The Migration Completed Successfully. Half the Names Are Now Question Marks.

Character-set data loss is the migration failure that passes every check: the load raises no error, the row counts match, and characters the target set can't represent are silently stripped to plain ASCII or replaced with '?'. What actually happens (narrowing loss vs the NLS_LANG pass-through trap vs byte expansion), why row counts never catch it, the round-trip scan that does (DMU/csscan), and a lab that reproduces the loss on Oracle Database Free then proves a Unicode target is clean.