Role of Cronbach's alpha in scale development

[Kurt Heisler]

I'm using confirmatory factor analysis to develop a multidimensional
instrument (i.e., several subscales). I plan on reporting subscale
scores.

What role should Cronbach's alpha play, if any, in scale development?
Or should it's use be restricted to simply reporting reliabilities of
subscales, *after* they have been developed based on factor loadings,
etc.

I'm aware that Cronbach's alpha assumes:

- no residual correlations
- that items have identical loadings
- unidimensionality

I've confirmed unidimensionality (for each subscale), and will
probably rule out correlated errors, so is it appropriate to report
Cronbach's alpha for each subscale? How "identical" do the loadings
need to be for Cronbach's to be accurate?

Thanks for your insights.