Here's a map to show the results of the different methodologies I'm considering, using Las Vegas as an example.
In red there's the boundaries based on 1960 densities (5000+ ppsm) - or in areas where only 1970 census tract data is available, then densities of 6000+ ppsm.
In yellow I've also allowed census tracts with densities above the MSA weighted average AND auto commute mode share below the MSA average to qualify. Also in MSAs where the urban core(s) is still less than 20% of the MSA, such as with Vegas, I've also lowered the 1960/1970 density requirements to 3000 ppsm and 4000 ppsm respectively.
In blue, in addition to allowing census tracts to qualify in all the ways above, I've also allowed them to qualify if the either the density OR non-auto commute mode share is at least 2x the MSA average and that variable is above the average by more than the other factor is below the average. The main purpose of this is for places like CBDs, river fronts, harbours, or in this case the Vegas strip that have low residential densities because most of the land is non-residential in use but very high non-auto commute mode share. You could also have census tracts especially in cities where auto mode share is high everywhere and not necessarily as good of a great predictor of urbanity (by that metro area's standards) as density.
Example:
Density is 2.3x more than the weighted average
Non-auto mode share is 2.2x less than average
= qualifies
Density is 1.9x more than the weighted average
Non-auto mode share is 1.8x less than average
= does not qualify
Non-auto mode share is 2.2x more than average
Density is 2.3x less than the weighted average
= does not qualify
In most cities this last criteria probably won't make much of a difference, but in Las Vegas it does because the strip has little housing (that isn't hotels) and also separates some census tract clusters from the main urban core that don't form a sufficiently large cluster to qualify as a secondary urban core (pop of 25k+), but which would otherwise qualify.