I had to spin sideways, and plot it out, to see what you mean.
Yes, "45" and "50" are both the same.
Also interesting that there seems to be one more switch position than you need to count 20...70 by 5. Some of them are not numbered. And 11-position switches are a little more common than 12-position switches.
Figuring just the front-end ideal gain and the between-stage loss pad, then assuming the gain in input transformer and output stage is 32.4dB (to make the "20" position correct), I get:
SW _ SA __ SB _ Gain __dB__ __ add32.4dB
20 - 2.4 - 0.10 = 0.24 = -12.4dB -- 20
?? - 4.3 - 0.10 = 0.43 = -7.3dB ---- 25
?? - 7.7 - 0.10 = 0.77 = -2.3dB ---- 30
?? - 13.8 - 0.10 = 1.38 = 2.8dB ---- 35
35 - 25.4 - 0.10 = 2.54 = 8.1dB ---- 40
40 - 47.3 - 0.10 = 4.73 = 13.5dB --- 46
45 - 92.6 - 0.10 = 9.26 = 19dB ----- 51.4
50 - 92.6 - 0.10 = 9.26 = 19dB ----- 51.4
55 - 92.6 - 0.18 = 16.7 = 24dB ----- 56.4
60 - 92.6 - 0.32 = 30 === 30dB ----- 62.4
65 - 92.6 - 0.56 = 52 === 35dB ----- 67.4
70 - 92.6 - 1.00 = 93 === 39dB ----- 71.4
I think he drew a 11-position network on a 12-position switch, and missed the mistake.
If you throw-out the "45" position, and then re-number down to "20", it is all pretty close.
The consistent error in the 45-70 range is due to me assuming an amplifier with infinite gain. The real amplifier's gain will be just about right to knock-off a dB or two, making it all exact within reasonable tolerance.
