feature that moved. A part that is 0.05 mm over on a wall thickness is a different diagnostic than a part that is 0.05 mm short on an overall length.

In which direction. Is the part bigger than the print or smaller? A wall thickness that grew (part thicker than print) is consistent with the die cavity opening up, which is washout or plastic deformation in the die. A wall thickness that shrank is rarer in forging and points at process changes, billet shrinkage from a cooler delivery temperature, or alignment shift.

Uniformly or locally. Is the drift the same on every part, or does it vary in a pattern? Uniform drift across hundreds of parts points at a system-wide change (process or die wear). Localized drift, where one corner of the part is out and the rest is in tolerance, points at a localized cause (uneven cooling, uneven lube, uneven wear on one feature).

Trending or step-change. Did the drift appear suddenly or build over the last few thousand hits? A step-change at a specific hit count points at a discrete event, a lube tank refill, a shift change, a billet lot change, a tool service. A slow trend points at progressive wear or a slowly drifting process variable.

A part that is 0.05 mm thick over tolerance, uniformly, with a sudden onset at a specific shift, with the rest of the dimensions in spec, is a different problem than a part that is 0.1 mm short on one feature, drifting over 5K hits, with surface finish degrading at the same time. The triage steps that follow are tuned by what the part is telling you.