Unfortunately the technique of using SP_REPL_ADDCOLUMN and SP_REPL_DROPCOLUMN with the dummy columns would allow data loss if you didn't take the system down. If you don't take the system down, and data is changed in any way from the time the "dummy" column is created, to the time that the new correct column is added, that data would be lost.

This could be avoided with a new snapshot, but there are many cases where this is not appropriate - such as Large Databases, slow networks, or purged systems.

How do others work around this without causing data loss?