Answer prose round 1: the spec names each title and destination, the writing verdict is its own decision citing item 100, items 99 and 101 state what holds, and the migration and changelog rows say what they replace

spec/flavour.md

@@ -310,12 +310,12 @@ the destination, `alt` the description's plain-text content when non-empty — a
 it contributing its own text, a node spelling its text in the content slot contributing that text,
 and a break of either kind a space. A non-empty title, `![alt](url "title")`, adds a `caption`
 holding the title as one text node, and a link whose entire text is the image,
-`[![alt](url)](href)`, marks the `media`. `adfToLosslessMarkdown` emits the image form only for the
-bare shape — those attrs and no others, no marks on the `mediaSingle`, no caption, and a
-`media` carrying nothing beyond `alt`, `type` and `url` and no mark but a link holding nothing
-beyond `href` and `title`, spelled `[![alt](url)](href "title")` — and only where CommonMark spells
-it: a destination, a title or a description the image form cannot hold, an empty `alt` included,
-takes the directive form instead.
+`[![alt](url)](href)`, marks the `media`. `adfToLosslessMarkdown` writes `![alt](url)` only for this
+shape: a `mediaSingle` with those attrs and no others, no marks and no caption, holding a `media`
+with no attrs beyond `alt`, `type` and `url`. The `media` may carry one mark, a link with no attrs
+beyond `href` and `title`, written `[![alt](url)](href "title")`. Where CommonMark cannot spell the
+image's alt text or URL, or the link's `href` or title, the directive form is written instead; an
+empty `alt` counts as one CommonMark cannot spell.
 
 `mediaInline` carries a media `collection` + `id`, never a URL, so anywhere else an image reads as
 its alt text linked to its destination, the link titled by the image's title: beside other