Teaching an agent to stop sounding like an agent
A case study in using Caliper to evaluate blader/humanizer, tighten voice calibration, and turn the improvement into an upstream contribution with regression coverage. With a postscript on what the maintainer did instead.
