When I was writing the third edition of my book I looked into this; assuming it hasn't changed in later releases I discovered in R80.30 w/ Gaia 3.10 that the per-minute value reported by cpview in historical mode is the average of two samples taken 30 seconds apart. Here are the relevant pages from my book comparing the historical granularity with sar, which can provide a nice "second opinion" in these types of historical cases:
Gateway Performance Optimization R81.20 Course
now available at maxpowerfirewalls.com