diff --git a/doc/Changelog.md b/doc/Changelog.md
index 60a6a82..aee9f7b 100644
--- a/doc/Changelog.md
+++ b/doc/Changelog.md
@@ -4,7 +4,7 @@ Changelog {#Changelog}
 # git master
 
 * [130](https://github.com/BlueBrain/Brion/pull/130):
-  Added brain::CompartmentReportReader and its Python wrapping.
+  Added brain::CompartmentReportReader and its Python wrapping
 * [120](https://github.com/BlueBrain/Brion/pull/120), [131](https://github.com/BlueBrain/Brion/pull/131):
   - New SpikeReport API.
   - Reimplementation of the high level SpikeReportReader. The new implementation
@@ -25,15 +25,15 @@ Changelog {#Changelog}
 * [117](https://github.com/BlueBrain/Brion/pull/117):
   Changes and fixes in Brain Python module:
   - Functions that take gids now preserve the iteration order of the input in
-    the output arrays/lists.
-  - Replaced Python sets with numpy arrays in functions returning GID sets.
-  - Circuit::getXYZNames functions renamed to Circuit::getXYZTypeNames.
+    the output arrays/lists
+  - Replaced Python sets with numpy arrays in functions returning GID sets
+  - Circuit::getXYZNames functions renamed to Circuit::getXYZTypeNames
 * [113](https://github.com/BlueBrain/Brion/pull/113):
-  Support for old circuits containing only synapse center positions.
+  Support for old circuits containing only synapse center positions
 * [110](https://github.com/BlueBrain/Brion/pull/110):
   Break PersistentMap out into keyv::Map
 * [107](https://github.com/BlueBrain/Brion/pull/107):
-  Added Sphinx generated documentation of the brain python module.
+  Added Sphinx generated documentation of the brain python module
 * [102](https://github.com/BlueBrain/Brion/pull/102):
   Use PersistentMap for caching synapse positions loaded from brain::Circuit
 * [94](https://github.com/BlueBrain/Brion/pull/94):
@@ -41,11 +41,11 @@ Changelog {#Changelog}
   creating invalid soma sections when the first order sections where connected
   to arbitrary soma sample points.
 * [89](https://github.com/BlueBrain/Brion/pull/89):
-  Python wrapping of brain classes.
+  Python wrapping of brain classes
 * [88](https://github.com/BlueBrain/Brion/pull/88):
-  - Brain namespace enums made strongly typed.
+  - Brain namespace enums made strongly typed
   - Fix for brain::Section::getSamples for sections where the first point
-    appears repeated.
+    appears repeated
 * [85](https://github.com/BlueBrain/Brion/pull/85):
   Implement brain synapses specification
 * [84](https://github.com/BlueBrain/Brion/pull/84):
diff --git a/doc/feature/compartmentReportPy.md b/doc/feature/compartmentReportPy.md
new file mode 100644
index 0000000..d49051b
--- /dev/null
+++ b/doc/feature/compartmentReportPy.md
@@ -0,0 +1,86 @@
+High-Level CompartmentReport API for Python {#compartment_report_py}
+============
+
+## Motivation
+
+Redesign the current BBPSDK-inspired API for also allowing voltage traces.
+The returned data should change to be a matrix[frame, compartment] instead
+of the current vector[compartments].
+
+## API
+
+### C++
+
+* rename brain::CompartmentReportReader to CompartmentReport
+* merge brion::CompartmentReport::loadFrame and loadNeuron in load( startTime, endTime )
+  * returns matrix in frame-major order
+* Change all timestamps to double
+
+### Python
+
+    class CompartmentReportView:
+        def load()
+            """Load all the available data from this view.
+
+            If the report is empty return None, otherwise return a tuple with a 1D
+            double ndarray for the frame timestamps and a 2D float ndarray, where
+            the first index is for frames and the second for compartment offsets."""
+
+        def load(start, end):
+            """Load all frames in a time range.
+
+            This function will wait if necessary until the requested time window becomes
+            available. The return value is the same as for load()."""
+
+        def load(timestamp):
+            """Load the frame that contains the given timestamp.
+
+            The caller will block until the frame on which the requested timestamp falls
+            is available.
+            The return type is a (double, nadarray) tuple where the first element is the
+            actual timestamp of the frame and the second one is a 1D double ndarray."""
+
+## Example
+
+    import brain
+
+    # voltage trace:
+    reader = brain.CompartmentReport("uri")
+    view = reader.create_view([1])
+    times, trace = view.load()
+    for time, frame in zip(times, trace):
+        print "{}, {}".format(time, frame[soma_index(42)])
+
+    # full frame of sub-target
+    view = reader.create_view(range(1, 1000))
+    time, frame = view.load( 42.f )
+    for gid in range(1, 1000):
+        print "{}, {}".format(gid, frame[soma_index(gid)])
+
+    def soma_index(gid):
+        return view.mapping.offsets[mapping.gids.index(gid)][0]
+
+## Issues
+
+### 1: Do we need separate functions for frame-based and trace-based access?
+
+_Resolution: No_
+
+The current CRView::load( start, end ) API already allows selecting a single
+Neuron and loading all frames. The brion API will need to change (see above).
+
+### 2: Do we provide column-major memory layout?
+
+_Resolution: No_
+
+We will not reformat the data layout to be column-major, i.e., packing all
+samples of a single compartment together. This is an optimization and we do
+not expect today to load more than a few neurons for voltage traces.
+
+### 3: Should we change the time stamps to double?
+
+_Resolution: Yes_
+
+Reasons: Not memory critical since it's not stored much. Python uses double too,
+and with plasticity simulations this becomes critical. This will create some
+intermediate inconsistency with other classes.
diff --git a/doc/feature/feature.md b/doc/feature/feature.md
index 2365018..cb91898 100644
--- a/doc/feature/feature.md
+++ b/doc/feature/feature.md
@@ -3,6 +3,7 @@ RFCs {#rfcs}
 
 * @subpage compartment_report
 * @subpage compartment_report_mpi
+* @subpage compartment_report_py
 * @subpage morphology11
 * @subpage spike_reports
 * @subpage brain_synapses