Long-term upper-tropospheric humidity (UTH) trends are sensitive not only to sampling and observing-system changes but also to the vertical definition used for cross-product comparison. We analysed the CM SAF version-2 microwave UTH climate record over 60° S–60° N and tested whether ERA5 relative humidity at 200–500 hPa, individually or as nonnegative vertical mixtures, could represent its 1996–2018 change. Area-weighted CM SAF UTH declined by 1.238 percentage points decade⁻¹ (95 % moving-block-bootstrap interval: −1.472 to −1.008; AR(1)-adjusted P = 0.003), with negative slopes in 97.3 % of valid grid cells. The sign persisted across seasons, latitude bands, coverage thresholds, later start years, and a platform-transition audit. ERA5 showed weak and uncertain trends at 200–300 hPa but significant moistening at 400–500 hPa. The CM SAF trend was 1.017 percentage points decade⁻¹ more negative than the most negative ERA5 level and therefore lay outside the envelope of every non-negative mixture of the five pressure levels. A mixture optimized for spatial trends retained only moderate correspondence (r = 0.474), a +0.991 percentage-points-decade⁻¹ bias, and 65.8 % sign agreement. A separately optimized mixture failed to predict detrended annual variability under leave-one-year-out validation (r = −0.098). Thus, the satellite decline is internally robust but is not representable by fixed-pressure ERA5 humidity or simple vertical averaging. Cross-product climate assessments require an observation-operator-aware comparison rather than selection or averaging of pressure levels.