diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
index 0dd35fc..937a908 100644
--- a/.github/workflows/ci.yaml
+++ b/.github/workflows/ci.yaml
@@ -49,6 +49,8 @@ jobs:
           python -m pip install --upgrade pip wheel 'setuptools<58'
           pip install -r requirements_dev.txt
           pip install -U codecov
+      - name: Print out pip freeze
+        run: pip freeze
       - name: Lint
         run: |
             pre-commit install
@@ -87,14 +89,16 @@ jobs:
         run: echo "$IS_PR_TARGETING_MASTER" "$MASTER_PUSH"
       - name: Run tests
         # use temporary directory cleaned up after every job
-        run: pytest --basetemp={runner.temp} --cov=./ --cov-report xml --mpl --mpl-results-path=tests/results
+        run: pytest --basetemp=${{ runner.temp }} --cov=./ --cov-report xml --mpl --mpl-results-path=tests/results
         env:
           MPLBACKEND: Agg
       - name: Upload pytest test result artifacts on failure
         uses: actions/upload-artifact@v2
         with:
           name: pytest-results-${{ matrix.python-version }}
-          path: tests/results
+          path: |
+            tests/results
+            ${{ runner.temp }}/test*current
         if: ${{ failure() }}
       - name: Upload coverage on success
         uses: codecov/codecov-action@v1
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 1ca88b2..096a961 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -11,6 +11,6 @@ repos:
     -   id: requirements-txt-fixer
     -   id: check-merge-conflict
 -   repo: https://github.com/psf/black
-    rev: 19.3b0
+    rev: 22.3.0
     hooks:
     -   id: black
diff --git a/Makefile b/Makefile
index 3d707f7..51c061b 100644
--- a/Makefile
+++ b/Makefile
@@ -72,11 +72,11 @@ test-without-figures:
 
 ## regenerate baseline figures
 regen-snapshot-figures: build-docker-test-image
-	docker run --rm -it -v $$(pwd):/src genetools-test pytest --mpl-generate-path=tests/baseline;
+	docker run --rm -it -v $$(pwd):/src genetools-test pytest --mpl-generate-path=tests/baseline --snapshot-update;
 
 ## regenerate saved test data (and baseline figures)
 regen-test-data: build-docker-test-image
-	docker run --rm -it -v $$(pwd):/src genetools-test pytest --mpl-generate-path=tests/baseline --regenerate-anndata;
+	docker run --rm -it -v $$(pwd):/src genetools-test pytest --mpl-generate-path=tests/baseline --snapshot-update --regenerate-anndata;
 
 coverage: ## check code coverage quickly with the default Python
 	coverage run --source genetools -m pytest
diff --git a/requirements_dev.txt b/requirements_dev.txt
index 8b83ccb..a161067 100644
--- a/requirements_dev.txt
+++ b/requirements_dev.txt
@@ -5,16 +5,16 @@ flake8==3.7.8
 h5py>=2.10.0
 igraph>=0.9.9
 louvain>=0.6.1
-matplotlib>=3.5.1
+matplotlib>=3.5.2
 # numba requires numpy<21
 numpy>=1.18.3,<1.21
 pandas>=1.3.5
 pip>=21.1.1
 pre-commit>=2.15.0
-pytest==4.6.5
-pytest-cov==2.10.0
+pytest-cov>=2.10.0
 pytest-mpl==0.11
 pytest-runner==5.1
+pytest>=5.1.0,<8.0.0
 recommonmark>=0.7.1
 requests>=2.26.0
 scanpy>=1.8.2
@@ -22,6 +22,7 @@ scikit-learn>=1.0.1
 scipy>=1.4.1
 seaborn>=0.11.2
 sphinx>=4
+syrupy>=2.3.1
 twine==1.14.0
 watchdog==0.9.0
 wheel==0.33.6
diff --git a/tests/__snapshots__/test_plots.ambr b/tests/__snapshots__/test_plots.ambr
new file mode 100644
index 0000000..74fec27
--- /dev/null
+++ b/tests/__snapshots__/test_plots.ambr
@@ -0,0 +1,3 @@
+# name: test_pdf_deterministic_output
+  'e8910b326f5ab726de325b29cf11f5a2'
+# ---
diff --git a/tests/test_plots.py b/tests/test_plots.py
index f54d5d0..aa7c4a8 100644
--- a/tests/test_plots.py
+++ b/tests/test_plots.py
@@ -257,10 +257,17 @@ def test_wrap_labels_overrides_any_linebreaks_in_labels():
     return fig
 
 
-def test_pdf_deterministic_output(tmp_path):
+def test_pdf_deterministic_output(tmp_path, snapshot):
     # Can't use snapshot_image here because pytest-mpl doesn't support PDF
-    # So we are doing our own snapshot test md5 checksum here
+    # So we are doing our own snapshot test md5 checksum here.
+    #
+    # Unlike pytest-mpl, which allows minute differences in the resulting image,
+    # the MD5 checksum relies on exact output from matplotlib, which is not guaranteed to be stable.
+    # The expected MD5 checksum will change based on matplotlib version unfortunately.
+    # We use syrupy to store the expected MD5 checksum. Regenerate with pytest --snapshot-update
+    #
     # This also allows us to test genetools.plots.savefig directly.
+    #
 
     fname = tmp_path / "test_pdf_determinstic_output.pdf"
 
@@ -277,12 +284,10 @@ def get_md5(fname):
         with open(fname, "rb") as f:
             return hashlib.md5(f.read()).hexdigest()
 
-    # A hacky snapshot test: put expected md5sum here
-    expected_md5 = "4fae9ebe9cb8f837aed495fee12ca179"
     observed_md5 = get_md5(fname)
     assert (
-        expected_md5 == observed_md5
-    ), f"{fname} md5sum mismatch: got {observed_md5}, expected {expected_md5}"
+        observed_md5 == snapshot
+    ), f"{fname} md5sum mismatch: got {observed_md5}, expected {snapshot}"
 
 
 @snapshot_image
diff --git a/tests/test_stats.py b/tests/test_stats.py
index 00023d8..6207dcc 100644
--- a/tests/test_stats.py
+++ b/tests/test_stats.py
@@ -50,8 +50,7 @@ def test_normalize_cols():
 
 
 def test_self_coclustering():
-    """Coclustering sanity check: dataset against itself should always yield 100%
-    """
+    """Coclustering sanity check: dataset against itself should always yield 100%"""
     cluster_ids = [1, 1, 2, 2]
     # 6 potential relationships in each dataset: 1-2, 1-3, 1-4, 2-3, 2-4, 3-4
     # Actual coclustering in each dataset: 1-2, 3-4