diff --git a/README.txt b/README.txt
new file mode 100644
index 0000000..7c15d55
--- /dev/null
+++ b/README.txt
@@ -0,0 +1,50 @@
+Seismic-2d
+==========
+This is a simplification of a model originally used for global-scale
+seismology.
+
+NET-VISA: Network Processing Vertically Integrated Seismic
+Analysis. Nimar S. Arora, Stuart Russell, Erik Sudderth. Bulletin of the
+Seismological Society of America (BSSA) April 2013, vol. 103 no. 2A
+pp709-729.
+
+Files
+=====
+
+description.odt -- a description of the model
+generate.py   -- generates the physics of a 2-D world and some episodes
+util.py       -- some geophysical utility functions
+solve.py      -- a sample solver that learns the physics and solves the episodes
+evaluate.py   -- evaluates a solution versus a reference
+mwmatching.py -- utility script for max-weight max cardinality matching
+training.data -- 10K episodes for training
+test.data     -- another 10K episodes for testing
+test.blind    -- the test data with the event-to-detection mapping omitted 
+test.solution -- the sample solution on the test data
+
+Overview
+========
+
+The model is completely described in ```description.odt``` and this should be
+translated in the Probabilistic Programming Language of your
+choosing. The unlabeled data in ```test.blind``` (and optionally the labeled
+data in ```training.data```) comprises the observations to the model. The
+query of interest is the seismic bulletin for each of the observed
+episodes.
+
+Once all the bulletins have been produced, the script ```evaluate.py```
+can be used to produce the reports on the accuracy versus the reference
+script ```test.data```. One can also compare the results versus the
+baseline in ```test.solution```.
+
+The files ```generate.py``` and ```solve.py``` have only been provided
+for convenience they shouldn't normally be used. However, if you want to
+check the performance of your model on more than just the provided data
+you may generate more as needed. The sample solver is based loosely on
+the published greedy algorithm, and may be used as a competitive
+baseline.
+
+Authors
+=======
+Nimar S. Arora, Bayesian Logic Inc., nimar.arora@gmail.com
+Stuart Russell, Deptt. of Computer Science, Berkeley.