diff --git a/workflows/prognostic_c48_run/docs/development.rst b/workflows/prognostic_c48_run/docs/development.rst
index e10c46386d..52d61d57ec 100644
--- a/workflows/prognostic_c48_run/docs/development.rst
+++ b/workflows/prognostic_c48_run/docs/development.rst
@@ -3,35 +3,26 @@
 Developer's Guide
 -----------------
 
-The prognostic run is developed via docker and docker-compose. This
-environment is based off the `prognostic_run` docker image, but has
-bind-mounts to the packages in "/external" of this repository and this
-directory, which allows locally developing this workflow and its
-dependencies.
+The prognostic run is developed via docker. This environment is based off the
+`prognostic_run` docker image, but has bind-mounts to the packages in "/external"
+of this repository and this directory, which allows locally developing this workflow
+and its dependencies.
 
 It is usually fastest to use the latest docker image from Google Container
 Repository. Pull the image::
 
-    docker pull us.gcr.io/vcm-ml/prognostic_run:latest
+    make pull_image_prognostic_run
 
 .. note::
 
     If you run into problems, it would be best to rebuild the docker image from scratch::
 
-        docker-compose build fv3
+        make build_image_prognostic_run
 
 Enter a bash shell in the image::
 
-    docker-compose run fv3net bash
+    make enter_prognostic_run
 
-.. note ::
-
-    This docker-compose will propagate key-based authentication to Google
-    Cloud Platform into the docker image. It expects that environmental variable
-    ``GOOGLE_APPLICATION_CREDENTIALS`` points to a json key. See Google's
-    `documentation <https://cloud.google.com/iam/docs/creating-managing-service-account-keys>`_
-    on how to generate one.
-
-Run the tests::
+Then run the tests::
 
     pytest
\ No newline at end of file
diff --git a/workflows/prognostic_c48_run/runtime/loop.py b/workflows/prognostic_c48_run/runtime/loop.py
index 40b1d5673d..b8fdb154e1 100644
--- a/workflows/prognostic_c48_run/runtime/loop.py
+++ b/workflows/prognostic_c48_run/runtime/loop.py
@@ -338,10 +338,6 @@ def _open_model(self, ml_config: MachineLearningConfig, step: str):
     def time(self) -> cftime.DatetimeJulian:
         return self._state.time
 
-    def cleanup(self):
-        self._print_global_timings()
-        self._fv3gfs.cleanup()
-
     def _step_dynamics(self) -> Diagnostics:
         self._log_debug(f"Dynamics Step")
         self._fv3gfs.step_dynamics()
@@ -378,17 +374,21 @@ def _apply_physics(self) -> Diagnostics:
             "total_precip_after_physics": self._state[TOTAL_PRECIP],
         }
 
-    def _print_timing(self, name, min_val, max_val, mean_val):
-        self._print(f"{name:<30}{min_val:15.4f}{max_val:15.4f}{mean_val:15.4f}")
-
-    def _print_global_timings(self, root=0):
-        is_root = self.rank == root
-        recvbuf = np.array(0.0)
-        reduced = {}
+    def _print_timings(self, reduced):
         self._print("-----------------------------------------------------------------")
         self._print("         Reporting clock statistics from python                  ")
         self._print("-----------------------------------------------------------------")
         self._print(f"{' ':<30}{'min (s)':>15}{'max (s)':>15}{'mean (s)':>15}")
+        for name, timing in reduced.items():
+            self._print(
+                f"{name:<30}{timing['min']:15.4f}"
+                f"{timing['max']:15.4f}{timing['mean']:15.4f}"
+            )
+
+    def log_global_timings(self, root=0):
+        is_root = self.rank == root
+        recvbuf = np.array(0.0)
+        reduced = {}
         for name, value in self._timer.times.items():
             reduced[name] = {}
             for label, op in [("min", MPI.MIN), ("max", MPI.MAX), ("mean", MPI.SUM)]:
@@ -396,10 +396,12 @@ def _print_global_timings(self, root=0):
                 if is_root and label == "mean":
                     recvbuf /= self.comm.Get_size()
                 reduced[name][label] = recvbuf.copy().item()
-            self._print_timing(
-                name, reduced[name]["min"], reduced[name]["max"], reduced[name]["mean"]
-            )
-        self._log_info(f"python_timing:{json.dumps(reduced)}")
+        self._print_timings(reduced)
+        log_out = {
+            "steps": reduced,
+            "units": "[s], cumulative and reduced across ranks",
+        }
+        self._log_info(json.dumps({"python_timing": log_out}))
 
     def _step_prephysics(self) -> Diagnostics:
 
@@ -529,6 +531,11 @@ def _apply_postphysics_to_dycore_state(self) -> Diagnostics:
         )
         return diagnostics
 
+    def _intermediate_restarts(self) -> Diagnostics:
+        self._log_info("Saving intermediate restarts if enabled.")
+        self._fv3gfs.save_intermediate_restart_if_enabled()
+        return {}
+
     def __iter__(
         self,
     ) -> Iterator[Tuple[cftime.DatetimeJulian, Dict[str, xr.DataArray]]]:
@@ -551,6 +558,7 @@ def __iter__(
                 ),
                 self._compute_postphysics,
                 self.monitor("python", self._apply_postphysics_to_dycore_state),
+                self._intermediate_restarts,
             ]:
                 with self._timer.clock(substep.__name__):
                     diagnostics.update(substep())
diff --git a/workflows/prognostic_c48_run/runtime/main.py b/workflows/prognostic_c48_run/runtime/main.py
index 7a3134f77f..2b58b2b820 100644
--- a/workflows/prognostic_c48_run/runtime/main.py
+++ b/workflows/prognostic_c48_run/runtime/main.py
@@ -71,6 +71,8 @@ def main():
     for diag_file in diag_files:
         diag_file.flush()
 
+    loop.log_global_timings()
+
 
 if __name__ == "__main__":