diff --git a/core/wren/src/wren/policy.py b/core/wren/src/wren/policy.py
index 4bf4158f70..d02e1a6f4a 100644
--- a/core/wren/src/wren/policy.py
+++ b/core/wren/src/wren/policy.py
@@ -161,6 +161,14 @@ def resolve_model_name(
     case-sensitively; an unquoted identifier prefers an exact case match but
     falls back to a case-insensitive scan. Returns ``None`` if no model
     matches.
+
+    When an unquoted reference matches more than one model case-insensitively
+    (e.g. both ``Users`` and ``users`` are defined), the one picked is the
+    lexicographically smallest rather than whichever ``model_names`` happens
+    to iterate first: for a ``set`` of strings that iteration order is not
+    the insertion order, it depends on string hashing and varies across
+    processes under ``PYTHONHASHSEED``, so the same query could silently
+    bind a different model on every run.
     """
     model_set = (
         model_names if isinstance(model_names, (set, frozenset)) else set(model_names)
@@ -170,10 +178,14 @@ def resolve_model_name(
     if quoted:
         return None
     name_lower = name.lower()
+    # Single pass, no intermediate list: a tie only needs `<` against the
+    # best candidate seen so far, same cost as the exact-match scan this
+    # replaces.
+    best: str | None = None
     for candidate in model_set:
-        if candidate.lower() == name_lower:
-            return candidate
-    return None
+        if candidate.lower() == name_lower and (best is None or candidate < best):
+            best = candidate
+    return best
 
 
 # Statement shapes that are read-only queries. ``SetOperation`` is the base class
diff --git a/core/wren/tests/unit/test_policy.py b/core/wren/tests/unit/test_policy.py
index 54cb2e9215..3fc9d59233 100644
--- a/core/wren/tests/unit/test_policy.py
+++ b/core/wren/tests/unit/test_policy.py
@@ -5,12 +5,16 @@
 
 from __future__ import annotations
 
+import os
+import subprocess
+import sys
+
 import pytest
 from sqlglot import exp, parse_one
 
 from wren.config import WrenConfig
 from wren.model.error import ErrorCode, WrenError
-from wren.policy import validate_planned_sql, validate_sql_policy
+from wren.policy import resolve_model_name, validate_planned_sql, validate_sql_policy
 
 pytestmark = pytest.mark.unit
 
@@ -669,3 +673,53 @@ def test_planned_sql_fails_open_when_unparseable():
     open where the input direction fails closed.
     """
     validate_planned_sql("SELCT a FRM orders", "postgres")
+
+
+# ── resolve_model_name, case-insensitive fallback ──────────────────────────
+
+
+def test_resolve_model_name_exact_match_wins():
+    assert resolve_model_name("orders", False, {"orders", "Orders"}) == "orders"
+
+
+def test_resolve_model_name_quoted_is_case_sensitive():
+    assert resolve_model_name("Orders", True, {"orders"}) is None
+
+
+def test_resolve_model_name_unquoted_case_insensitive_fallback():
+    assert resolve_model_name("ORDERS", False, {"orders"}) == "orders"
+
+
+def test_resolve_model_name_no_match_returns_none():
+    assert resolve_model_name("missing", False, {"orders"}) is None
+
+
+def test_resolve_model_name_case_collision_is_deterministic():
+    """When two model names differ only in case, an unquoted reference that
+    matches neither exactly must resolve the same way every time.
+
+    Before the fix this depended on ``set`` iteration order, which varies by
+    process under ``PYTHONHASHSEED``, not within one: a fixed hash seed makes
+    iteration order stable for the lifetime of a process, so a same-process
+    loop passes on the buggy code too, every time. The bug only shows up
+    across fresh interpreters, so the regression spawns one per
+    ``PYTHONHASHSEED`` instead, matching how this was reproduced originally
+    (8 fresh subprocesses split 6/2 between ``'Orders'`` and ``'orders'``).
+    Sorted order is an arbitrary tie-break (the two names are equally
+    "right", this pair may legitimately be two distinct models, see
+    test_case_sensitivity.py); the point is only that it no longer varies.
+    """
+    script = (
+        "from wren.policy import resolve_model_name; "
+        "print(resolve_model_name('ORDERS', False, {'Orders', 'orders'}))"
+    )
+    for seed in range(8):
+        env = {**os.environ, "PYTHONHASHSEED": str(seed)}
+        result = subprocess.run(
+            [sys.executable, "-c", script],
+            env=env,
+            capture_output=True,
+            text=True,
+            check=True,
+        )
+        assert result.stdout.strip() == "Orders"