diff --git a/FuzzySearch.Test/FuzzyTests.cs b/FuzzySearch.Test/FuzzyTests.cs
index 178a0ae..e31c606 100644
--- a/FuzzySearch.Test/FuzzyTests.cs
+++ b/FuzzySearch.Test/FuzzyTests.cs
@@ -449,6 +449,76 @@ public void CalculateScore_EmptyPattern_ZeroScoreAndPatternPresent()
 		Assert.AreEqual(0, score);      // No characters to match, so score is 0
 	}
 
+	[TestMethod]
+	public void CalculateScore_LongUnmatchedPrefix_PenaltyFlattensAtTheCap()
+	{
+		// Arrange: the same single-character match, preceded by ever more junk
+		string pattern = "y";
+
+		// Act
+		int atCap = Fuzzy.CalculateScore("xxxxxy", pattern, out bool atCapPresent);
+		int pastCap = Fuzzy.CalculateScore("xxxxxxxxxxxxy", pattern, out bool pastCapPresent);
+		int wellPastCap = Fuzzy.CalculateScore(new string('x', 100) + "y", pattern, out bool wellPastCapPresent);
+
+		// Assert: all still match, and the prefix cost stops growing once the cap is reached
+		Assert.IsTrue(atCapPresent, "The pattern is present regardless of the prefix length.");
+		Assert.IsTrue(pastCapPresent, "The pattern is present regardless of the prefix length.");
+		Assert.IsTrue(wellPastCapPresent, "The pattern is present regardless of the prefix length.");
+
+		Assert.AreEqual(atCap, pastCap, "A prefix beyond the cap threshold must not cost any more than one at it.");
+		Assert.AreEqual(atCap, wellPastCap, "The prefix penalty must stay flat however long the prefix grows.");
+	}
+
+	[TestMethod]
+	public void CalculateScore_PrefixPenalty_NeverExceedsTheDocumentedCap()
+	{
+		// Arrange: a single-character pattern matched at the very end earns no bonuses, so the prefix
+		// penalty is the only thing moving the score and the cap is directly observable as a floor
+		string pattern = "y";
+
+		// Act & Assert
+		for (int prefixLength = 1; prefixLength <= 20; prefixLength++)
+		{
+			int score = Fuzzy.CalculateScore(new string('x', prefixLength) + "y", pattern, out _);
+
+			Assert.IsGreaterThanOrEqualTo(Fuzzy.maxPrefixPenalty, score,
+				$"A prefix of {prefixLength} characters scored {score}, beyond the documented cap of {Fuzzy.maxPrefixPenalty}.");
+		}
+	}
+
+	[TestMethod]
+	public void CalculateScore_LongerPrefix_NeverScoresHigherThanAShorterOne()
+	{
+		// Arrange
+		string pattern = "y";
+		int previous = Fuzzy.CalculateScore("y", pattern, out _);
+
+		// Act & Assert: the penalty is monotonic — a longer prefix is never rewarded, only flattened
+		for (int prefixLength = 1; prefixLength <= 20; prefixLength++)
+		{
+			int score = Fuzzy.CalculateScore(new string('x', prefixLength) + "y", pattern, out _);
+			Assert.IsLessThanOrEqualTo(previous, score,
+				$"A prefix of {prefixLength} characters scored higher than the shorter prefix before it.");
+			previous = score;
+		}
+	}
+
+	[TestMethod]
+	public void CalculateScore_NonMatchingSubject_StillReflectsHowMuchWasSkipped()
+	{
+		// Arrange: the pattern appears in neither subject, so the prefix refund must not apply
+		string pattern = "y";
+
+		// Act
+		int shortSubject = Fuzzy.CalculateScore("xxx", pattern, out bool shortPresent);
+		int longSubject = Fuzzy.CalculateScore("xxxxxxxxxxxx", pattern, out bool longPresent);
+
+		// Assert
+		Assert.IsFalse(shortPresent, "The pattern is not present in a subject without it.");
+		Assert.IsFalse(longPresent, "The pattern is not present in a subject without it.");
+		Assert.IsLessThan(shortSubject, longSubject, "A longer non-matching subject should still score lower.");
+	}
+
 	#endregion
 
 	#region Unicode Normalization Tests
diff --git a/FuzzySearch/Fuzzy.cs b/FuzzySearch/Fuzzy.cs
index 95f13b4..b62af59 100644
--- a/FuzzySearch/Fuzzy.cs
+++ b/FuzzySearch/Fuzzy.cs
@@ -42,7 +42,11 @@ public static class Fuzzy
 	/// <summary>The penalty for each unmatched character at the beginning of the string.</summary>
 	internal const int unmatchedPrefixLetterPenalty = -1;
 
-	/// <summary>The maximum prefix penalty that can be applied.</summary>
+	/// <summary>
+	/// The maximum prefix penalty that can be applied. Once the unmatched prefix is this expensive, a longer
+	/// prefix costs no more, so a match preceded by a long irrelevant prefix is deprioritized rather than
+	/// punished without bound.
+	/// </summary>
 	internal const int maxPrefixPenalty = -5;
 
 	/// <summary>The penalty for each unmatched character in the string.</summary>
@@ -184,6 +188,12 @@ internal static int CalculateScoreCore(ReadOnlySpan<char> subject, ReadOnlySpan<
 		int bestLetterLength = 0;
 		int bestLetterScore = 0;
 
+		// The generic per-codepoint penalty below also charges the codepoints before the first match. That is
+		// refunded when the first match lands, so PenalizeNonPatternCharacters stays the sole — and therefore
+		// capped — source of prefix cost. Without the refund the two stack and maxPrefixPenalty bounds only one
+		// of them, leaving the score falling without bound as the prefix grows.
+		int prefixPenaltyCharged = 0;
+
 		// Loop over codepoints in subject
 		while (strIdx != strLength)
 		{
@@ -214,6 +224,14 @@ internal static int CalculateScoreCore(ReadOnlySpan<char> subject, ReadOnlySpan<
 			{
 				int newScore = 0;
 
+				if (patternIdx == 0)
+				{
+					// First pattern codepoint matched: hand back the uncapped per-codepoint cost of the prefix,
+					// so that the capped penalty applied next is all the prefix is charged.
+					score -= prefixPenaltyCharged;
+					prefixPenaltyCharged = 0;
+				}
+
 				score = PenalizeNonPatternCharacters(score, patternIdx, strCodepointIdx);
 
 				newScore = ApplyBonuses(prevMatched, prevLower, prevSeparator, strChar, strLower, strUpper, newScore);
@@ -243,6 +261,15 @@ internal static int CalculateScoreCore(ReadOnlySpan<char> subject, ReadOnlySpan<
 			else
 			{
 				score += unmatchedLetterPenalty;
+
+				if (patternIdx == 0)
+				{
+					// Still before the first match, so this is prefix cost. Remember it for the refund above.
+					// A subject that never matches the pattern keeps these penalties, so its score still
+					// reflects how much was skipped.
+					prefixPenaltyCharged += unmatchedLetterPenalty;
+				}
+
 				prevMatched = false;
 			}