diff --git a/datajunction-server/datajunction_server/sql/decompose.py b/datajunction-server/datajunction_server/sql/decompose.py
index fc4236e8d..b95672d3a 100644
--- a/datajunction-server/datajunction_server/sql/decompose.py
+++ b/datajunction-server/datajunction_server/sql/decompose.py
@@ -1156,12 +1156,13 @@ def _expand_template(self, template: str, args: list) -> str:
     def _short_hash(self, expression: str, query_ast: ast.Query) -> str:
         """Generate a short hash for the given expression.
 
-        ``str(query_ast.select.from_)`` is whitespace-normalized so cosmetic
-        changes to AST string rendering (pretty-printing) don't perturb hashes
-        and rename every derived measure. ``expression`` is passed through
-        unchanged to preserve hash compatibility for expressions that already
-        contain newlines (e.g. CASE, IF).
+        Both the expression and the FROM clause are whitespace-normalized so
+        cosmetic changes to AST string rendering (pretty-printing, indentation,
+        newlines around AND/OR conjuncts) don't perturb hashes and silently
+        rename every derived measure. Component names are durable identifiers
+        for materialized columns, so the hash must be invariant to formatting.
         """
+        expression_normalized = " ".join(expression.split())
         from_str = " ".join(str(query_ast.select.from_).split())
-        signature = expression + from_str
+        signature = expression_normalized + from_str
         return hashlib.md5(signature.encode("utf-8")).hexdigest()[:8]
diff --git a/datajunction-server/tests/sql/decompose_test.py b/datajunction-server/tests/sql/decompose_test.py
index b4cf1ff4d..f8633316d 100644
--- a/datajunction-server/tests/sql/decompose_test.py
+++ b/datajunction-server/tests/sql/decompose_test.py
@@ -795,6 +795,33 @@ async def test_unsupported_aggregation_function(session: AsyncSession, create_me
     )
 
 
+@pytest.mark.asyncio
+async def test_count_if_hash_invariant_to_formatting(
+    session: AsyncSession,
+    create_metric,
+):
+    """
+    Component hashes must be stable across cosmetic AST rendering changes.
+
+    A previous regression (pretty-printing PR) caused ``str(arg)`` of a
+    BinaryOp containing AND/OR to emit newlines between conjuncts, which
+    silently renamed every ``count_if`` measure with a multi-conjunct
+    condition. Component names are durable identifiers for materialized
+    columns, so the same metric query must always produce the same hash —
+    even if the same logical condition is parsed from differently-formatted
+    SQL text.
+    """
+    single_line = await create_metric(
+        "SELECT COUNT_IF(a = 1 AND b = 2 AND c = 3) FROM parent_node",
+    )
+    multi_line = await create_metric(
+        "SELECT COUNT_IF(\n  a = 1\n  AND b = 2\n  AND c = 3\n) FROM parent_node",
+    )
+    one, _ = await MetricComponentExtractor(single_line.id).extract(session)
+    two, _ = await MetricComponentExtractor(multi_line.id).extract(session)
+    assert [c.name for c in one] == [c.name for c in two]
+
+
 @pytest.mark.asyncio
 async def test_count_if(session: AsyncSession, create_metric):
     """