Bob Benedict, KD8CGH
Well reversion to the mean finally hit me. I tested a 30 m LP filter that also looked great. Then I tested my 20 m LP filter and it has a first shoulder at about 10 MHz and then a second shoulder at 15 MHz so the loss at 14 MHz is about 4.4 dB. I rechecked the parts placement, torroid winding and solder connections and all seem fine. I made a model of the filter in SimSmith and played around with the values. The only way I got something like what I saw was to reduce L1 and L3 down to about 500nH.
I then pulled L1 and tested it in the impedance tester daughter board of an Analog Discovery 2. Doing a sweep after calibration it says the inductance at 14 MHz is 876 nH, too high! If I remove one turn the measured inductance goes down to 797nH. Unfortunately, according to the SimSmith model T1 being too high doesn't lead to the two shoulders, it only moves the shoulder a bit down in frequency.