The editor stage always used the same fixed 16:9 box regardless of the recording's actual shape, so a portrait recording was letterboxed down to a small strip in the middle of a wide landscape frame. Overlay positions were always calculated against the video's real rendered rect rather than the container, so nothing was actually misplaced - it was just awkward to work with.
The stage now reads the recording's real dimensions once its metadata loads and sizes itself to match, so a portrait recording gets a portrait editing frame (capped and centered like the record page's own portrait preview) instead of being squeezed into a landscape one.