diff --git a/corpus/commonmark-spec/exceptions.json b/corpus/commonmark-spec/exceptions.json
index 4e4af4d..9464e01 100644
--- a/corpus/commonmark-spec/exceptions.json
+++ b/corpus/commonmark-spec/exceptions.json
@@ -465,14 +465,14 @@
     "check": "count",
     "divergence": "em 1/3",
     "example": 638,
-    "kind": "mark-model",
-    "reason": "The mark spans adjacent text nodes, counted once per node where CommonMark nests one element."
+    "kind": "pending",
+    "reason": "The hard break carries the em mark, which ADF's schema forbids; todo.md item 87 strips it."
   },
   {
     "check": "count",
     "divergence": "em 1/3",
     "example": 639,
-    "kind": "mark-model",
-    "reason": "The mark spans adjacent text nodes, counted once per node where CommonMark nests one element."
+    "kind": "pending",
+    "reason": "The hard break carries the em mark, which ADF's schema forbids; todo.md item 87 strips it."
   }
 ]
diff --git a/todo.md b/todo.md
index c302574..ef25184 100644
--- a/todo.md
+++ b/todo.md
@@ -72,7 +72,7 @@
 `*foo  \nbar*` builds a `hardBreak` marked `em` (CommonMark examples 638 and 639), which Atlassian's
 schema refuses: its `hardBreak_node` holds no `marks` (`spec/adf-schema/full.json`). So
 `commonMarkToAdf` builds a document a site may reject from plain CommonMark. The suite's `em 1/3`
-exceptions for 638 and 639 go back to `em 1/2`.
+`pending` exceptions for 638 and 639 then return to `mark-model` and `em 1/2`.
 
 ### 88. Build a `blockTaskItem` holding only the paragraphs and extensions Atlassian's schema allows, in `portableMarkdownToAdf`.