diff --git a/DATASTRUCTURE.md b/DATASTRUCTURE.md
new file mode 100644
index 0000000..20d1cb1
--- /dev/null
+++ b/DATASTRUCTURE.md
@@ -0,0 +1,55 @@
+# Relations between text types and collections/DBs in BaseX
+
+The vicav app treats different TEI text types differently and they are stored in different collections.  
+Each collection can be listed which means that the teiHeader metadata is retruned as JSON. This can be consumed by a frontend to show lists an filter them.  
+By default a virtual teiCorpus document will be created from the teiHeader data in a collection.
+If there is a document containing a teiCorpus in a collection this will be used instead.  
+The idea is to have such a document whenever there are resources of that type that can only be presented as their metadata.  
+For example texts that only exist as audio recordings and were not transcribed or recordings that can not be presented due to legal and/or privacy protection reasons.
+
+## Supported TEI text types
+
+* Meta texts (About, News, etc.) -> vicav_texts
+* Bibliographic entries -> vicav_biblio
+* Language profiles -> vicav_profiles
+* Sample texts -> vicav_samples
+* Linguisitic feature descriptions -> vicav_lingfeatures
+* Text cropora -> vicav_cropus
+
+### Recommended structure for meta texts
+
+For a schema example see: 
+
+### Recommeded structure for bibliographic entries
+
+We usually collect bibliographic entries in Zotero and export them to TEI.
+
+For a schema example see: 
+
+### Recommeded structure for language profiles
+
+For a schema example see: 
+
+### Recommeded structure for sample texts
+
+For a schema example see:
+
+### Recommeded structure for linguisitic feature descriptions
+
+For a schema example see:
+
+### Recommeded structure for text cropora
+
+We use NoSketchEngine as the search engine beckend.   
+There is a workflow that takes TEI texts or ELAN files and converts them to TEI with the text tokenized.  
+Search results from NoSketchEngine are resolved to w-tags in XML files that are genereated using the above workflow that also generates the NoSketchEngine verticals.
+The xml:id attributes on any w-tag in the vicav_corpus collection needs to be unique within the collection. We therefore usually prefix the token ID with a document ID.
+
+For a schema example see:
+
+## Other collections
+
+* vicav_projects -> settings for a particular instance of the vicav-app
+* prerendered_json -> the project config including all metadata that is requested via the settings. created on build.
+* dict_users -> users for the dictionary part. See the vleserver-basex API
+* Dictionary collections -> See the vleserver-basex API