diff --git a/.github/workflows/documentation.yml b/.github/workflows/documentation.yml
index c39ca04..197a774 100644
--- a/.github/workflows/documentation.yml
+++ b/.github/workflows/documentation.yml
@@ -21,7 +21,6 @@ jobs:
         make doc-install
     - name: Build sphinx documentation
       run: |
-        ls ./docs
         make documentation
     - name: Configure AWS Credentials
       uses: aws-actions/configure-aws-credentials@v1
@@ -31,7 +30,6 @@ jobs:
         aws-region: us-east-1
     - name: Deploy static site to S3 bucket
       run: |
-        ls ./docs
         aws s3 rm s3://${{ secrets.AWS_DOCUMENTATION_BUCKET }}/documentation --recursive
         aws s3 sync ./docs/documentation/ s3://${{ secrets.AWS_DOCUMENTATION_BUCKET }}/deep-experiments/ --delete
 #    - name: gcloud auth
diff --git a/Makefile b/Makefile
index 3e09790..35183ef 100644
--- a/Makefile
+++ b/Makefile
@@ -29,6 +29,7 @@ streamlit-deploy:
 	docker push 961104659532.dkr.ecr.us-east-1.amazonaws.com/streamlit
 
 documentation:
+	rm -rf docs/documentation
 	sphinx-build -b html docs/source docs/documentation
 
 documentation-push:
diff --git a/docs/source/dfs-data/description.rst b/docs/source/dfs-data/description.rst
new file mode 100644
index 0000000..21b89b6
--- /dev/null
+++ b/docs/source/dfs-data/description.rst
@@ -0,0 +1,57 @@
+Data
+====
+
+The data for the project is in the folder `data`.
+The `immap` subfolder contains only IMMAP data, while `frameworks_data` contains all the rest.
+
+What is our data?
+------------------------
+
+Our data is composed of ~100.000 sentences extracted from PDF and web articles. Each sentence has been manually
+labeled by taggers. Multiple labels are associated to the same sentence:
+
+- Sector
+- Pillar
+- Subpillar
+
+We want to predict all of them. Each subpillar belongs to one and only one pillar.
+More labels are coming.
+
+How good is the data
+--------------------
+
+Not so much, some classes are ambiguous and, since we have multiple taggers, their tags are not
+always consistent.
+However, we have a lot of data, which is good.
+
+
+Which data should I use?
+------------------------
+
+We are currently working only with `frameworks_data` and the most recent data version.
+
+We advise you to import the file:
+
+.. code-block:: python
+
+    from deep.constants import *
+
+The variable ``LATEST_DATA_PATH`` points to the most recent version of the data.
+
+What are the differences between the data versions?
+----------------------------------------------------
+
+Alongside the dataset queried (IMMAP vs all) mainly bug fixes and better definition of the classes.
+Please use the latest version.
+
+How do I get the most recent version?
+-------------------------------------
+
+We use `DVC <https://dvc.org>`_ to deal with data. Simply run
+
+.. code-block:: bash
+
+    dvc pull
+
+to get the most recent version.
+
diff --git a/docs/source/index.rst b/docs/source/index.rst
index 87f3764..ecf8b2f 100644
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -9,7 +9,7 @@
    :maxdepth: 1
    :caption: Data:
 
-   data/description
+   dfs-data/description
 
 .. toctree::
    :maxdepth: 1