Answer auto-reviewer round 1: the README says to retry the engine-limit refusal once on JavaScriptCore, and the merged-text test says why it stays cheap

README.md

@@ -410,7 +410,8 @@ typed on `ConvertError` takes both.
 A call reports the first refusal in document order and stops, so a document with several surfaces
 them one per call. Every refusal is deterministic on one engine — there is no I/O anywhere — so a
 retry returns the identical error: fix the input, or set the document aside. Only the engine-limit
-refusal under `unsupported-node-shape` differs between engines.
+refusal under `unsupported-node-shape` differs between engines; on JavaScriptCore (Bun, Safari),
+retry it once, since that engine reports a heap running short the same way.
 
 ### Parsing
 

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

@@ -70,6 +70,7 @@ test('refuses what the document guard refuses, and nothing else', () => {
   assert.match(portableDocument(document(paragraph(deepInline))), /^unsupported-nesting-depth at /)
 })
 
+// The merge throws at its concatenation, before any text is scanned, so no engine holds more than the one string.
 test('refuses a paragraph whose merged text passes the engine\'s longest string', () => {
   const long = 'a'.repeat(2 ** 28)
   const merged = paragraph(...Array.from({ length: 9 }, () => text(long)))