Signed, notarised and stapled. Download, unzip, drag to Applications.
The SNR readout was inverted. It showed 0 dB on a station with a locked pilot and 12 dB on bare noise. The channel measurement was correct all along — it was the noise reference that was wrong, by about 35 dB.
The noise floor was sampled from a narrow window at each end of the capture, positioned by a helper that describes the Airspy HF+'s filter skirts and returns zero for everything else. On an RTL dongle that put the window on the absolute outermost bins — at 1.2 MSPS, exactly the anti-alias roll-off and fold-back, the one part of the span where the tuner's output means nothing. Every bin it looked at, the quietest included, sat 20–45 dB above the real noise.
The floor is now a low percentile of the whole usable span rather than two windows at the edges, and the excluded edge zone has a sensible minimum on radios that declare nothing wider. Measured on air: the reported floor moved from −55 dB to −95 dB, and the meter now agrees with the pilot-derived MPX S/N.
The same value feeds the AGC's floor-rise test, which was therefore judging roll-off rather than noise. That should improve too, though it has not been measured.
Everything from 4.1.8 is included.