What is the practical difference between Type A and Type B uncertainty evaluation?

The distinction is about how you obtained the information, not about the nature of the effect:

• Type A: you evaluated it from your own repeated observations in this measurement — standard deviation of the mean, ANOVA of a nested design, regression residuals. Comes with degrees of freedom = n−1 (or from the design).
• Type B: you took it from elsewhere — a calibration certificate (divide U by its k), a spec limit (rectangular, divide by √3), resolution (half-interval rectangular), a handbook CTE, an assumed distribution.

Once each is expressed as a standard uncertainty they combine identically (RSS with sensitivity coefficients). The only place the A/B origin matters afterwards is degrees of freedom for Welch–Satterthwaite — Type B components are often treated as having large ν unless you doubt the limits.

Budget-writing tip: one line per physical effect, note evaluation type, distribution, divisor, and ν. Auditors read budgets by those columns.