diff --git a/Proj2.hs b/Proj2.hs
index ff9f91e..6d4c01c 100644
--- a/Proj2.hs
+++ b/Proj2.hs
@@ -54,7 +54,7 @@ import Debug.Trace
 
 -- ==== DATA STRUCTURES =======================================================
 
--- TODO: INSERT COMMENT HERE
+-- contains possible guesses. a possible guess is consistent with all current -- and previous feedback given by the composer
 --
 type GameState = [[Pitch]]
 
@@ -155,10 +155,19 @@ initialGuess = (chord, chords)
   where
     chord : chords = allChords
 
--- TODO: implement me
+-- takes in the previous guess, the game state, and the feedback for the
+-- previous guess and outputs the next guess and a (reduced in size) game state
+--
+-- strategy:
+--  1. reduce the size of the search space by removing all guesses inconsistent
+--     with the answer received for the previous guess.
+--  2. TODO: mini-max?
 --
 nextGuess :: ([Pitch], GameState) -> (Int, Int, Int) -> ([Pitch], GameState)
-nextGuess (prevGuess, chord : chords) _ = (chord, chords)
+nextGuess (prevGuess, chords) answer = (guess, consistentChords)
+  where
+    guess : consistentChords = filter consistentWithAnswer chords
+    consistentWithAnswer chord = answer == feedback prevGuess chord
 
 -- ==== HELPER FUNCTIONS ======================================================
 
diff --git a/README.md b/README.md
index 3809ffd..1490f0c 100644
--- a/README.md
+++ b/README.md
@@ -1,18 +1,5 @@
 # Spec
 ## Tips to improve `nextGuess`
-A better approach would be to only make guesses that are consistent with the
-answers you have received for previous guesses. You can do this by computing
-the list of possible targets, and removing elements that are inconsistent with
-any answers you have received to previous guesses. A possible target is
-inconsistent with an answer you have received for a previous guess if the
-answer you would receive for that guess and that (possible) target is different
-from the answer you actually received for that guess.
-
-You can use your GameState type to store your previous guesses and the
-corresponding answers. Or, more efficient and just as easy, store the list of
-remaining possible targets in your GameState, and pare it down each time you
-receive feedback for a guess.
-
 The best results can be had by carefully choosing each guess so that it is most
 likely to leave a small remaining list of possible targets. You can do this by
 computing for each remaining possible target the maximum number of possible