diff --git a/src/Memoizer.ts b/src/Memoizer.ts
index 0ace2982418b420bd56fdc07195cbf9b1f5a687e..81c2adfc023760a02149c449b4080a0273da37e0 100644
--- a/src/Memoizer.ts
+++ b/src/Memoizer.ts
@@ -100,6 +100,9 @@ export class Memoizer {
     // much effort), but any extra re-renders get eliminated with AST diffing.
     const processor = UNWRAP(this.#astProcessor);
     const parsed = this.#parsed;
+    const replacedTailAffectedDocument = parsed
+      .slice(blockStart)
+      .some((block) => block.some(nodeAffectsDocument));
     let parsedTree = processor.parse(content.slice(parseOffset));
     let newPositions = parsedTree.children.map(
       (child) => UNWRAP(UNWRAP(child.position).start.offset) + parseOffset,
@@ -118,7 +121,7 @@ export class Memoizer {
     // If a reflink is involved, fall back to a full reparse. Rehydrating a
     // synthetic root from cached blocks can lose source offsets on transformed
     // top-level nodes, which would scramble block bucketing and key identity.
-    if (nodeAffectsDocument(parsedTree)) {
+    if (replacedTailAffectedDocument || nodeAffectsDocument(parsedTree)) {
       blockStart = 0;
       parseOffset = 0;
       parsedTree = processor.parse(content);
diff --git a/tests/Memoizer.test.tsx b/tests/Memoizer.test.tsx
index 0e3df355c15afa3d828131e6aa77ff685d56a532..f34696681e9f4e1329d2a4240ffa9d63bf2bd721 100644
--- a/tests/Memoizer.test.tsx
+++ b/tests/Memoizer.test.tsx
@@ -140,6 +140,20 @@ describe("Memoizer incremental rendering", () => {
     expect(after).toContain('href="/two"');
   });
 
+  it("keeps earlier gfm footnotes rendered when a later footnote definition is removed", () => {
+    const incremental = createMemoizer(gfmProcessor);
+
+    incremental.update("a[^b]\n\n[^b]: bee\n\nc");
+    incremental.update("a[^b]\n\n[^b]: bee\n\nc[^d]\n\n[^d]: dee");
+    const after = renderNodes(incremental.update("a[^b]\n\n[^b]: bee\n\nc[^d]"));
+
+    const fresh = createMemoizer(gfmProcessor);
+    const expected = renderNodes(fresh.update("a[^b]\n\n[^b]: bee\n\nc[^d]"));
+
+    expect(after).toBe(expected);
+    expect(after).toContain('id="user-content-fn-b"');
+  });
+
   it.each([
     {
       content: ["alpha\n\nbeta", "alpha\nbeta", "alpha\n\nbeta\n\ngamma", "alpha\n\nbeta"],