From ac3465c656d1cb5bb2503d60d6e8ef5d1cb3813e Mon Sep 17 00:00:00 2001
From: Ajay Krishnan <ajay@krishnan.ca>
Date: Sat, 25 Jul 2026 09:00:59 -0700
Subject: [PATCH] Fix ID collisions for repeated section titles

Large llms-full.txt feeds repeat section headings, which made document
identity hashes collide within one source and abort indexing on the
documents primary key. Include each document's ordinal in the identity.
---
 docker/docs/context_docs/refresh.py |  6 ++++--
 docker/docs/tests/test_refresh.py   | 18 ++++++++++++++++++
 2 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/docker/docs/context_docs/refresh.py b/docker/docs/context_docs/refresh.py
index 92259d6..0fe46df 100644
--- a/docker/docs/context_docs/refresh.py
+++ b/docker/docs/context_docs/refresh.py
@@ -65,10 +65,12 @@ class RefreshCoordinator:
             vectors = await self.embedder.encode_documents(texts) if texts else []
             documents: list[PreparedDocument] = []
             source_host = (urlparse(response.resolved_url).hostname or "").lower()
-            for parsed_document, vector in zip(parsed.documents, vectors, strict=True):
+            for ordinal, (parsed_document, vector) in enumerate(zip(parsed.documents, vectors, strict=True)):
                 content_hash = hashlib.sha256(parsed_document.content.encode()).hexdigest()
+                # Include the ordinal so repeated section titles (common in large
+                # llms-full.txt feeds) cannot collide on the primary key.
                 identity = "\0".join(
-                    [source, parsed_document.canonical_url, parsed_document.heading_path, str(parsed_document.chunk_index)]
+                    [source, str(ordinal), parsed_document.canonical_url, parsed_document.heading_path, str(parsed_document.chunk_index)]
                 )
                 documents.append(
                     PreparedDocument(
diff --git a/docker/docs/tests/test_refresh.py b/docker/docs/tests/test_refresh.py
index fbffd2c..c1a6c3b 100644
--- a/docker/docs/tests/test_refresh.py
+++ b/docker/docs/tests/test_refresh.py
@@ -91,6 +91,24 @@ class RefreshTest(unittest.IsolatedAsyncioTestCase):
         self.assertEqual(1, self.store.list_sources()[0].doc_count)
         self.assertTrue(self.store.lexical_search("Original", limit=5))
 
+    async def test_repeated_section_titles_index_without_id_collisions(self) -> None:
+        body = "# Basic syntax\n\nFirst variant.\n\n# Basic syntax\n\nSecond variant.\n"
+        fetcher = FakeFetcher([FakeFetch(200, body)])
+        coordinator = RefreshCoordinator(
+            store=self.store,
+            fetcher=fetcher,
+            embedder=FakeEmbedder(),
+            parser=parse_llms_text,
+            ttl_seconds=3600,
+            now=lambda: 100.0,
+        )
+
+        outcome = await coordinator.refresh(self.source, force=True)
+
+        self.assertEqual("updated", outcome.status)
+        self.assertEqual(2, outcome.document_count)
+        self.assertEqual(2, self.store.list_sources()[0].doc_count)
+
     async def test_empty_success_response_preserves_previous_content(self) -> None:
         fetcher = FakeFetcher(
             [