ZegnatIt is just a very hard problem. textContent is what DOM gives us, but people find the results unexpected. But that leaves us with either emulating the browser’s innerText (no clear algo for non-CSS implementers) or having our own living algo that changes when more user examples come in :(
