diff --git a/README.md b/README.md
index 2ac0236..0a0b648 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,19 @@
 # gptc_scratch
 
-GPTC, ported to Scratch
\ No newline at end of file
+gptc_scratch uses the same algorithm as GPTC, in a Scratch project. However, it
+only supports two categories, and they are numbered (1 and 2) rather than
+named.
+
+Models are formatted differently than in the Python version of GPTC. A model
+consists of two files, each containing the text for a category. gptc_scratch
+does not have a tokenizer, so each line must consist of one lowercase word with
+no other characters. To load the model, right-click "Category 1 Words" in the
+project, click "import," then choose one of the files. Do the same thing again
+with "Category 2 Words" and the other file.
+
+The text to classify must be stored in the same format as the model files.
+Follow the same procedure that you used for loading the model files, but with
+"Words in text to classify" and the file containing the text to classify. Then,
+click "Compile." If the button changes to a lighter color, wait for it to
+change back. Next, click "classify." This should update the "Cat1" and "Cat2"
+variables with the confidence valued for the given text and categories.
diff --git a/gptc_scratch.sb3 b/gptc_scratch.sb3
index 3f29c68..bf437b2 100644
Binary files a/gptc_scratch.sb3 and b/gptc_scratch.sb3 differ