A text error catches the eye while reading; a number error does not. When a sum, a date or an account number is carried over wrongly, the sentence still reads flawlessly. That is why numeric fields are separated from the text read and verified by a method of their own.
Why a separate step
Numeric errors cannot be inferred from context, because the sentence looks right with either value. Reading the text cannot catch them. A separate step is mandatory.
The list method
All numeric fields are gathered in a separate list and compared with the original. Checking by eye inside the text is not enough. The method is systematic.
What is covered
Number, date, time, sum, rate, reference number and unit fields are in scope. The list is defined at the outset. None is skipped.
Date fields
The order of day and month and the format of the year are checked. Silent slips are most frequent in these fields. Attention is concentrated here.
Decimal and thousands separators
Separator conversions are verified separately. A wrong conversion changes the value a thousandfold. The check is meticulous.
Unit fields
It is checked that units are carried over correctly with their values. A missing unit is more common than a numeric error. The field is looked at separately.
Currency codes
The unit codes next to sums are verified. The right figure with the wrong unit is meaningless. The check is made.
Identifying numbers
Tax, registry, account and document numbers are compared character by character. A one-digit deviation makes the document unusable. Meticulous care is needed.
Double check
Payment and identity fields are compared by two separate people. The method is simple but the most effective. It is standard in corporate jobs.
Counting table rows
The number of rows and columns in tables is compared with the original. A lost row is not noticed by eye. A count is made.
Totals are not recalculated
Totals in the document are not recalculated and are carried over as they stand. A total that does not add up is reported as an observation. No correction is made.
Leading zeros
It is checked that leading zeros in numbers are preserved. Tools delete these zeros silently. The fields are kept as text.
Automatic format conversions
Conversions made by spreadsheet tools are looked for. These conversions produce no warning. The check is made deliberately.
Numbers inside images
Values inside charts and diagrams are checked too. These fields do not appear in the automatic list. They are added by hand.
Errors from character recognition
In scanned documents, digits may have been recognised wrongly. Values are verified against the image. The text layer is not trusted.
Recording the errors found
The deviations identified are recorded and their source is examined. A recurring error type points to a process problem. The record gains value.
Timing of the check
This check is made after the text is final. A check made early is invalidated by changes. The sequence is observed.
The next step
When the numeric fields are verified, terminology compliance is scanned with a tool. The two checks target different risks. The sequence is kept.