feat(emotion): reshape the feelings board and expand to 8 targets (#36)

Redraw the home board's grid (slowly→yikes, the old "!" tile → little) and
grow the collectible set from 3 to 8 emotions: happy, elated, irked, sarcastic,
distraught, furious, betrayed, surprised. The judge's candidate labels track
the targets.

Update the solution-path tests that referenced the removed "!" tile and refresh
the judge live-eval's board sentences for the same reason. All level invariants
still hold — grid rectangular, start blank, every target scoreable, solvable
within budget.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

levels/emotion.py

@@ -3,7 +3,7 @@
 Grid geometry note: a 4x4 grid graph is bipartite, so start/⏎ placement
 fixes sentence-length parity. With start at (1,1) and ⏎ at (2,3) (Manhattan
 distance 3), walks have odd hop counts and sentences even word counts -- which
-is what makes "not sad" and "great !" reachable.
+is what makes a two-word read like "not sad" reachable.
 """
 
 from levels._base import Level
@@ -12,14 +12,14 @@ LEVEL = Level(
     id="emotion",
     title="feelings",
     grid=[
-        ["slowly", "very", "fast", "annoy"],
-        ["yikes", "start", "great", "!"],
+        ["yikes", "very", "fast", "annoy"],
+        ["little", "start", "great", "little"],
         ["hurt", "not", "sad", "⏎"],
         ["again", "?", "angry", "disturb"],
     ],
     start=(1, 1),  # row, col of the start tile
-    targets=["happy", "betrayed", "surprised"],  # collect them all
-    labels=["happy", "betrayed", "surprised"],
+    targets=["happy", "elated", "irked", "sarcastic", "distraught", "furious", "betrayed", "surprised"],  # collect them all
+    labels=["happy", "elated", "irked", "sarcastic", "distraught", "furious", "betrayed", "surprised"],
     budget=10,
     order=0,
     home=True,

tests/test_judge.py

@@ -29,12 +29,10 @@ def test_renormalize_handles_large_negative_logprobs():
 # --- Live eval (slow, needs the Modal judge endpoint) -----------------------
 
 BOARD_SENTENCES = [
-    ("great!", "happy", True),
-    ("very great!", "happy", True),
     ("not sad", "happy", True),
     ("hurt", "betrayed", True),
     ("yikes", "surprised", True),
-    ("great!", "betrayed", False),
+    ("not sad", "betrayed", False),
     ("yikes", "happy", False),
 ]
 

tests/test_level.py

@@ -123,7 +123,7 @@ def emotion_walk(path):
 
 
 def test_emotion_quick_win_path():
-    assert assemble_sentence(emotion_walk(["great", "!", "⏎"])) == "great!"
+    assert assemble_sentence(emotion_walk(["great", "sad", "⏎"])) == "great sad"
 
 
 def test_emotion_clever_win_path():
@@ -131,7 +131,7 @@ def test_emotion_clever_win_path():
 
 
 def test_emotion_start_revisit_win_path():
-    assert assemble_sentence(emotion_walk(["very", "start", "great", "!", "⏎"])) == "very great!"
+    assert assemble_sentence(emotion_walk(["very", "start", "great", "sad", "⏎"])) == "very great sad"
 
 
 # ---- animal board: empty tiles are walkable but wordless -------------------