diff --git a/reasoning_gym/arithmetic/power_function.py b/reasoning_gym/arithmetic/power_function.py
index dc9e10f8..c5db17a0 100644
--- a/reasoning_gym/arithmetic/power_function.py
+++ b/reasoning_gym/arithmetic/power_function.py
@@ -1,4 +1,4 @@
-"""Computhe the power of a number."""
+"""Compute the power of a number."""
 
 from dataclasses import dataclass
 from decimal import Decimal
@@ -37,22 +37,26 @@ class PowerFunctionDataset(ProceduralDataset):
     def __init__(self, config: PowerFunctionConfig):
         super().__init__(config=config, seed=config.seed, size=config.size)
 
+    def _format_sig_figs(self, x: Decimal, sig: int) -> Decimal:
+        """Format a Decimal to exactly 'sig' significant figures, keeping trailing zeros."""
+        if x.is_zero():
+            return "0." + "0" * (sig - 1)
+
+        exp = x.adjusted()
+        shift = sig - exp - 1
+        rounded = x.quantize(Decimal("1e{}".format(-shift)))
+        return Decimal(rounded)
+
     def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
         """Score the answer by checking if it matches the expected answer to 3 significant figures."""
         oracle_answer = entry["answer"]
         if answer is not None:
             try:
-                user_answer = Decimal(answer)
-                oracle_value = Decimal(oracle_answer)
-
-                if oracle_value == 0:
-                    return 1.0 if user_answer == 0 else 0.01
-
-                user_sig_figs = f"{user_answer:.3g}"
-                oracle_sig_figs = f"{oracle_value:.3g}"
+                user_answer = self._format_sig_figs(Decimal(answer), 3)
+                oracle_answer = self._format_sig_figs(Decimal(oracle_answer), 3)
 
                 # Check if they match to 3 significant figures
-                if user_sig_figs == oracle_sig_figs:
+                if user_answer == oracle_answer:
                     return 1.0
                 else:
                     return 0.01
diff --git a/tests/test_power_function.py b/tests/test_power_function.py
index 22c5327e..f56a9ce9 100644
--- a/tests/test_power_function.py
+++ b/tests/test_power_function.py
@@ -1,5 +1,7 @@
 """Tests for Power Function questions generation"""
 
+from decimal import Decimal
+
 import pytest
 
 from reasoning_gym.arithmetic import PowerFunctionConfig, PowerFunctionDataset
@@ -82,3 +84,46 @@ def test_power_function_curriculum():
 
     increased_cfg = curriculum.generate_configuration(base_value)
     assert increased_cfg.min_exponent == 2 and increased_cfg.max_exponent == 4
+
+
+# Test score_answer function with various answers
+def test_power_function_score_answer_for_edge_cases():
+    """Test score_answer function for edge cases"""
+    config = PowerFunctionConfig(seed=42)
+    dataset = PowerFunctionDataset(config)
+
+    # Case 1: Match with trailing zeros
+    item = dataset[0].copy()
+    user_answer = "1.000e+00"
+    # Let's change the oracle answer for edge case testing
+    item["answer"] = "1.0"
+    score = dataset.score_answer(user_answer, item)
+    assert score == 1.0, f"Expected score 1.0, got {score}"
+
+    # Case 2: Rounding up at edge of significant figures
+    item = dataset[0].copy()
+    item["answer"] = str(Decimal("0.9999") ** 1)  # Close to 1.000
+    user_answer = "1.00"
+    score = dataset.score_answer(user_answer, item)
+    assert score == 1.0, f"Expected score 1.0, got {score}"
+
+    # Case 3: Negative base, valid exponent
+    item = dataset[0].copy()
+    item["answer"] = str(Decimal("-2.00") ** 3)  # -8.0
+    user_answer = "-8.00"
+    score = dataset.score_answer(user_answer, item)
+    assert score == 1.0, f"Expected score 1.0, got {score}"
+
+    # Case 4: Very small number with exponent notation
+    item = dataset[0].copy()
+    item["answer"] = str(Decimal("1e-6"))  # 1e-6
+    user_answer = "1.00e-6"
+    score = dataset.score_answer(user_answer, item)
+    assert score == 1.0, f"Expected score 1.0, got {score}"
+
+    # Case 5: Incorrect answer should yield low score
+    item = dataset[0].copy()
+    item["answer"] = "1000.0"
+    user_answer = "999.0"
+    score = dataset.score_answer(user_answer, item)
+    assert score == 0.01, f"Expected low score 0.01, got {score}"