QlikView and Qlik Sense uses the same underlying Qlik Indexing (QIX) engine, which stores all decimal values as floating point values, based on IEEE 754 double-precision floating point standard. Floating point value is a common way to handle decimal values in computers, because it allows for very efficient processing of decimal values and thereby gives good calculation performance.
The principle is to store the value as an integer part, called mantissa, and then define the decimal position by an exponent.
A floating point value is always stored with a static length. 64-bit environments allows for a 17 digit mantissa, which means that a value like 1.75 is stored as 1.7500000000000000.
Floating point values are by definition not exact values and the stored value can vary on the last decimal digits. For example the value 1.75 can be rounded upwards or downwards, but still represent the same decimal value.
The variation on the last decimal digit can occur at any time when the value is being processed. This also includes when a value is loaded and stored, as well as when a calculation or a modification of the value is performed.
The precision of all numerical values in QIX engine is 14 digits. This means that the last three (17-14=3) digits of the decimal value are subject for rounding, and the last three digits can therefore fluctuate during processing. For this reason decimal values should only be presented with max 14 digits, including integer and decimal part.
Num() function can be used to format a loaded value to present within the 14 digit precision. It is possible to present more than 14 digits by statically expanding the number of decimal digits, which means all 17 stored digit can be revealed. It is not recommended to show more than 14 digits to the end-user as this information can be misleading.
NOTE1: The leading zero in decimal values smaller than 1 is not included in the 14 digit precision.
NOTE2: Values smaller than 100 have more than 14 digits in the decimal part and there for these values can not be completely revealed.
References to more details about floating point values;
IEEE 754 Standard; http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=4610933