All the URI rewriting can be controlled by options in the protocol.rewrite option tree.
In the case of sd, a URI rewriting mechanism called dumb prefixes is used. If protocol.rewrite.enable-dumb is true, ELinks looks at the contents of the protocol.rewrite.dumb.* tree, and if it finds option protocol.rewrite.dumb.sd, it will use its value as the target URI.
Therefore, more generally speaking, if ELinks looks at protocol.rewrite.dumb.<typed_string>, and if it exists, will replace the entered URI with it. These dumb prefixes can be used as a kind of alternative bookmark systen - if you want to have fast access to a certain site and you don't want to spend a while navigating the bookmarks manager, you can just fire up the Goto URL dialog, type the shortcut and there you go.
A dumb prefix can contain %c, which will be replaced by the URI of the current document. This is used, for example, in the arc dumb-prefix, which provides a shortcut to the Wayback machine at archive.org.
In the case of g, a slightly different URI rewriting mechanism called smart prefixes is used. If protocol.rewrite.enable-smart is true, ELinks looks at the contents of the protocol.rewrite.smart tree and, if it finds option protocol.rewrite.smart.g, it will use its value as the target URI.
The difference is that the smart prefixes can take arguments and reference them in the target URI. As with dumb prefixes, %c in the URI means the current address. Additionally, %s will be substituted by the whole string following the prefix (elinks in the g example above; %s is by far the most frequently used expansion), while you can reference individual space-separated arguments with the codes %0 thru %9. Finally, if you want a plain % to the resulting URI, use %%.
Note that the prefix searched in the protocol.rewrite.smart tree is everything in the typed address up to the first space or colon (so bug:123 and bug 123 are identical). These prefixes are generally useful for searching anything very fast - be it Google (g:foo, gi:foo, gr:foo, …), ELinks Bugzilla bugs (bug:1234), the RFC database (cr:foo), or the Czech-English dictionary (czen:foo). The only limit is your imagination.