diff --git a/AGENTS.md b/AGENTS.md
index 93056359c..64b03ef7e 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -168,7 +168,7 @@ Exception: when a backend caps page size below `DEFAULT_LIMIT` (1000), set `limi
   - Shared `seeded_*` fixtures are read-only; update/delete tests create private entities and
     clean up in `try/finally`.
   - Search assertions must be scoped to `seed_prefix`, filtered to the fixture's ids, and
-    wrapped in `poll_until` (`tests/integration/utils/wait.py`); `text`/`name` params are
+    wrapped in `poll_until` (`tests/utils/wait.py`); `text`/`name` params are
     fuzzy full-text queries, and other workers delete their seeds mid-run.
   - No exact-count asserts on unscoped queries.
   - Seed helpers live in `tests/integration/seeding.py`; new seed entities are appended
diff --git a/tests/integration/TESTING.md b/tests/integration/TESTING.md
index d57423536..21b7212f9 100644
--- a/tests/integration/TESTING.md
+++ b/tests/integration/TESTING.md
@@ -90,10 +90,10 @@ For any search-based assertion:
 
 1. Scope the query with `text=`/`name=` set to `seed_prefix` (reduces noise).
 2. **Filter results to the ids owned by the fixture** (correctness).
-3. Wrap the fetch in `poll_until` from `tests/integration/utils/wait.py` (index lag).
+3. Wrap the fetch in `poll_until` from `tests/utils/wait.py` (index lag).
 
 ```python
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 
 def test_hydrate_project(client: Albert, seed_prefix: str, seeded_projects: list[Project]):
@@ -108,6 +108,29 @@ def test_hydrate_project(client: Albert, seed_prefix: str, seeded_projects: list
     assert projects, "Expected at least one project in search results"
 ```
 
+`poll_until` stops at the first **non-empty** result by default. Our NoSQL store plus
+search index is eventually consistent, so freshly seeded items become visible one by one
+and a partial page is normal during indexing. One rule decides how to poll: **whatever
+stops the polling must imply the assertion that follows.** When asserting on a specific
+item, filter the fetched page to that item's id, so a non-empty result means the item is
+visible. When the assertion needs the **complete** expected set (exact-equality or
+exact-count asserts against your fixture's ids), pass a `predicate` so polling continues
+until the set is whole; without it the test flakes the moment one item indexes before
+the rest:
+
+```python
+expected_ids = {f"INV{lot.inventory_id}" for lot in seeded_lots if ...}
+results = poll_until(
+    lambda: search_scoped(...),
+    predicate=lambda results: {f"INV{p.id}" for p in results} == expected_ids,
+)
+assert {f"INV{p.id}" for p in results} == expected_ids
+```
+
+Even with a predicate, a rare timeout is expected behavior of an eventually consistent
+index. A timeout usually means indexing lag, so re-run once. A repeat failure means the
+predicate or the code under test is wrong; investigate it like any other failure.
+
 Also:
 
 - Never assert **exact counts** of unscoped `search()`/`get_all()` results; other workers
@@ -156,7 +179,7 @@ group's files with `-n 4` to catch cross-worker races.
 
 - [ ] `pytestmark = pytest.mark.xdist_group("...")` chosen per Rule 1 (or deliberately unmarked because only `client` is used)
 - [ ] No mutation of `seeded_*` entities; private entities cleaned up in `try/finally`
-- [ ] Search assertions scoped, id-filtered, and wrapped in `poll_until`
+- [ ] Search assertions scoped, id-filtered, and wrapped in `poll_until` (with a `predicate` when asserting a complete expected set)
 - [ ] No exact-count asserts on global queries
 - [ ] Names of created entities include `seed_prefix` or `TEST - <uuid>`
 - [ ] Ran the file plus its group with `-n 4` locally
diff --git a/tests/integration/collections/test_attributes.py b/tests/integration/collections/test_attributes.py
index c0ece31f6..24b638dfd 100644
--- a/tests/integration/collections/test_attributes.py
+++ b/tests/integration/collections/test_attributes.py
@@ -18,7 +18,7 @@
 from albert.resources.inventory import InventoryItem
 from albert.resources.lots import Lot
 from albert.resources.parameter_groups import DataType, EnumValidationValue, Operator
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("inventory")
 
diff --git a/tests/integration/collections/test_btinsight.py b/tests/integration/collections/test_btinsight.py
index 0df349e15..a480ce976 100644
--- a/tests/integration/collections/test_btinsight.py
+++ b/tests/integration/collections/test_btinsight.py
@@ -2,7 +2,7 @@
 
 from albert import Albert
 from albert.resources.btinsight import BTInsight, BTInsightCategory, BTInsightRegistry
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("bt")
 
diff --git a/tests/integration/collections/test_custom_templates.py b/tests/integration/collections/test_custom_templates.py
index 7516e6419..f3203b514 100644
--- a/tests/integration/collections/test_custom_templates.py
+++ b/tests/integration/collections/test_custom_templates.py
@@ -9,7 +9,7 @@
     _CustomTemplateDataUnion,
 )
 from albert.resources.users import User
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("customtemplates")
 
diff --git a/tests/integration/collections/test_data_columns.py b/tests/integration/collections/test_data_columns.py
index 8f7819119..daaaa0ce0 100644
--- a/tests/integration/collections/test_data_columns.py
+++ b/tests/integration/collections/test_data_columns.py
@@ -2,7 +2,7 @@
 
 from albert import Albert
 from albert.resources.data_columns import DataColumn
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("datatemplates")
 
diff --git a/tests/integration/collections/test_data_templates.py b/tests/integration/collections/test_data_templates.py
index 365f3a2fa..2596f248f 100644
--- a/tests/integration/collections/test_data_templates.py
+++ b/tests/integration/collections/test_data_templates.py
@@ -25,7 +25,7 @@
 from albert.resources.tags import Tag
 from albert.resources.units import Unit
 from albert.resources.users import User
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("datatemplates")
 
diff --git a/tests/integration/collections/test_files.py b/tests/integration/collections/test_files.py
index a155343b0..8e3e80ee1 100644
--- a/tests/integration/collections/test_files.py
+++ b/tests/integration/collections/test_files.py
@@ -7,7 +7,7 @@
 from albert import Albert
 from albert.exceptions import NotFoundError
 from albert.resources.files import FileNamespace
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 
 def test_file_round_trip(client: Albert):
diff --git a/tests/integration/collections/test_inventory.py b/tests/integration/collections/test_inventory.py
index 44881f355..6cf3a617d 100644
--- a/tests/integration/collections/test_inventory.py
+++ b/tests/integration/collections/test_inventory.py
@@ -22,7 +22,7 @@
 from albert.resources.storage_locations import StorageLocation, StorageLocationFilter
 from albert.resources.tags import Tag
 from albert.resources.users import User
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("inventory")
 
@@ -62,6 +62,8 @@ def filter_seeded(items):
         return [item for item in items if normalize_inv_id(item.id) in seeded_ids]
 
     def scoped_search(*, created_by=None, updated_by=None):
+        # Every seeded item is created by static_user and matches text=seed_prefix, so
+        # the complete scoped set is seeded_ids; wait for all of it before comparing.
         return poll_until(
             lambda: filter_seeded(
                 list(
@@ -72,7 +74,10 @@ def scoped_search(*, created_by=None, updated_by=None):
                         max_items=100,
                     )
                 )
-            )
+            ),
+            predicate=lambda results: (
+                {normalize_inv_id(item.id) for item in results} == seeded_ids
+            ),
         )
 
     results = poll_until(
@@ -101,17 +106,17 @@ def scoped_search(*, created_by=None, updated_by=None):
     assert test_item.created and test_item.created.at
     from_created_at = test_item.created.at.date().isoformat()
     recently_created = poll_until(
-        lambda: filter_seeded(
-            list(
-                client.inventory.search(
-                    text=seed_prefix,
-                    from_created_at=from_created_at,
-                    max_items=100,
-                )
+        lambda: [
+            item
+            for item in client.inventory.search(
+                text=seed_prefix,
+                from_created_at=from_created_at,
+                max_items=100,
             )
-        )
+            if normalize_inv_id(item.id) == test_item.id
+        ]
     )
-    assert test_item.id in {normalize_inv_id(item.id) for item in recently_created}
+    assert recently_created, "Expected the seeded item in from_created_at search results"
 
     hydrated_by_creator = poll_until(
         lambda: filter_seeded(
@@ -130,9 +135,13 @@ def scoped_search(*, created_by=None, updated_by=None):
     assert facets
 
     search_hits = poll_until(
-        lambda: filter_seeded(list(client.inventory.search(text=test_item.name, max_items=10)))
+        lambda: [
+            item
+            for item in client.inventory.search(text=test_item.name, max_items=10)
+            if normalize_inv_id(item.id) == test_item.id
+        ]
     )
-    hit = next(item for item in search_hits if normalize_inv_id(item.id) == test_item.id)
+    hit = search_hits[0]
     assert hit.manufacturer is not None
     company_name = (
         test_item.company.name if isinstance(test_item.company, Company) else test_item.company
@@ -185,12 +194,16 @@ def search_scoped(**kwargs):
         ]
 
     filter_results = poll_until(
-        lambda: search_scoped(storage_location=[StorageLocationFilter(name=unit.name)])
+        lambda: search_scoped(storage_location=[StorageLocationFilter(name=unit.name)]),
+        predicate=lambda results: {f"INV{p.id}" for p in results} == expected_ids,
     )
     assert {f"INV{p.id}" for p in filter_results} == expected_ids
 
     # The full StorageLocation object from a lookup remains accepted.
-    object_results = poll_until(lambda: search_scoped(storage_location=unit))
+    object_results = poll_until(
+        lambda: search_scoped(storage_location=unit),
+        predicate=lambda results: {f"INV{p.id}" for p in results} == expected_ids,
+    )
     assert {f"INV{p.id}" for p in object_results} == expected_ids
 
 
diff --git a/tests/integration/collections/test_notebooks.py b/tests/integration/collections/test_notebooks.py
index 154f4d1fd..bab8dd0b7 100644
--- a/tests/integration/collections/test_notebooks.py
+++ b/tests/integration/collections/test_notebooks.py
@@ -15,7 +15,7 @@
 )
 from albert.resources.projects import Project
 from tests.integration.seeding import generate_notebook_block_seeds, generate_notebook_seeds
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("projects")
 
@@ -111,18 +111,16 @@ def test_update_block_content_raises_exception(client: Albert, seeded_notebook:
 def test_search(client: Albert, seed_prefix: str, seeded_notebooks: list[Notebook]):
     """Test search finds seeded notebook block content scoped to the seed project."""
     nb = seeded_notebooks[0]
-    seeded_ids = {n.id for n in seeded_notebooks}
     hits = poll_until(
         lambda: [
             hit
             for hit in client.notebooks.search(
                 text=seed_prefix, project_id=nb.parent_id, max_items=50
             )
-            if hit.notebook_id in seeded_ids and hit.block_id
+            if hit.notebook_id == nb.id and hit.block_id
         ]
     )
-    assert hits, "Expected at least one notebook search hit"
-    assert any(hit.notebook_id == nb.id for hit in hits)
+    assert hits, "Expected seeded notebook in search results"
 
 
 def test_get_block_by_id(client: Albert, seeded_notebooks: list[Notebook]):
diff --git a/tests/integration/collections/test_parameter_groups.py b/tests/integration/collections/test_parameter_groups.py
index 370f34aa1..d56cca922 100644
--- a/tests/integration/collections/test_parameter_groups.py
+++ b/tests/integration/collections/test_parameter_groups.py
@@ -13,7 +13,7 @@
 )
 from albert.resources.tags import Tag
 from albert.resources.units import Unit
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("datatemplates")
 
@@ -85,7 +85,6 @@ def test_parameter_group_search(
     parameter = pg.parameters[0].name
     assert tag and parameter
 
-    seeded_ids = {item.id for item in seeded_parameter_groups}
     hits = poll_until(
         lambda: [
             hit
@@ -112,11 +111,11 @@ def test_parameter_group_search(
                 additional_field=["owner", "tags", "createdByName"],
                 max_items=50,
             )
-            if hit.id in seeded_ids
+            if hit.id == pg.id
         ]
     )
     assert_valid_parameter_groups(results, ParameterGroupSearchItem)
-    assert pg.id in {hit.id for hit in results}
+    assert results[0].id == pg.id
 
 
 def test_hydrate_pg(client: Albert, seed_prefix: str, seeded_parameter_groups):
diff --git a/tests/integration/collections/test_parameters.py b/tests/integration/collections/test_parameters.py
index 318221180..260726198 100644
--- a/tests/integration/collections/test_parameters.py
+++ b/tests/integration/collections/test_parameters.py
@@ -4,7 +4,7 @@
 
 from albert.client import Albert
 from albert.resources.parameters import Parameter
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("datatemplates")
 
@@ -35,6 +35,7 @@ def test_parameter_get_all_by_ids(client: Albert, seeded_parameters: list[Parame
     ids = [x.id for x in seeded_parameters]
     results = poll_until(
         lambda: [p for p in client.parameters.get_all(ids=ids, max_items=10) if p.id in set(ids)],
+        predicate=lambda results: len(results) == len(ids),
         timeout=15.0,
         interval=1.0,
     )
@@ -53,6 +54,7 @@ def test_get_by_ids(client: Albert, seeded_parameters: list[Parameter]):
     ids = [x.id for x in seeded_parameters]
     results = poll_until(
         lambda: [p for p in client.parameters.get_by_ids(ids=ids) if p.id in set(ids)],
+        predicate=lambda results: len(results) == len(ids),
         timeout=15.0,
         interval=1.0,
     )
@@ -74,6 +76,7 @@ def test_get_by_ids_omits_unknown_ids(client: Albert, seeded_parameters: list[Pa
             for p in client.parameters.get_by_ids(ids=[*known_ids, "PRM0"])
             if p.id in set(known_ids)
         ],
+        predicate=lambda results: len(results) == len(known_ids),
         timeout=15.0,
         interval=1.0,
     )
diff --git a/tests/integration/collections/test_product_design.py b/tests/integration/collections/test_product_design.py
index 403f5a7d6..79cc8b4a4 100644
--- a/tests/integration/collections/test_product_design.py
+++ b/tests/integration/collections/test_product_design.py
@@ -4,7 +4,7 @@
 from albert.resources.inventory import InventoryItem
 from albert.resources.product_design import UnpackedProductDesign
 from albert.resources.worksheets import Worksheet
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("sheets")
 
@@ -16,7 +16,7 @@ def test_search(
     seeded_worksheet: Worksheet,
 ):
     """Test search finds the seeded formula on the product design grid within its project."""
-    seeded_ids = {p.id for p in seeded_products}
+    product = seeded_products[0]
     hits = poll_until(
         lambda: [
             hit
@@ -25,11 +25,11 @@ def test_search(
                 project_id=seeded_worksheet.project_id,
                 max_items=50,
             )
-            if hit.id in seeded_ids
+            if hit.id == product.id
         ]
     )
     assert hits, "Expected seeded formula in product design search results"
-    assert seeded_products[0].id in {hit.id for hit in hits}
+    assert hits[0].id == product.id
 
 
 def test_get_unpacked(client: Albert, seeded_products: list[InventoryItem]):
diff --git a/tests/integration/collections/test_projects.py b/tests/integration/collections/test_projects.py
index 46e22f9b7..15bb44783 100644
--- a/tests/integration/collections/test_projects.py
+++ b/tests/integration/collections/test_projects.py
@@ -8,7 +8,7 @@
 from albert.resources.acls import ACL, AccessControlLevel
 from albert.resources.attachments import Attachment
 from albert.resources.projects import DocumentSearchItem, Project, ProjectSearchItem
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("projects")
 
diff --git a/tests/integration/collections/test_reports.py b/tests/integration/collections/test_reports.py
index aaf4a3b3b..183603b48 100644
--- a/tests/integration/collections/test_reports.py
+++ b/tests/integration/collections/test_reports.py
@@ -12,7 +12,7 @@
     FullAnalyticalReport,
 )
 from albert.resources.tasks import BaseTask
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("tasks")
 
@@ -24,7 +24,6 @@ def test_search_reports(
 ):
     """Test searching reports finds seeded reports scoped to their project."""
     expected = seeded_reports[0]
-    seeded_ids = {r.id for r in seeded_reports}
     hits = poll_until(
         lambda: [
             hit
@@ -33,13 +32,12 @@ def test_search_reports(
                 project_id=expected.project_id,
                 max_items=50,
             )
-            if hit.id in seeded_ids
+            if hit.id == expected.id
         ]
     )
-    hit_ids = {hit.id for hit in hits}
-    assert expected.id in hit_ids
+    assert hits, "Expected seeded report in search results"
 
-    hit = next(item for item in hits if item.id == expected.id)
+    hit = hits[0]
     assert hit.name is not None
     assert seed_prefix in hit.name
     assert hit.project_id == expected.project_id
diff --git a/tests/integration/collections/test_synthesis.py b/tests/integration/collections/test_synthesis.py
index 08f5ec91f..eb59ae8d8 100644
--- a/tests/integration/collections/test_synthesis.py
+++ b/tests/integration/collections/test_synthesis.py
@@ -5,7 +5,7 @@
 from albert import Albert
 from albert.exceptions import AlbertException
 from albert.resources.notebooks import Notebook
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("projects")
 
diff --git a/tests/integration/collections/test_tasks.py b/tests/integration/collections/test_tasks.py
index cd7c703fe..78acf91dc 100644
--- a/tests/integration/collections/test_tasks.py
+++ b/tests/integration/collections/test_tasks.py
@@ -26,7 +26,7 @@
 from albert.resources.worker_jobs import WorkerJob
 from albert.resources.workflows import Workflow
 from tests.integration.utils.metadata import change_metadata, make_metadata_update_assertions
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("tasks")
 
diff --git a/tests/integration/collections/test_teams.py b/tests/integration/collections/test_teams.py
index d65bdd714..b08efebfc 100644
--- a/tests/integration/collections/test_teams.py
+++ b/tests/integration/collections/test_teams.py
@@ -7,7 +7,7 @@
 from albert.exceptions import AlbertException, NotFoundError
 from albert.resources.teams import Team, TeamMember
 from albert.resources.users import User
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("teams")
 
diff --git a/tests/integration/collections/test_workflows.py b/tests/integration/collections/test_workflows.py
index 7184efa53..ace5c516c 100644
--- a/tests/integration/collections/test_workflows.py
+++ b/tests/integration/collections/test_workflows.py
@@ -4,7 +4,7 @@
 from albert.resources.facet import FacetItem
 from albert.resources.parameters import ParameterCategory
 from albert.resources.workflows import Workflow, WorkflowSearchItem
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("tasks")
 
diff --git a/tests/integration/collections/test_worksheet.py b/tests/integration/collections/test_worksheet.py
index 4c5245b16..027b3fd62 100644
--- a/tests/integration/collections/test_worksheet.py
+++ b/tests/integration/collections/test_worksheet.py
@@ -3,7 +3,7 @@
 from albert import Albert
 from albert.resources.inventory import InventoryItem
 from albert.resources.worksheets import Worksheet
-from tests.integration.utils.wait import poll_until
+from tests.utils.wait import poll_until
 
 pytestmark = pytest.mark.xdist_group("sheets")
 
diff --git a/tests/integration/utils/wait.py b/tests/integration/utils/wait.py
deleted file mode 100644
index 2809e1082..000000000
--- a/tests/integration/utils/wait.py
+++ /dev/null
@@ -1,16 +0,0 @@
-import time
-from collections.abc import Callable
-
-
-def poll_until(fetch: Callable[[], list], *, timeout: float = 30.0, interval: float = 1.0) -> list:
-    """Poll ``fetch`` until it returns a non-empty result or the timeout elapses.
-
-    Search-index-backed endpoints lag behind seeding; tests asserting on fresh
-    seeds poll instead of assuming immediate visibility. Returns the last result.
-    """
-    deadline = time.monotonic() + timeout
-    while True:
-        result = fetch()
-        if result or time.monotonic() >= deadline:
-            return result
-        time.sleep(interval)
diff --git a/tests/unit/TESTING.md b/tests/unit/TESTING.md
index 56e5f4abf..e0d4396f9 100644
--- a/tests/unit/TESTING.md
+++ b/tests/unit/TESTING.md
@@ -32,7 +32,7 @@ The conftest guards enforce the first two mechanically:
 - **Open a network connection.** `_block_network` raises on any real `connect`.
 - **Read credentials or `.env`.** `_strip_albert_env` removes every `ALBERT_*` variable.
 - **Import from `tests/integration/`** (`client`, `seeded_*`, `static_*`, `seed_prefix`,
-  `poll_until`, `seeding`).
+  `seeding`). Shared, network-free helpers live in `tests/utils/` (e.g. `poll_until`).
 - **Encode assumed server behavior.** A fake response may be shaped like a real one, but the
   assertion is about what the SDK sends or how it reacts, never "the API returns X".
 - **Assert trivia.** That a Pydantic field exists, that a default is `None`, that a
diff --git a/tests/unit/utils/test_wait.py b/tests/unit/utils/test_wait.py
new file mode 100644
index 000000000..9cd6a7406
--- /dev/null
+++ b/tests/unit/utils/test_wait.py
@@ -0,0 +1,94 @@
+import json
+
+import pytest
+import requests
+
+from albert.exceptions import InternalServerError
+from tests.utils.wait import poll_until
+
+
+def _make_server_error() -> InternalServerError:
+    req = requests.PreparedRequest()
+    req.method = "GET"
+    req.url = "https://example.com/api/v3/x"
+    req.body = None
+    resp = requests.Response()
+    resp.status_code = 500
+    resp.reason = "Internal Server Error"
+    resp.request = req
+    resp._content = json.dumps({"errors": "transient failure"}).encode()
+    resp.encoding = "utf-8"
+    return InternalServerError(resp)
+
+
+def test_poll_until_returns_first_non_empty_by_default():
+    """Test that polling stops at the first non-empty result with no predicate."""
+    calls = iter([[], ["a"], ["a", "b"]])
+    assert poll_until(lambda: next(calls), interval=0) == ["a"]
+
+
+def test_poll_until_predicate_waits_for_satisfaction():
+    """Test that polling continues until the predicate accepts the result."""
+    calls = iter([["a"], ["a", "b"], ["a", "b"]])
+    result = poll_until(
+        lambda: next(calls),
+        predicate=lambda result: len(result) == 2,
+        interval=0,
+    )
+    assert result == ["a", "b"]
+
+
+def test_poll_until_returns_last_result_on_timeout():
+    """Test that an unsatisfied predicate returns the last result after the timeout."""
+    result = poll_until(lambda: [], predicate=lambda result: False, timeout=0.01, interval=0.005)
+    assert result == []
+
+
+def test_poll_until_returns_last_partial_result_on_timeout():
+    """Test that a never-satisfied predicate returns the latest partial result."""
+    partials = [["a"], ["a", "b"], ["a", "b", "c"]]
+    calls = iter(partials)
+    result = poll_until(
+        lambda: next(calls, partials[-1]),
+        predicate=lambda result: len(result) == 4,
+        timeout=0.05,
+        interval=0.01,
+    )
+    assert result == ["a", "b", "c"]
+
+
+def test_poll_until_recovers_from_transient_fetch_error():
+    """Test that a one-off AlbertServerError does not fail the poll."""
+    outcomes = iter([_make_server_error(), ["a"]])
+
+    def fetch():
+        outcome = next(outcomes)
+        if isinstance(outcome, Exception):
+            raise outcome
+        return outcome
+
+    assert poll_until(fetch, interval=0) == ["a"]
+
+
+def test_poll_until_reraises_fetch_error_after_timeout():
+    """Test that a fetch that always raises AlbertServerError propagates after the timeout."""
+
+    def fetch():
+        raise _make_server_error()
+
+    with pytest.raises(InternalServerError):
+        poll_until(fetch, timeout=0.01, interval=0.005)
+
+
+def test_poll_until_does_not_catch_non_server_errors():
+    """Test that non-AlbertServerError exceptions propagate immediately without polling."""
+    calls = [0]
+
+    def fetch():
+        calls[0] += 1
+        raise AttributeError("lambda typo")
+
+    with pytest.raises(AttributeError, match="lambda typo"):
+        poll_until(fetch, timeout=30.0, interval=0)
+
+    assert calls[0] == 1
diff --git a/tests/utils/__init__.py b/tests/utils/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/tests/utils/wait.py b/tests/utils/wait.py
new file mode 100644
index 000000000..ac965b877
--- /dev/null
+++ b/tests/utils/wait.py
@@ -0,0 +1,39 @@
+import time
+from collections.abc import Callable
+
+from albert.exceptions import AlbertServerError
+
+
+def poll_until(
+    fetch: Callable[[], list],
+    *,
+    predicate: Callable[[list], bool] | None = None,
+    timeout: float = 30.0,
+    interval: float = 1.0,
+) -> list:
+    """Poll ``fetch`` until the result satisfies ``predicate`` or the timeout elapses.
+
+    Search-index-backed endpoints lag behind seeding; tests asserting on fresh
+    seeds poll instead of assuming immediate visibility. With no ``predicate``,
+    any non-empty result ends polling. Returns the last result.
+
+    Pass a ``predicate`` when the assertion needs the complete expected set: a
+    non-empty but partially indexed result satisfies the default check while
+    remaining items are still becoming visible.
+
+    ``AlbertServerError`` exceptions raised by ``fetch`` (for example a transient
+    5xx) do not end polling: the poll retries until the deadline and re-raises the
+    last exception if the deadline expires on an erroring attempt.
+    """
+    deadline = time.monotonic() + timeout
+    while True:
+        try:
+            result = fetch()
+        except AlbertServerError:
+            if time.monotonic() >= deadline:
+                raise
+        else:
+            ready = predicate(result) if predicate is not None else bool(result)
+            if ready or time.monotonic() >= deadline:
+                return result
+        time.sleep(interval)