Add some introductory paragraphs to the Data model section.

README.md

@@ -22,3 +22,7 @@ To generate a WD snapshot, run e.g.:
     $ ./snapshot.sh WD 2016-01-01 2015-12-08
 
 Also see https://github.com/w3c/webvtt/commit/754f13e3cf03d6036c3e4628c6920d17b412f778 for manual fixup of the generated output.
+
+To format the index.bs file, run:
+
+  $ ./format.py index.bs

index.bs

@@ -778,6 +778,18 @@ precomposed character).</p>
 <h2 id=data-model>Data model</h2>
 <!-- Describe metadata, caption/subtitle, chapter & description cues -->
 
+<p class=note>The box model of WebVTT consists of three key elements: the video viewport, cues, and
+regions. The video viewport is the rendering area into which cues and regions are rendered. Cues are
+boxes consisting of a set of cue lines. Regions are subareas of the video viewport that are used to
+group cues together. Cues are positioned either inside the video viewport directly or inside a
+region, which is positioned inside the video viewport.</p>
+
+<p class=note>The position of a cue inside the video viewport is defined by a set of cue settings.
+The position of a region inside the video viewport is defined by a set of region settings. Cues that
+are inside regions can only use a limited set of their cue settings. Specifically, if the cue has a
+"vertical", "line" or "size" setting, the cue drops out of the region. Otherwise, the cue's width is
+calculated to be relative to the region width rather than the viewport. </p>
+
 <h3 id=model-overview>Overview</h3>
 
 <p><i>This section is non-normative.</i></p>
@@ -1324,6 +1336,10 @@ together, which is particularly important when scrolling up.</p>
  <dd>
   <p>A number giving the number of lines of the box within which the text of each line of the
   containing cues is to be rendered. Defaults to 3.</p>
+
+  <p class="note">Since a WebVTT region defines a fixed rendering area, a cue that has more lines
+  than the region allows will be clipped. For scrolling regions, the clipping happens at the top,
+  for non-scrolling regions it happens at the bottom.</p>
  </dd>
 
  <dt><dfn lt="WebVTT region anchor">A region anchor point</dfn></dt>
@@ -2988,6 +3004,10 @@ run the following steps:</p>
        "<code>lr</code>", then let |cue|'s <a>WebVTT cue writing direction</a> be <a lt="WebVTT cue
        vertical growing right writing direction">vertical growing right</a>.</p></li>
 
+       <li><p>If |cue|'s <a>WebVTT cue writing direction</a> is not <a lt="WebVTT cue horizontal
+       writing direction">horizontal</a>, let |cue|'s <a>WebVTT cue region</a> be null (there are no
+       vertical regions).</p></li>
+
       </ol>
 
      </dd>
@@ -3063,6 +3083,10 @@ run the following steps:</p>
        <li><p>If the last character in |linepos| is a U+0025 PERCENT SIGN character (%), then let
        |cue|'s <a>WebVTT cue snap-to-lines flag</a> be false. Otherwise, let it be true.</p></li>
 
+       <li><p>If |cue|'s <a>WebVTT cue line</a> is not <a lt="WebVTT cue line automatic">auto</a>,
+       let |cue|'s <a>WebVTT cue region</a> be null (the cue has been explicitly positioned with a
+       line offset and thus drops out of the region).</p></li>
+
       </ol>
 
      </dd>
@@ -3117,6 +3141,9 @@ run the following steps:</p>
 
        <li><p>Let |cue|'s <a>WebVTT cue size</a> be |number|.</p></li>
 
+       <li><p>If |cue|'s <a>WebVTT cue size</a> is not 100, let |cue|'s <a>WebVTT cue region</a> be
+       null (the cue has been explicitly sized and thus drops out of the region).</p></li>
+
       </ol>
 
      </dd>