diff --git a/core/linguistics/BUILD.bazel b/core/linguistics/BUILD.bazel
index bea67e1..0bd2c5d 100644
--- a/core/linguistics/BUILD.bazel
+++ b/core/linguistics/BUILD.bazel
@@ -18,3 +18,14 @@ kt_jvm_test(
         "@maven//:junit_junit",
     ],
 )
+
+kt_jvm_test(
+    name = "sentence_scorer_test",
+    size = "small",
+    srcs = ["src/test/kotlin/com/anysoftkeyboard/calque/linguistics/SentenceScorerTest.kt"],
+    test_class = "com.anysoftkeyboard.calque.linguistics.SentenceScorerTest",
+    deps = [
+        ":linguistics",
+        "@maven//:junit_junit",
+    ],
+)
diff --git a/core/linguistics/src/main/kotlin/com/anysoftkeyboard/calque/linguistics/SentenceScorer.kt b/core/linguistics/src/main/kotlin/com/anysoftkeyboard/calque/linguistics/SentenceScorer.kt
new file mode 100644
index 0000000..9cfe284
--- /dev/null
+++ b/core/linguistics/src/main/kotlin/com/anysoftkeyboard/calque/linguistics/SentenceScorer.kt
@@ -0,0 +1,177 @@
+package com.anysoftkeyboard.calque.linguistics
+
+import java.text.BreakIterator
+import java.util.Locale
+
+/** Difficulty tier for curated reading lessons. */
+enum class SentenceLevel {
+    LEVEL_0,
+    LEVEL_1,
+    LEVEL_2,
+}
+
+/**
+ * Word properties required for sentence scoring.
+ *
+ * @property frequencyRank Subtitle frequency rank (lower number = more common, e.g. 1..1500).
+ * @property isCognate Whether this word is a transparent cognate with English.
+ * @property grammar Morphological and grammatical tags (e.g., tense, mood).
+ */
+data class ScoredWordInfo(
+    val frequencyRank: Int? = null,
+    val isCognate: Boolean = false,
+    val grammar: String? = null,
+)
+
+/** Result of scoring a Spanish sentence for lesson suitability. */
+data class SentenceScoreResult(
+    val level: SentenceLevel?,
+    val score: Double,
+    val wordCount: Int,
+    val rareWordRatio: Double,
+    val cognateRatio: Double,
+    val hasComplexTense: Boolean,
+)
+
+/**
+ * Sentence readability scorer and level bucketer.
+ *
+ * Implements Step 4 from INGESTION.md to evaluate sentence difficulty based on length, vocabulary
+ * rarity (SUBTLEX-ESP frequency), cognate transparency, and grammatical complexity.
+ */
+object SentenceScorer {
+    const val MIN_WORD_COUNT = 4
+    const val MAX_WORD_COUNT = 14
+    const val RARE_WORD_FREQ_THRESHOLD = 1500
+
+    const val LEVEL_0_MAX_SCORE = 2.0
+    const val LEVEL_1_MAX_SCORE = 4.0
+    const val LEVEL_2_MAX_SCORE = 6.5
+
+    private const val LENGTH_WEIGHT = 0.25
+    private const val RARE_RATIO_WEIGHT = 5.0
+    private const val COGNATE_RATIO_WEIGHT = 2.5
+    private const val COMPLEX_TENSE_PENALTY = 3.0
+
+    private val COMPLEX_TENSE_KEYWORDS =
+        listOf("subjunctive", "conditional", "subjuntivo", "condicional")
+
+    /**
+     * Tokenizes a Spanish sentence into lowercased word tokens using Unicode-aware BreakIterator.
+     */
+    fun tokenizeWords(sentence: String, locale: Locale = Locale("es", "ES")): List<String> {
+        val iterator = BreakIterator.getWordInstance(locale)
+        iterator.setText(sentence)
+        val words = mutableListOf<String>()
+
+        var start = iterator.first()
+        var end = iterator.next()
+        while (end != BreakIterator.DONE) {
+            val token = sentence.substring(start, end).trim()
+            if (token.isNotEmpty() && token.any { it.isLetter() }) {
+                words.add(token.lowercase(locale))
+            }
+            start = end
+            end = iterator.next()
+        }
+        return words
+    }
+
+    /**
+     * Scores a Spanish sentence using the provided word lookup function.
+     *
+     * @param sentence The raw Spanish sentence to evaluate.
+     * @param wordLookup Function providing [ScoredWordInfo] for a given lowercased word.
+     * @return [SentenceScoreResult] containing the computed score, metrics, and level (or null if
+     *   discarded).
+     */
+    fun scoreSentence(
+        sentence: String,
+        wordLookup: (String) -> ScoredWordInfo?,
+    ): SentenceScoreResult {
+        val words = tokenizeWords(sentence)
+        val wordCount = words.size
+
+        if (wordCount < MIN_WORD_COUNT || wordCount > MAX_WORD_COUNT) {
+            return SentenceScoreResult(
+                level = null,
+                score = 0.0,
+                wordCount = wordCount,
+                rareWordRatio = 0.0,
+                cognateRatio = 0.0,
+                hasComplexTense = false,
+            )
+        }
+
+        val metrics = calculateMetrics(words, wordLookup)
+        var score =
+            (wordCount * LENGTH_WEIGHT) + (metrics.rareWordRatio * RARE_RATIO_WEIGHT) -
+                (metrics.cognateRatio * COGNATE_RATIO_WEIGHT)
+
+        if (metrics.hasComplexTense) {
+            score += COMPLEX_TENSE_PENALTY
+        }
+
+        return SentenceScoreResult(
+            level = determineLevel(score),
+            score = score,
+            wordCount = wordCount,
+            rareWordRatio = metrics.rareWordRatio,
+            cognateRatio = metrics.cognateRatio,
+            hasComplexTense = metrics.hasComplexTense,
+        )
+    }
+
+    private fun calculateMetrics(
+        words: List<String>,
+        wordLookup: (String) -> ScoredWordInfo?,
+    ): IntermediateMetrics {
+        var rareWordCount = 0
+        var cognateCount = 0
+        var hasComplexTense = false
+
+        for (word in words) {
+            val info = wordLookup(word)
+            if (info == null || isRare(info.frequencyRank)) {
+                rareWordCount++
+            }
+            if (info?.isCognate == true) {
+                cognateCount++
+            }
+            if (info?.grammar != null && isComplexTense(info.grammar)) {
+                hasComplexTense = true
+            }
+        }
+
+        return IntermediateMetrics(
+            rareWordRatio = rareWordCount.toDouble() / words.size,
+            cognateRatio = cognateCount.toDouble() / words.size,
+            hasComplexTense = hasComplexTense,
+        )
+    }
+
+    private fun isRare(frequencyRank: Int?): Boolean {
+        return frequencyRank == null || frequencyRank > RARE_WORD_FREQ_THRESHOLD
+    }
+
+    private fun determineLevel(score: Double): SentenceLevel? {
+        return when {
+            score <= LEVEL_0_MAX_SCORE -> SentenceLevel.LEVEL_0
+            score <= LEVEL_1_MAX_SCORE -> SentenceLevel.LEVEL_1
+            score <= LEVEL_2_MAX_SCORE -> SentenceLevel.LEVEL_2
+            else -> null
+        }
+    }
+
+    private fun isComplexTense(grammar: String): Boolean {
+        return COMPLEX_TENSE_KEYWORDS.any { keyword ->
+            grammar.contains(keyword, ignoreCase = true)
+        }
+    }
+
+    private data class IntermediateMetrics(
+        val rareWordRatio: Double,
+        val cognateRatio: Double,
+        val hasComplexTense: Boolean,
+    )
+}
diff --git a/core/linguistics/src/test/kotlin/com/anysoftkeyboard/calque/linguistics/SentenceScorerTest.kt b/core/linguistics/src/test/kotlin/com/anysoftkeyboard/calque/linguistics/SentenceScorerTest.kt
new file mode 100644
index 0000000..3dc374b
--- /dev/null
+++ b/core/linguistics/src/test/kotlin/com/anysoftkeyboard/calque/linguistics/SentenceScorerTest.kt
@@ -0,0 +1,185 @@
+package com.anysoftkeyboard.calque.linguistics
+
+import org.junit.Assert.assertEquals
+import org.junit.Assert.assertFalse
+import org.junit.Assert.assertNull
+import org.junit.Assert.assertTrue
+import org.junit.Test
+
+class SentenceScorerTest {
+    private val delta = 0.0001
+
+    @Test
+    fun tokenizeWords_handlesSpanishPunctuationAndInvertedMarks() {
+        val sentence = "¡Hola! ¿Cómo estás, amigo mío?"
+        val tokens = SentenceScorer.tokenizeWords(sentence)
+
+        assertEquals(listOf("hola", "cómo", "estás", "amigo", "mío"), tokens)
+    }
+
+    @Test
+    fun scoreSentence_discardsSentencesWithFewerThanFourWords() {
+        val shortSentence = "Hola, mi amigo."
+        val result =
+            SentenceScorer.scoreSentence(shortSentence) {
+                ScoredWordInfo(frequencyRank = 100, isCognate = false, grammar = null)
+            }
+
+        assertEquals(3, result.wordCount)
+        assertNull(result.level)
+    }
+
+    @Test
+    fun scoreSentence_discardsSentencesWithMoreThanFourteenWords() {
+        val longSentence =
+            "Uno dos tres cuatro cinco seis siete ocho nueve diez once doce trece catorce quince."
+        val result =
+            SentenceScorer.scoreSentence(longSentence) {
+                ScoredWordInfo(frequencyRank = 100, isCognate = false, grammar = null)
+            }
+
+        assertEquals(15, result.wordCount)
+        assertNull(result.level)
+    }
+
+    @Test
+    fun scoreSentence_assignsLevel0ForSimpleCognateSentences() {
+        // 4 words, all common (freq rank <= 1500), 2 cognates, no complex tense:
+        // score = 4 * 0.25 + 0 * 5.0 - (2/4) * 2.5 = 1.0 - 1.25 = -0.25 <= 2.0 -> LEVEL_0
+        val sentence = "La música es fantástica."
+        val dict =
+            mapOf(
+                "la" to ScoredWordInfo(frequencyRank = 5, isCognate = false),
+                "música" to ScoredWordInfo(frequencyRank = 500, isCognate = true),
+                "es" to ScoredWordInfo(frequencyRank = 10, isCognate = false),
+                "fantástica" to ScoredWordInfo(frequencyRank = 1200, isCognate = true),
+            )
+
+        val result = SentenceScorer.scoreSentence(sentence) { dict[it] }
+
+        assertEquals(4, result.wordCount)
+        assertEquals(SentenceLevel.LEVEL_0, result.level)
+        assertEquals(-0.25, result.score, delta)
+        assertEquals(0.0, result.rareWordRatio, delta)
+        assertEquals(0.5, result.cognateRatio, delta)
+        assertFalse(result.hasComplexTense)
+    }
+
+    @Test
+    fun scoreSentence_assignsLevel1ForModerateVocabulary() {
+        // 8 words, 3 rare words (rank > 1500), 1 cognate:
+        // rare ratio = 3/8 = 0.375
+        // cognate ratio = 1/8 = 0.125
+        // score = (8 * 0.25) + (0.375 * 5.0) - (0.125 * 2.5) = 2.0 + 1.875 - 0.3125 = 3.5625 <= 4.0
+        // -> LEVEL_1
+        val words = listOf("uno", "dos", "tres", "cuatro", "cinco", "seis", "siete", "ocho")
+        val sentence = words.joinToString(" ") + "."
+        val dict =
+            mapOf(
+                "uno" to ScoredWordInfo(frequencyRank = 2000, isCognate = false),
+                "dos" to ScoredWordInfo(frequencyRank = 3000, isCognate = false),
+                "tres" to ScoredWordInfo(frequencyRank = 4000, isCognate = false),
+                "cuatro" to ScoredWordInfo(frequencyRank = 100, isCognate = true),
+                "cinco" to ScoredWordInfo(frequencyRank = 200, isCognate = false),
+                "seis" to ScoredWordInfo(frequencyRank = 300, isCognate = false),
+                "siete" to ScoredWordInfo(frequencyRank = 400, isCognate = false),
+                "ocho" to ScoredWordInfo(frequencyRank = 500, isCognate = false),
+            )
+
+        val result = SentenceScorer.scoreSentence(sentence) { dict[it] }
+
+        assertEquals(8, result.wordCount)
+        assertEquals(SentenceLevel.LEVEL_1, result.level)
+        assertEquals(3.5625, result.score, delta)
+    }
+
+    @Test
+    fun scoreSentence_assignsLevel2ForAdvancedVocabularyWithoutComplexTense() {
+        // 10 words, 5 rare (0.5), 0 cognates:
+        // score = (10 * 0.25) + (0.5 * 5.0) - 0 = 2.5 + 2.5 = 5.0 <= 6.5 -> LEVEL_2
+        val words = (1..10).map { "palabra$it" }
+        val sentence = words.joinToString(" ") + "."
+        val dict =
+            words
+                .mapIndexed { index, word ->
+                    word to
+                        ScoredWordInfo(
+                            frequencyRank = if (index < 5) 5000 else 100,
+                            isCognate = false,
+                        )
+                }
+                .toMap()
+
+        val result = SentenceScorer.scoreSentence(sentence) { dict[it] }
+
+        assertEquals(10, result.wordCount)
+        assertEquals(SentenceLevel.LEVEL_2, result.level)
+        assertEquals(5.0, result.score, delta)
+    }
+
+    @Test
+    fun scoreSentence_addsPenaltyForSubjunctiveOrConditionalTense() {
+        // 5 words, 0 rare, 0 cognates:
+        // base score = 5 * 0.25 = 1.25 (would be LEVEL_0)
+        // with subjunctive: + 3.0 = 4.25 -> LEVEL_2 (since 4.0 < 4.25 <= 6.5)
+        val sentence = "Espero que vengas pronto mañana."
+        val dict =
+            mapOf(
+                "espero" to ScoredWordInfo(frequencyRank = 50, isCognate = false),
+                "que" to ScoredWordInfo(frequencyRank = 2, isCognate = false),
+                "vengas" to
+                    ScoredWordInfo(
+                        frequencyRank = 100,
+                        isCognate = false,
+                        grammar = "present subjunctive",
+                    ),
+                "pronto" to ScoredWordInfo(frequencyRank = 80, isCognate = false),
+                "mañana" to ScoredWordInfo(frequencyRank = 60, isCognate = false),
+            )
+
+        val result = SentenceScorer.scoreSentence(sentence) { dict[it] }
+
+        assertEquals(5, result.wordCount)
+        assertTrue(result.hasComplexTense)
+        assertEquals(4.25, result.score, delta)
+        assertEquals(SentenceLevel.LEVEL_2, result.level)
+    }
+
+    @Test
+    fun scoreSentence_discardsSentencesWithScoreExceedingMaxLevel() {
+        // 14 words, all rare, no cognates, with complex tense:
+        // score = 14 * 0.25 + 1.0 * 5.0 - 0 + 3.0 = 3.5 + 5.0 + 3.0 = 11.5 > 6.5 -> DISCARD
+        val words = (1..14).map { "rara$it" }
+        val sentence = words.joinToString(" ") + "."
+        val dict =
+            words
+                .mapIndexed { index, word ->
+                    word to
+                        ScoredWordInfo(
+                            frequencyRank = 10000,
+                            isCognate = false,
+                            grammar = if (index == 0) "conditional" else null,
+                        )
+                }
+                .toMap()
+
+        val result = SentenceScorer.scoreSentence(sentence) { dict[it] }
+
+        assertNull(result.level)
+        assertEquals(11.5, result.score, delta)
+    }
+
+    @Test
+    fun scoreSentence_treatsUnrecognizedWordsAsRare() {
+        // Words missing from lookup should default to rare
+        val sentence = "Palabra desconocida misteriosa extraña."
+        val result = SentenceScorer.scoreSentence(sentence) { null }
+
+        assertEquals(4, result.wordCount)
+        assertEquals(1.0, result.rareWordRatio, delta)
+        assertEquals(0.0, result.cognateRatio, delta)
+        // score = 4 * 0.25 + 1.0 * 5.0 = 6.0 <= 6.5 -> LEVEL_2
+        assertEquals(6.0, result.score, delta)
+        assertEquals(SentenceLevel.LEVEL_2, result.level)
+    }
+}