Overhead ratios need a same-build full-length baseline, not a scaled prefix

When a metric is 'post-processing CPU / simulation CPU without dumping' (e.g. converter cost relative to the sim that made the dump), the denominator must come from the same build, test and cycle count that produced the dump. A short prefix run (e.g. 25K of 6.5M cycles) includes startup and different activity phases, so scaling it linearly gives only an ESTIMATE. Often the full no-dump run is cheap (minutes) compared with the dump run, so ask for one rather than publishing a scaled ratio. If the original build is gone, rerun on the current build and label the ratio 'current build, not the original receipt'.