cd - behavior: It undoes the last directory change irrespective of whether you did it with builtin cd or via sd. If you do "sd project-a" and then "cd -", you go back to where you were before. The sd function is a "thin" wrapper that ultimately executes "command cd <pathname>", so it respects the shell's native $OLDPWD logic perfectly.

pushd and "odds": In its current implementation, sd only records a visit when you specifically use the sd command (or its cd alias) to move. If you use pushd or popd directly, sd isn't invoked and doesn't "see" that movement. This keeps manual stack operations entirely separate from the automated ranking. To be clear: the shell managed pushd/popd stack is distinct from the dynamic score-sorted directory stack maintained/utilized by sd. Personally I simply do not have any more use for the former so would argue that sd not seeing cd actions initiated by push/popd is not an issue for the tool.

pushd replacement: sd doesn't replace pushd/popd/dirs. Think of pushd as a manual bookmarking system and sd as a temporal index. They coexist: pushd is for explicit, manual stack management (presuming you still want/need it in addition to sd), while sd is for "get me to this project's source" jumps using shorthand.

The design goal is to be a "well-behaved citizen" of the shell. It doesn't break standard invariants; it just adds a shorthand layer that requires zero manual maintenance.