User Input Verification?
CSC 205/206
Professor Stearns

User Input Verification is complex for two major reasons:

1. User differences
   Some users are invested in the input; they are entering their own data and care that it is done correctly. Other users are minimum-wage data entry workers who can't wait to go home. The latter type of users have a high error rate (20% is an oft-quoted number)

2. Range checks may be impossible
   There is no easy way to distinguish between a correctly entered grade (e.g. 84) and an incorrect grade (e.g. 81).

There are some basic principles that will increase the accuracy of user-entered data. It is rarely (never?) acceptable to assume the user will enter valid data.

1. Enter data twice and cross check them.
   This is standard industry practice for uninvested users; make this a requirement if you are specifying a system with such users.

2. Do type checks
   If a number is input, verify that it is a valid number.
   Specify valid numbers in the data dictionary.
   Fundamental rule: read all input as a String and convert it to the proper type. All class libraries provide conversion and error-checking code; don't do it yourself! (e.g. see the parseInt method in Java.lang.Integer).

3. Do range checks
   This is obvious but often ignored. Be sure to specify ranges in the data dictionary. Don't forget the 8-day week story.

4. Get human feedback
   Provide a feedback mechanism for data validation. For example, you can verify a mailing list by mailing something and asking recipients to check their data.

5. Confirmation Box
   Ask the user to confirm the input data. Although commonly used, such a confirmation may have little effect on accuracy because users tend to press OK spontaneously.

6. Verify computed result
   Use the data to compute some known result; then validate the correctness and range of the result. If the computed result is bad, it is likely due to bad input.

7. Data cleaning
   If there is a large batch input file (e.g. bank check records), it may be desirable to "clean" the input data by removing suspect records for human inspection. The purpose is to keep bad records from entering the system at all.