https://html.spec.whatwg.org/#text-track-model

"Whenever a text track that is in a media element's list of text tracks has its text track mode change value, the user agent must run the following steps for the media element:"

This means that if a <track> element is first created and its text track's mode set to "showing" before appending it to the media element, these steps will not be run, and thus the "time marches on steps" steps which will cause any cues within to be shown will also not be run.

This illustrates the problem:

<video preload="auto">
  <source src="video.mp4" type="video/mp4">
  <source src="video.ogv" type="video/ogg">
</video>
<script>
var video = document.querySelector("video");

video.addEventListener("loadedmetadata", function() {
  video.currentTime = 1;
});

video.addEventListener("seeked", function() {
  var bug = true;
  var track;
  if (bug) {
    var trackElement = document.createElement("track");
    trackElement.kind = "subtitles";
    track = trackElement.track;
  } else {
    track = video.addTextTrack("subtitles");
  }
  track.addCue(new VTTCue(0, 10, "text"));
  track.mode = "showing";
  if (bug) {
    video.appendChild(trackElement);
  }
});
</script>

The main point being that it's unfortunate that the order of setting track.mode and appending trackElement should matter at all.

Of course, script-created tracks "should" be added with HTMLMediaElement.addTextTrack(), but it is still possible to add <track> elements and use their internal TextTracks instead.