diff --git a/README.md b/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..fd12b9637469a5961c6214921748950af86983f4
--- /dev/null
+++ b/README.md
@@ -0,0 +1,52 @@
+# MCC experiments
+
+## Usage
+
+install the dependencies and run
+
+```
+./mcc.py test.nt 500
+```
+
+## Notes
+
+### 20/11/24
+
+#### Effect of independent attributes
+
+Consider the following example:
+
+D^*
+```
+a na c1
+a na c2
+a na c3
+```
+
+and the join probability P(b1|a) = 2/3 and P(b2|a) = 1/3. So, the value of B is independent from the attribute C.
+
+There is a rewriting for the following query : 
+
+```sql 
+SELECT B FROM T WHERE A=a
+```
+
+but there is no rewriting with a condition on the independent attribute, e.g.
+
+```sql
+SELECT B FROM T WHERE C=c2
+```
+
+#### Remarks about the BID
+
+- It allows to represent together the part of the database that is certain of $D^*$ with the part is known only under some probabilities.
+- For example by projecting out the null values on one world of the BID, the results is the same as on $D^*$. It is not the case when by doing so on a database generated from the join distribution. 
+- Using the BID only, it is impossible to distinguish the queries that can be rewritten from the others.
+
+#### Open question
+
+Given a MG and $D^*$ is there a class $C$ such that : 
+
+1. the distance between the join distribution and the empirical distribution of $C$ is minimal
+2. the probability of the class $C$ is strictly positive
+