Preventing Loss Of Data Because Of NLS_LANG Setting?

I have a database using character set AL32UTF8. The database contains character strings (VARCHAR2 colums) that may contain both Western European and Eastern European characters (and may be even other kinds of characters such as Cyrillic or Asian).

Suppose a client application has set NLS_LANG character set to WE8ISO8859P1. By this Western European characters will be shown correctly, while Eastern European characters which do not compare with WE8ISO8859P1 will be converted and shown as '?' (question marks) in the client application. If a user of this application fetches a record with Eastern European characters, modifies the record and then rewrites it to the database, the Eastern European characters with be rewritten to the database as question marks, i.e. Eastern European data have been corrupted.

I would like to prevent this by detecting that data were not converted properly during the fetch and then show the record to the user in read-only mode in order to avoid data loss, but I have not been able to detect the conversion error.

The application fetches data through the OCI interface using the "ofetch" function. The error code set by ofetch is the same (i.e. no error) regardless of whether the record contains Eastern European characters or not.

I thought I could manage this by setting the database parameter NLS_NCHAR_CONV_EXCP to TRUE, but this has no effect. Apparently this only deals with operations directly in the database.