Compare performance only when wear conditions are reasonably similar. Record projection and persistence in separate fields, then reject a ranking if sessions differ too much in application, parent resource environment, or timing. Context mismatch should remain a limitation, not be hidden.