Answer prose round 1: one sentence per idea for the image readings, a decision title that states the rule, item 86 names its shapes

src/markdown/parse/inline-content.ts

@@ -22,7 +22,7 @@ import { readDirectiveMark } from './directive-marks.ts'
 import { readInlineDirectiveNode } from './directive-nodes.ts'
 import { readTextDirective } from '../text-directive.ts'
 
-// `image` is the `mediaSingle` a lone image builds, for the one container that holds it: a plain paragraph. `nodes` reads it as its linked alt text.
+// Only a `paragraph` block takes `image`, the `mediaSingle` a lone image builds; every other container takes `nodes`, where the image is its linked alt text.
 export type InlineContent = { image?: AdfNode; nodes: AdfNode[] }
 
 // The text break leaf: it holds no marks, and stands between the nodes it parts until the outermost content checks and drops it.

src/markdown/parse/markdown-to-adf.ts

@@ -318,7 +318,7 @@ function codeBlockNode(language: string, text: string, reading: Reading, path: C
   return success(text === '' ? node : { ...node, content: [{ text, type: 'text' }] })
 }
 
-// spec/flavour.md, The CommonMark image: only a plain paragraph gives a lone image the block it needs.
+// spec/flavour.md, The CommonMark image: only a `paragraph` block gives a lone image the block it needs.
 function paragraphNode(text: string, reading: Reading, path: ConvertErrorPath): Result<AdfNode> {
   const content = parseInlineContent(text, reading.definitions, path, 'paragraph', reading.claims)
   if (!content.ok) return content