The compilation of 1604 did not begin from nothing. Collections of the Gurus' compositions were already circulating, and the making of the volume was as much an act of selection and authentication as of assembly.
Two things were at stake. One was completeness: the compositions of four predecessors had to be gathered from wherever they were held, and manuscript scholarship has identified early collections that appear to have been available to the compilers. The other was exclusion. Compositions were circulating under the Gurus' names that the compilers did not accept, and the volume as fixed is a judgement about which were genuine.
The Granth records that judgement in its own way. Its arrangement is so regular, and its headings so systematic, that later additions are difficult to make without leaving a trace, and the second stage of the canon, when the ninth Guru's compositions were inserted, is visible precisely because the system is tight.
Manuscript evidence for the earlier stages is a live field rather than a settled one. Early manuscripts survive, their relationships to one another and to the 1604 compilation are argued over, and the arguments are not merely academic, since the standing of particular manuscripts touches on the authority of the text for a living community. The atlas notes the field is active and does not adjudicate it.