I did some more digging into Ford steering gear parts and came across this web page
with drawings from the Ford parts catalogs. It appears the 4-digit numbers pertain to part categories rather than specific parts (e.g. #3575 for the sector is used from 1932 to 1953, no matter whether it was a fixed sector (up to 1936) or roller sector (from 1937 onward), as well as on F1 and F100 pickup trucks from 1940 to 1960.
with suggestions for steering gear adjustment states: "The worm and roller type gear was used from 1937 to 1948. Ford used the Gemmer model 305 in Fords from '37 to '40 and in Mercurys from '39 to '42. From 1941 to '48 (Ford) and from 1946 to '48 (Mercury) Ford used their own manufactured version which was similar to the Gemmer 305 gearset. Some minor differences in parts occurred over the years." I have a hunch these changes may have been quite significant.
It seems to be the prefix that makes the part specific, e.g. prefix 68 for Ford Model 68 in 1936, and prefix 78 for Ford Model 78 in 1937. However, some prefixes I found do not make sense (I was not able t match prefix "B7A" to a Ford model designation.
Looking for 3-tooth rolling sectors on the Macs web site yielded several results:
Steering Gearbox Sector Shaft Gear And Pin - 0.442 Pin OD
, Part #: 49-19363-1, Alt Part #: 8A-3575-X
Steering Gearbox Sector Shaft Gear And Pin - 0.472 Pin OD
, Part #: 49-25200-1, Alt Part #: AB-3575-X
Ford Pickup Sector Roller & Pin - F100 & F250
, Part #: 48-30119-1, Alt Part #: B7C-3575-X
Furthermore, this search also revealed additional worms (e.g. 3 worms, including an RHD worm for Pickup trucks -- presumably all with 3-tooth sector rollers), but I'm mentioning this passenger car one because it specifically mentions a 3-tooth roller:
Steering Shaft Worm Gear - For 3 Tooth Sector
, Part #: 49-29680-1, Alt Part #: B7A-3524-X
In order to assess parts interchangeability, it seems to be important to make a list of worm shaft diameters and roller pin diameters used on various Alfa models (e.g. 1900, 102, and 106) and Ford models.
The good news is we seem to have some options. The bad news is it requires some effort to figure out which options are valid.