diff --git a/locators/evaluations_locators.py b/locators/evaluations_locators.py
index 132d8f6..e7b1671 100644
--- a/locators/evaluations_locators.py
+++ b/locators/evaluations_locators.py
@@ -125,6 +125,11 @@ class EvaluationsLocators:
     # while it is empty.
     MODAL_OBJECTIVE_TEXTAREA = "[role='dialog'] textarea"
     MODAL_STEP2_HEADING = "[role='dialog'] :text('I am evaluating as')"
+    # Step-2 "Evaluation Domain" select — only rendered when the selected
+    # model has a domain AND the backend returns non-empty audit domain
+    # options for it (ModelSelectionModal.tsx `requiresEvaluationDomain`).
+    # Added for test-sync/ParakhAI-frontend-pr441 (dropdown alphabetical sort).
+    MODAL_EVALUATION_DOMAIN_DROPDOWN = "select[name='evaluationDomain']"
 
     # Modal dropdown option lists — at least one <option> or listbox item must be present
     # NOTE: If dropdowns are custom (React-Select / Radix), add data-testid="model-option"
diff --git a/pages/new_evaluation_page.py b/pages/new_evaluation_page.py
index 462fd83..e24455c 100644
--- a/pages/new_evaluation_page.py
+++ b/pages/new_evaluation_page.py
@@ -171,6 +171,30 @@ def get_modal_version_options(self) -> list[str]:
         ).all()
         return [o.inner_text().strip() for o in opts]
 
+    def is_modal_evaluation_domain_dropdown_visible(self) -> bool:
+        """Return True if step 2's 'Evaluation Domain' select is rendered.
+
+        It only appears when the selected model has a domain and the backend
+        returns at least one audit domain option for it — absent for models
+        without one, which is a legitimate case, not a bug.
+        """
+        return self.is_visible(
+            EvaluationsLocators.MODAL_EVALUATION_DOMAIN_DROPDOWN, timeout=3_000
+        )
+
+    def get_modal_evaluation_domain_options(self) -> list[str]:
+        """Return the visible text of every option in step 2's 'Evaluation
+        Domain' select (excluding option index 0 if it's a disabled
+        placeholder with no value)."""
+        opts = self.page.locator(
+            f"{EvaluationsLocators.MODAL_EVALUATION_DOMAIN_DROPDOWN} option"
+        ).all()
+        return [
+            o.inner_text().strip()
+            for o in opts
+            if (o.get_attribute("value") or "").strip() != ""
+        ]
+
     def select_model_by_index(self, index: int = 1) -> None:
         """Select a model by option index. All options are real registered models —
         the 'New AI Model' ghost placeholder that used to occupy index 0 is gone
diff --git a/tests/e2e/test_new_evaluation_dropdown_sort.py b/tests/e2e/test_new_evaluation_dropdown_sort.py
new file mode 100644
index 0000000..1b18778
--- /dev/null
+++ b/tests/e2e/test_new_evaluation_dropdown_sort.py
@@ -0,0 +1,144 @@
+"""
+Pending coverage for CivicDataLab/ParakhAI-frontend#441
+("fix: sort AI model and evaluation domain dropdowns alphabetically").
+
+Open `dev` PR, no coverage yet (test-sync §9 sweep, 2026-10-08). Gated with
+`pending_pr` — skipped in CI until GitHub reports #441 merged.
+
+PR #441 head: acbf053512e5a2bfa7a87136d2105f74eb5e95c9 (as of 2026-10-08).
+If review drags on, re-check the head hasn't moved before this merges.
+
+What changed: the "Select AI Model" dropdown (New Evaluation modal, step 1)
+and the "Evaluation Domain" dropdown (step 2) used to render options in
+whatever order the API returned them. #441 sorts both by visible label with
+`localeCompare`.
+
+UNVERIFIED — not executed, no credentials in this cloud sandbox. Both
+dropdowns live behind the AI Maker dashboard (Keycloak login via
+`authenticated_page`), which this environment cannot reach. A human with
+TEST_EMAIL_1/TEST_PASSWORD_1 must run this locally to get the real
+red/green proof:
+
+    pytest tests/e2e/test_new_evaluation_dropdown_sort.py -v --tb=short
+
+Red/green proof to perform, since #441 isn't on dev yet:
+  1. Run against dev now -> expect FAIL (old code serves API order, which
+     is not guaranteed alphabetical -- fails whenever the live model/domain
+     list isn't already sorted; see the note on test_model_dropdown's skip
+     condition below for the one case this can't show anything).
+  2. `gh pr checkout 441` (frontend repo) and serve locally against the dev
+     API per SKILL.md section 6's local-serve recipe, point HOME_URL_DEV at
+     the local port -> expect PASS.
+  3. Restore pointing at dev, confirm the test is correctly gated
+     (`pending_pr ParakhAI-frontend#441: not merged yet` skip reason).
+"""
+
+import pytest
+from playwright.sync_api import Page
+
+from pages.new_evaluation_page import NewEvaluationPage
+
+pytestmark = [
+    pytest.mark.e2e,
+    pytest.mark.auth,
+    pytest.mark.pending_pr("ParakhAI-frontend#441"),
+    pytest.mark.timeout(180),
+]
+
+
+class TestDropdownsSortedAlphabetically:
+    """Both dropdowns must render options sorted by visible label, not API order."""
+
+    def test_model_dropdown_options_sorted_alphabetically(self, authenticated_page: Page):
+        """
+        'Select an AI Model' options (modal step 1) must be in ascending,
+        case-insensitive alphabetical order by label.
+
+        This only has teeth when the live model list has 2+ distinct labels
+        whose case-sensitive ('<') and case-insensitive order differ, or
+        whose API order isn't already alphabetical -- otherwise an
+        unsorted-but-coincidentally-alphabetical API response would pass
+        either way. Skip rather than pass vacuously when there's too little
+        data to distinguish the two.
+        """
+        nep = NewEvaluationPage(authenticated_page)
+        nep.go_to_evaluations_list()
+        nep.click_new_evaluation()
+
+        if not nep.is_modal_visible():
+            pytest.skip("New Evaluation modal did not open — cannot reach the dropdown")
+
+        options = nep.get_modal_model_options()
+        labels = [text for (_value, text) in options if text.strip()]
+
+        if len(labels) < 2:
+            nep.click_modal_cancel()
+            pytest.skip(
+                f"Only {len(labels)} model option(s) on dev — not enough data to "
+                "distinguish sorted from unsorted"
+            )
+
+        expected = sorted(labels, key=lambda s: (s.casefold(), s))
+        nep.click_modal_cancel()
+
+        assert labels == expected, (
+            "'Select an AI Model' options are not sorted alphabetically by "
+            f"label: got {labels!r}, expected {expected!r} (PR #441)"
+        )
+
+    def test_evaluation_domain_dropdown_options_sorted_alphabetically(
+        self, authenticated_page: Page
+    ):
+        """
+        'Evaluation Domain' options (modal step 2) must be sorted the same
+        way. The dropdown only renders for models that have a domain with
+        non-empty backend audit-domain options -- try a few models before
+        concluding none qualify, rather than skipping on the first miss.
+        """
+        nep = NewEvaluationPage(authenticated_page)
+        nep.go_to_evaluations_list()
+        nep.click_new_evaluation()
+
+        if not nep.is_modal_visible():
+            pytest.skip("New Evaluation modal did not open — cannot reach the dropdown")
+
+        options = nep.get_modal_model_options()
+        valid_indices = [
+            i
+            for i, (_value, text) in enumerate(options)
+            if text.strip() and text.strip() not in nep.EXCLUDED_MODEL_NAMES
+        ]
+        if not valid_indices:
+            nep.click_modal_cancel()
+            pytest.skip("No usable model options on dev — cannot select a model")
+
+        domain_labels: list[str] | None = None
+        for idx in valid_indices[:5]:
+            nep.select_model_by_index(idx)
+            nep.click_modal_next()
+            if nep.is_modal_evaluation_domain_dropdown_visible():
+                domain_labels = nep.get_modal_evaluation_domain_options()
+                break
+            nep.click_modal_back()
+
+        if domain_labels is None:
+            nep.click_modal_cancel()
+            pytest.skip(
+                "No model among the first candidates renders an 'Evaluation "
+                "Domain' dropdown on dev — cannot exercise PR #441's domain-sort change"
+            )
+
+        if len(domain_labels) < 2:
+            nep.click_modal_cancel()
+            pytest.skip(
+                f"Only {len(domain_labels)} domain option(s) on dev — not enough "
+                "data to distinguish sorted from unsorted"
+            )
+
+        expected = sorted(domain_labels, key=lambda s: (s.casefold(), s))
+        nep.click_modal_cancel()
+
+        assert domain_labels == expected, (
+            "'Evaluation Domain' options are not sorted alphabetically by "
+            f"label: got {domain_labels!r}, expected {expected!r} (PR #441)"
+        )