diff --git a/tutorials/AMB2018-02-B/README.md b/tutorials/AMB2018-02-B/README.md
index ffa1705..210a5a7 100644
--- a/tutorials/AMB2018-02-B/README.md
+++ b/tutorials/AMB2018-02-B/README.md
@@ -57,7 +57,16 @@ Defines the laser path, laser power, and scan speed or dwell time.
 constant/dynamicMeshDict
 ```
 
-Defines the dynamic mesh/refinement settings.
+Defines the dynamic mesh/refinement settings and cell-count-based runtime
+redistribution.
+
+```text
+system/decomposeParDict
+```
+
+Selects Scotch for the initial decomposition and PT-Scotch for optional runtime
+redistribution. It also enables the `refinementHistory` constraint required to
+keep refinement siblings together for unrefinement.
 
 ```text
 system/blockMeshDict
@@ -152,6 +161,17 @@ refinement
 The refinement region projects along the heat source path and targets a desired
 average cell load per processor.
 
+Runtime redistribution is disabled by default with `type none` in
+`constant/dynamicMeshDict`. Change it to `type distributor` to enable the
+provided PT-Scotch configuration. The mesh distributor then checks every 10
+time steps and redistributes when an active refinement model creates more than
+10% cell-count imbalance. These are starting values; production cases should
+tune `redistributionInterval` and `maxImbalance` for their mesh and MPI count.
+
+The tutorial uses OpenFOAM's generic `distributor`, which balances cell counts.
+Do not change its type to `loadBalancer`: that component requires registered
+per-cell `cpuLoad` objects that AdditiveFOAM does not currently provide.
+
 ### Coefficients
 
 `model`
@@ -233,6 +253,11 @@ enabled false;
 `solidificationData` writes solidification events for CET analysis.
 `ExaCA` writes temperature history data for ExaCA input files.
 
+The ExaCA function object supports refinement on a fixed processor
+decomposition but not runtime mesh redistribution. Keep the distributor set to
+`type none` in `constant/dynamicMeshDict` when combining ExaCA with an active
+refinement model.
+
 The `Allrun` script calls the reconstruction helpers after the solver finishes:
 
 ```sh
diff --git a/tutorials/AMB2018-02-B/constant/dynamicMeshDict b/tutorials/AMB2018-02-B/constant/dynamicMeshDict
index 5def0f0..0ba4eb7 100644
--- a/tutorials/AMB2018-02-B/constant/dynamicMeshDict
+++ b/tutorials/AMB2018-02-B/constant/dynamicMeshDict
@@ -36,4 +36,17 @@ topoChanger
     dumpLevel       true;
 }
 
+// Cell-count-based redistribution for parallel AMR.
+// To enable, change type from "none" to "distributor".
+distributor
+{
+    type                    none;
+
+    libs                    ("libfvMeshDistributors.so");
+
+    redistributionInterval  10;
+
+    maxImbalance            0.10;
+}
+
 // ************************************************************************* //
diff --git a/tutorials/AMB2018-02-B/system/decomposeParDict b/tutorials/AMB2018-02-B/system/decomposeParDict
index da6a847..bdb38ce 100644
--- a/tutorials/AMB2018-02-B/system/decomposeParDict
+++ b/tutorials/AMB2018-02-B/system/decomposeParDict
@@ -17,6 +17,18 @@ FoamFile
 
 numberOfSubdomains  8;
 
-method              scotch;
+decomposer          scotch;
+
+distributor         ptscotch;
+
+libs                ("libptscotchDecomp.so");
+
+constraints
+{
+    refinementHistory
+    {
+        type            refinementHistory;
+    }
+}
 
 // ************************************************************************* //