Each record begins with an integer code indicating what kind of record it is. This record ID can also be used as an ordering index. Currently, the order of records in a dataset is fixed, and this ordering function is somewhat redundant.
Some record types and record groups can be repeated (i.e., DESCRIP1, DESCRIP3, COMMENT, BADVAL, Procgroup, Auxgroup, and Packgroup); these records all include a RECSORT field which is used to associate related records within a group and to order different records (or record groups) which are of the same type. Except for BADVAL and DIMSPEC1 records (whose multiple records are not allowed to overlap), this ordering may be very important: where conflicts arise in regions where two records' domains overlap, the record with the higher-numbered index takes precedence.
For example, the COMMENT records are ordered by their indices such that, even if they are written to the dataset in a random order, a reader program can know how to put them back in order, so that the sentences, tabular values, or other such information in the comments makes sense.
The RECSORT fields in the Procgroup and Packgroup records specify the order in which subsets of the data were processed or packed. (These subsets are specified by other fields, START and END, which are described in Section 2.3.2). The Auxgroup RECSORT fields are mainly a convenience to distinguish one record from another within an Auxgroup. The information contained in the Procgroup and Auxgroup records is usually not needed to read the data and could be ignored; this information is often, however, needed to use the data wisely. The Packgroup records, on the other hand, contain information which is critical to reading the data, and their ordering must be specified with care.
The DESCRIP1 and DESCRIP3 records use the RECSORT field to distinguish between descriptors of different subsets the same dimension index. The order of the DESCRIP1 records is not important since there can be no overlap in specification. The order of the DESCRIP3 records is important and indicates the order in which the data were averaged.