diff --git a/graphblas/ss/_core.py b/graphblas/ss/_core.py
index 5e851b7cc..f036b3bb7 100644
--- a/graphblas/ss/_core.py
+++ b/graphblas/ss/_core.py
@@ -259,9 +259,13 @@ class About(Mapping):
     _modes = {
         lib.GrB_NONBLOCKING: "nonblocking",
         lib.GrB_BLOCKING: "blocking",
-        lib.GxB_NONBLOCKING_GPU: "nonblocking_gpu",
-        lib.GxB_BLOCKING_GPU: "blocking_gpu",
     }
+    # GxB_NONBLOCKING_GPU and GxB_BLOCKING_GPU were modes of their own in SS:GraphBLAS <=10.5.1.
+    # In 10.5.2 they became aliases of GrB_NONBLOCKING and GrB_BLOCKING, which python-suitesparse-
+    # graphblas leaves out; `setdefault` keeps "nonblocking" and "blocking" if they are aliases.
+    if hasattr(lib, "GxB_NONBLOCKING_GPU"):
+        _modes.setdefault(lib.GxB_NONBLOCKING_GPU, "nonblocking_gpu")
+        _modes.setdefault(lib.GxB_BLOCKING_GPU, "blocking_gpu")
     _mode_options = {
         "mode": lib.GxB_MODE,
     }
diff --git a/graphblas/tests/test_matrix.py b/graphblas/tests/test_matrix.py
index 4befea2cf..ea8a29f26 100644
--- a/graphblas/tests/test_matrix.py
+++ b/graphblas/tests/test_matrix.py
@@ -2084,6 +2084,9 @@ def test_ss_import_export(A, do_iso, methods):
         assert A5.isequal(A)
         assert A5.ss.is_iso is do_iso
     d["bitmap"] = np.concatenate([d["bitmap"], d["bitmap"]], axis=0)
+    if not do_iso:
+        # Values for the whole bitmap; GraphBLAS reads them all
+        d["values"] = np.concatenate([d["values"], d["values"]], axis=0)
     B5b = Matrix.ss.import_any(**d)
     if in_method == "import":
         # B5b == [A, A] (i.e, get 2d shape from bitmap first if possible)
diff --git a/graphblas/tests/test_ss_utils.py b/graphblas/tests/test_ss_utils.py
index 2c9e01a1a..b1469b7a0 100644
--- a/graphblas/tests/test_ss_utils.py
+++ b/graphblas/tests/test_ss_utils.py
@@ -223,6 +223,8 @@ def test_about():
     with pytest.raises(KeyError):
         about["badkey"]
     assert "SuiteSparse" in about["library_name"]
+    # SS:GraphBLAS 10.5.2 made the GPU modes aliases of these; neither may read as a GPU mode
+    assert about["mode"] == ("blocking" if gb._init_params["blocking"] else "nonblocking")
     with pytest.raises(TypeError):
         del about["library_name"]  # pylint: disable=unsupported-delete-operation
     assert "library_name" in repr(about)
diff --git a/scripts/check_versions.sh b/scripts/check_versions.sh
index d9959556d..f702f2f41 100755
--- a/scripts/check_versions.sh
+++ b/scripts/check_versions.sh
@@ -13,6 +13,7 @@
 conda search 'flake8-bugbear[channel=conda-forge]>=26.9.9'
 conda search 'flake8-simplify[channel=conda-forge]>=0.30.0'
 conda search 'numpy[channel=conda-forge]>=2.5'
+# ci_pick_versions.py caps an unpinned pandas below 3.1 for now (see its MAINT note).
 conda search 'pandas[channel=conda-forge]>=3.0'
 conda search 'scipy[channel=conda-forge]>=1.18'
 conda search 'networkx[channel=conda-forge]>=3.6'
diff --git a/scripts/ci_pick_versions.py b/scripts/ci_pick_versions.py
index ebd003d96..8c7d4786c 100755
--- a/scripts/ci_pick_versions.py
+++ b/scripts/ci_pick_versions.py
@@ -83,6 +83,14 @@
     "2.x": ["2.2", "2.3", "3.0", ""],
 }
 
+# Caps on an unpinned ("" = latest) pick, as conda specs. Pins in the pools are not
+# affected; a capped package's newest release stays out of CI until the cap goes.
+# MAINT: 2026-10-22 pandas 3.1.0 fits fewer columns in a repr's auto-width
+# (display.max_columns=0), and warns (Pandas4Warning) when the pandas-based repr renderer
+# appends its "..." row to a numeric column; both fail test_formatting.py. Remove the cap
+# once #605 (the repr renderer that does not use pandas) is on main.
+LATEST_CEILINGS = {"pandas": "<3.1"}
+
 # The "2.x" pool starts at 2.6 because awkward <2.6 still uses numpy.AxisError,
 # which numpy 2.0 removed. Only >=2.10 pins survive numpy >=2.5; see the awkward
 # constraints in `apply_constraints`.
@@ -557,14 +565,17 @@ def format_output(v):
     """Format version dict as bash-eval-safe key=value lines.
 
     Values use '=X.Y' prefix for conda install (e.g., npver='=2.2').
-    Empty means latest, NA means skip.
+    Empty means latest, NA means skip. Latest is capped by LATEST_CEILINGS
+    (e.g., pdver='<3.1').
     psg already has its prefix baked in.
     """
     lines = []
     for key in ("numpy", "scipy", "pandas", "awkward", "networkx", "pyyaml", "sparse", "numba"):
         var = _VAR_NAMES[key]
         val = v[key]
-        if val in ("NA", ""):
+        if val == "" and key in LATEST_CEILINGS:
+            lines.append(f"{var}='{LATEST_CEILINGS[key]}'")
+        elif val in ("NA", ""):
             lines.append(f"{var}='{val}'")
         else:
             lines.append(f"{var}='={val}'")
@@ -598,7 +609,7 @@ def format_summary(v):
         if val == "NA":
             parts.append(f"{name}=NA")
         elif val == "":
-            parts.append(f"{name}=latest")
+            parts.append(f"{name}=latest{LATEST_CEILINGS.get(key, '')}")
         else:
             parts.append(f"{name}={val}")
     return "versions: " + " ".join(parts)