In framesets, parent-child relations were absolutely defined and stable, as were the paths between individual frames.
(Current frame is "self" or "window", parent frame or frameset is "parent", and the top most entry point into the hierarchy "top". Moreover, "self", transcending the window context, is also the only reliable reference to the global object, thus also providing a valid reference to the context of a worker. Specifically, it was for framesets that the notion of hierarchy was introduced, which eventually resulted in the concept of the DOM. Some inconsistencies to this concept of strict parent-child relations were actually introduced by early implementations of the iframe-element, which is, BTW, still a valid HTML element.)
That said, there was a small inconsistency with an early subversion of Netscape 3, regarding, whether the frame source would be relative to any current location of the frame or rather relative to the frameset. (But this was an issue for a rather short period of time, two months or so.) A major difference in styling was the implementation of frame borders, if they would be entirely invisible by just specifying `border="0"` (Netscape and others) or, if they required the two attributes `frameborder="0"` and `framespacing="0"` (MS IE). In practice, next to all sites specified both schemes. And jet another, but minor implementation specific detail was the sizing of framesets: While Netscape Navigator supported, like all other browsers, a size specified in pixels, this was internally translated to percents of the total width. Therefor, depending on rounding to integers, the presentation in the Netscape browser could be off by a pixel or two.
(The latter was, in deed, not unusual behavior at the time, just like MS Word and RTF used to translate any measurements internally to "tips" or twentieths of a point.)