Data Conversion: Doing it Right the First Time - High-Performance Web Databases: Design, Development, and Deployment - page 596

made to correct the data, it probably will have to be reentered. Again, in
some cases, additional programming can help remedy the problems.
Programming for Data Correction
Some simple automated routines can make the job of data correction
much easier. If they require no manual intervention, it could be advanta-
geous to simply put them into the main conversion program. However, the
program may require that a user make the decision.
If the existing data entry programs are not adequate for large-scale data
correction efforts, some additional programs might have to be written for
error repair. For example, the existing system may not allow the display of
records with a referential integrity problem, which are probably the very
records that need correction. Custom programming will be required to
make the change.
SPECIFYING THE MAPPING
Often, crucial information needed for the conversion will be missing. If
the old system can accommodate the missing information, it may be a mat-
ter of keying it in from original paper records. However, the original infor-
mation may not be available anymore, or it may never have been collected.
In that case, it may be necessary to put in special markers to show that the
information is not available.
Model Mismatches
It can be difficult to go from an non-normalized structure to a normalized
structure because of the potential for problems in mapping from old to
new. Many problems are the result of inconsistent and redundant data, a
poor key structure, or missing information. If there is a normalized struc-
ture in the old system, there probably will not be as many difficulties. Other
problems result from changed assumptions about the cardinality of rela-
tionships or actual changes in the business rules.
Discovered Requirements
The requirements of a system almost never are understood fully by the
user or the developer prior to construction of the system. Some of the data
requirements do not become clear until the test conversions are being run.
At that point, it may be necessary to go back and revisit the whole devel-
opment effort. Standard change and scope control techniques apply.
Existing Documentation
Data requirements are rarely right the first time because the initial doc-
umentation is seldom correct. There may be abandoned fields, mystery