The format uses a formatted, direct access file to contain the information. The header file contains exactly the same information as the Version 2 binary database. The file is created by writing the information to a character buffer using a 1PE12.5 format for reals and a format just large enough to hold the integers. The number of digits stored for a real number can be set through the inform file. The raising of this value will increase the accuracy of the representation of the real value but will also increase the file size. The information written to the is zipped using the following rules. Each bit of the first byte of a section is used to indicate whether the next 8 packets represent a literal or a repeat. A 0 is stored for a literal, a 1 for a repeat. In a literal packet the character is written. In a repeat packet the next 12 bits contain how far back in the previous entries is the start of the matching string (this limits the search for a matching string to the last 4096 characters). If the length of the matching string is 17 or less the length minus 2 is stored in the next 4 bits. If the length of the matching string is greater than 17 then the next 4 bits are set to 0 and the next bytes is set to the length of the matching string minus 18 (this limits the length of the matching string to 255+17 characters). Repeat strings must be of length greater than or equal to 3 any match shorter than this is ignored.