Answer principles round 3: 0 🔴, 1 🟡 (0 misread); item 154 says what the pairing reads off the built nodes, and the fallback test spells its runs

src/markdown/portable/adf-to-portable-markdown.test.ts

@@ -397,7 +397,7 @@ test('drops the mark of the run that cannot close where it sits, and keeps the l
 test('gives up every highlight run whose own delimiter cannot flank in one fallback, so a line of them re-emits once however many it holds', () => {
   const highlight: AdfMark = { type: 'backgroundColor' }
   const line = [text('a '), text('x', highlight), text('a '), text('x', highlight), text('a '), text('x', highlight), text('a')]
-  const runs = [1, 3, 5].map((index) => ({ depth: 0, first: index, last: index }))
+  const runs = [{ depth: 0, first: 1, last: 1 }, { depth: 0, first: 3, last: 3 }, { depth: 0, first: 5, last: 5 }]
   assert.deepEqual(portableLineFallback(line, 'paragraph', []), { ok: true, value: { kind: 'unspellable-run', runs } })
 })
 

todo.md

@@ -218,8 +218,8 @@ and hostile. Technical principle: for pure logic over many inputs, prefer a prop
 
 A comprehension seat found `src/markdown/inline-content.ts` `resolveNodes` / `pairedHighlights`
 hard: an empty text node stands in for each `==` until pairing, then a cursor walk overwrites it.
-The pieces now carry their offsets, so the pairing can run on them first. Make that place easier to
-read.
+The pairing reads each `==`'s position, stretch and marks off the built nodes; give the pieces what
+it reads, so it can run on them first. Make that place easier to read.
 
 ### 155. Have `closeBracket` return the link, image or literal bracket it builds, and apply it to the scan in one caller.