The traditional implementation (e.g., fig-Forth) has PAD at some distance from HERE: In traditional systems WORD copies the data to HERE, then there is the pictured numeric output buffer, and then there is PAD. The pictured numeric output buffer grows downwards from PAD-1. So the distance between HERE and PAD in those systems accomodates the size of the pictured numeric output buffer and maybe the strings scanned by WORD (if the system wants to support using <#...#> while parsing with WORD).
