diff --git a/CHANGELOG.rst b/CHANGELOG.rst
index 960d389..54620f8 100644
--- a/CHANGELOG.rst
+++ b/CHANGELOG.rst
@@ -3,8 +3,9 @@ Changelog
 
 Version 0.2.1
 -------------
-* The following `cell-density` sub-commands can now optionally take a group_ids_config:
-   `cell-density`, `glia-cell-densities`, `inhibitory-and-excitatory-neuron-densities`, `fit-average-densities`
+* The following *cell-density* sub-commands can now optionally take a *group_ids_config*:
+   *cell-density*, *glia-cell-densities*, *inhibitory-and-excitatory-neuron-densities*, *fit-average-densities*
+
    Otherwise they use the default config for the AIBS atlas.
 
 Version 0.2.0
diff --git a/atlas_densities/app/cell_densities.py b/atlas_densities/app/cell_densities.py
index 55e7e95..11ae1d1 100644
--- a/atlas_densities/app/cell_densities.py
+++ b/atlas_densities/app/cell_densities.py
@@ -855,6 +855,7 @@ def fit_average_densities(
     - some regions can have NaN density values for one or more cell types because they are not
     covered by the selected slices of the volumetric gene marker intensities.
     """
+    Path(fitted_densities_output_path).mkdir(parents=True, exist_ok=True)
 
     L.info("Loading annotation ...")
     annotation = VoxelData.load_nrrd(annotation_path)
diff --git a/atlas_densities/app/data/metadata/README.txt b/atlas_densities/app/data/metadata/README.txt
index 6faa21c..dc8d151 100644
--- a/atlas_densities/app/data/metadata/README.txt
+++ b/atlas_densities/app/data/metadata/README.txt
@@ -36,3 +36,12 @@ A clause is formed of dictionary of the form:
 The attribute name is used for the atlas lookup, things like "name" or "acronym" are valid.
 
 Finally, one can refer to a previous stanza by preceding it with a `!`.
+
+The current `group_ids.json was discussed here:
+    https://github.com/BlueBrain/atlas-densities/pull/51#issuecomment-1748445813
+    In very short, in older versions of the annotations, fibers and main regions were in separated files (ccfv2 / ccfbbp).
+    In the main regions, the space allocated to the fibers were annotated to Basic cell groups and regions.
+    For some weird reasons, the fibers do not take the entire space allocated for them.
+    Which means that for most of the voxels annotated to Basic cell groups and regions, they correspond to fibers.
+    I say most of the voxels because there are also voxels at the frontier between two main regions (e.g. Cerebellum) that were not annotated to their closest leaf region.
+    These voxels are gone in ccfv3.