From fa68138c15f30e0b6dafd72864c8e86d7b94015b Mon Sep 17 00:00:00 2001
From: genki <123@1234.com>
Date: Wed, 21 Jan 2026 15:59:41 -0500
Subject: [PATCH] discover dez

---
 .idea/deploymentTargetSelector.xml            |   2 +-
 .../sherpai2/data/local/AppDatabase.kt        |  55 +-
 .../sherpai2/data/local/dao/Facecachedao.kt   | 287 ++++++-
 .../data/local/dao/Userfeedbackdao.kt         | 212 +++++
 .../data/local/entity/Userfeedbackentity.kt   | 161 ++++
 .../placeholder/sherpai2/di/DatabaseModule.kt |  11 +-
 .../sherpai2/di/RepositoryModule.kt           |  39 +-
 .../clustering/Clusterrefinementservice.kt    | 415 +++++++++
 .../clustering/Faceclusteringservice.kt       | 479 +++++++----
 .../domain/clustering/Facequalityfilter.kt    | 193 +++++
 .../clustering/Temporalclusteringservice.kt   | 597 +++++++++++++
 .../ui/discover/Discoverpeoplescreen.kt       | 159 +++-
 .../ui/discover/Discoverpeopleviewmodel.kt    | 164 +++-
 .../ui/discover/Temporalnamingdialog.kt       | 353 ++++++++
 .../ui/discover/Validationpreviewscreen.kt    | 784 +++++++++++-------
 15 files changed, 3402 insertions(+), 509 deletions(-)
 create mode 100644 app/src/main/java/com/placeholder/sherpai2/data/local/dao/Userfeedbackdao.kt
 create mode 100644 app/src/main/java/com/placeholder/sherpai2/data/local/entity/Userfeedbackentity.kt
 create mode 100644 app/src/main/java/com/placeholder/sherpai2/domain/clustering/Clusterrefinementservice.kt
 create mode 100644 app/src/main/java/com/placeholder/sherpai2/domain/clustering/Facequalityfilter.kt
 create mode 100644 app/src/main/java/com/placeholder/sherpai2/domain/clustering/Temporalclusteringservice.kt
 create mode 100644 app/src/main/java/com/placeholder/sherpai2/ui/discover/Temporalnamingdialog.kt

diff --git a/.idea/deploymentTargetSelector.xml b/.idea/deploymentTargetSelector.xml
index 627d767..92cafd5 100644
--- a/.idea/deploymentTargetSelector.xml
+++ b/.idea/deploymentTargetSelector.xml
@@ -4,7 +4,7 @@
     <selectionStates>
       <SelectionState runConfigName="app">
         <option name="selectionMode" value="DROPDOWN" />
-        <DropdownSelection timestamp="2026-01-18T23:43:22.974426869Z">
+        <DropdownSelection timestamp="2026-01-21T20:49:28.305260931Z">
           <Target type="DEFAULT_BOOT">
             <handle>
               <DeviceId pluginId="LocalEmulator" identifier="path=/home/genki/.android/avd/Medium_Phone.avd" />
diff --git a/app/src/main/java/com/placeholder/sherpai2/data/local/AppDatabase.kt b/app/src/main/java/com/placeholder/sherpai2/data/local/AppDatabase.kt
index f7f4355..a128c5c 100644
--- a/app/src/main/java/com/placeholder/sherpai2/data/local/AppDatabase.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/data/local/AppDatabase.kt
@@ -10,6 +10,11 @@ import com.placeholder.sherpai2.data.local.entity.*
 /**
  * AppDatabase - Complete database for SherpAI2
  *
+ * VERSION 10 - User Feedback Loop
+ * - Added UserFeedbackEntity for storing user corrections
+ * - Enables cluster refinement before training
+ * - Ground truth data for improving clustering
+ *
  * VERSION 9 - Enhanced Face Cache
  * - Added FaceCacheEntity for per-face metadata
  * - Stores quality scores, embeddings, bounding boxes
@@ -38,14 +43,15 @@ import com.placeholder.sherpai2.data.local.entity.*
         FaceModelEntity::class,
         PhotoFaceTagEntity::class,
         PersonAgeTagEntity::class,
-        FaceCacheEntity::class,          // NEW: Per-face metadata cache
+        FaceCacheEntity::class,
+        UserFeedbackEntity::class,      // NEW: User corrections
 
         // ===== COLLECTIONS =====
         CollectionEntity::class,
         CollectionImageEntity::class,
         CollectionFilterEntity::class
     ],
-    version = 9,  // INCREMENTED for face cache
+    version = 10,  // INCREMENTED for user feedback
     exportSchema = false
 )
 abstract class AppDatabase : RoomDatabase() {
@@ -63,7 +69,8 @@ abstract class AppDatabase : RoomDatabase() {
     abstract fun faceModelDao(): FaceModelDao
     abstract fun photoFaceTagDao(): PhotoFaceTagDao
     abstract fun personAgeTagDao(): PersonAgeTagDao
-    abstract fun faceCacheDao(): FaceCacheDao  // NEW
+    abstract fun faceCacheDao(): FaceCacheDao
+    abstract fun userFeedbackDao(): UserFeedbackDao  // NEW
 
     // ===== COLLECTIONS DAO =====
     abstract fun collectionDao(): CollectionDao
@@ -185,6 +192,10 @@ val MIGRATION_8_9 = object : Migration(8, 9) {
                 hasGoodLighting INTEGER NOT NULL,
                 embedding TEXT,
                 confidence REAL NOT NULL,
+                imageWidth INTEGER NOT NULL DEFAULT 0,
+                imageHeight INTEGER NOT NULL DEFAULT 0,
+                cacheVersion INTEGER NOT NULL DEFAULT 1,
+                cachedAt INTEGER NOT NULL DEFAULT 0,
                 PRIMARY KEY(imageId, faceIndex),
                 FOREIGN KEY(imageId) REFERENCES images(imageId) ON DELETE CASCADE
             )
@@ -197,13 +208,47 @@ val MIGRATION_8_9 = object : Migration(8, 9) {
     }
 }
 
+/**
+ * MIGRATION 9 → 10 (User Feedback Loop)
+ *
+ * Changes:
+ * 1. Create user_feedback table for storing user corrections
+ */
+val MIGRATION_9_10 = object : Migration(9, 10) {
+    override fun migrate(database: SupportSQLiteDatabase) {
+
+        // Create user_feedback table
+        database.execSQL("""
+            CREATE TABLE IF NOT EXISTS user_feedback (
+                id TEXT PRIMARY KEY NOT NULL,
+                imageId TEXT NOT NULL,
+                faceIndex INTEGER NOT NULL,
+                clusterId INTEGER,
+                personId TEXT,
+                feedbackType TEXT NOT NULL,
+                originalConfidence REAL NOT NULL,
+                userNote TEXT,
+                timestamp INTEGER NOT NULL,
+                FOREIGN KEY(imageId) REFERENCES images(imageId) ON DELETE CASCADE,
+                FOREIGN KEY(personId) REFERENCES persons(id) ON DELETE CASCADE
+            )
+        """)
+
+        // Create indices for fast lookups
+        database.execSQL("CREATE INDEX IF NOT EXISTS index_user_feedback_imageId ON user_feedback(imageId)")
+        database.execSQL("CREATE INDEX IF NOT EXISTS index_user_feedback_clusterId ON user_feedback(clusterId)")
+        database.execSQL("CREATE INDEX IF NOT EXISTS index_user_feedback_personId ON user_feedback(personId)")
+        database.execSQL("CREATE INDEX IF NOT EXISTS index_user_feedback_feedbackType ON user_feedback(feedbackType)")
+    }
+}
+
 /**
  * PRODUCTION MIGRATION NOTES:
  *
- * Before shipping to users, update DatabaseModule to use migration:
+ * Before shipping to users, update DatabaseModule to use migrations:
  *
  * Room.databaseBuilder(context, AppDatabase::class.java, "sherpai.db")
- *     .addMigrations(MIGRATION_7_8)  // Add this
+ *     .addMigrations(MIGRATION_7_8, MIGRATION_8_9, MIGRATION_9_10)  // Add all migrations
  *     // .fallbackToDestructiveMigration()  // Remove this
  *     .build()
  */
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/data/local/dao/Facecachedao.kt b/app/src/main/java/com/placeholder/sherpai2/data/local/dao/Facecachedao.kt
index ea6cfc2..e4a0f2d 100644
--- a/app/src/main/java/com/placeholder/sherpai2/data/local/dao/Facecachedao.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/data/local/dao/Facecachedao.kt
@@ -8,7 +8,21 @@ import androidx.room.Update
 import com.placeholder.sherpai2.data.local.entity.FaceCacheEntity
 
 /**
- * FaceCacheDao - Face detection cache with NEW queries for two-stage clustering
+ * FaceCacheDao - YEAR-BASED filtering for temporal clustering
+ *
+ * NEW STRATEGY: Cluster by YEAR to handle children
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * Problem: Child's face changes dramatically over years
+ * Solution: Cluster each YEAR separately
+ *
+ * Example:
+ * - 2020 photos → Emma age 2
+ * - 2021 photos → Emma age 3
+ * - 2022 photos → Emma age 4
+ *
+ * Result: Multiple clusters of same child at different ages
+ * User names: "Emma, Age 2", "Emma, Age 3", etc.
+ * System creates: Emma_Age_2, Emma_Age_3 submodels
  */
 @Dao
 interface FaceCacheDao {
@@ -27,17 +41,161 @@ interface FaceCacheDao {
     suspend fun update(faceCache: FaceCacheEntity)
 
     // ═══════════════════════════════════════
-    // NEW CLUSTERING QUERIES ⭐
+    // YEAR-BASED QUERIES (NEW - For Children)
     // ═══════════════════════════════════════
 
     /**
-     * Get high-quality solo faces for Stage 1 clustering
+     * Get premium solo faces from a SPECIFIC YEAR
      *
-     * Filters:
-     * - Solo photos (faceCount = 1)
-     * - Large faces (faceAreaRatio >= minFaceRatio)
-     * - Has embedding
+     * Use Case: Cluster children by age
+     * - Cluster 2020 photos separately from 2021 photos
+     * - Same child at different ages = different clusters
+     * - User names each: "Emma Age 2", "Emma Age 3"
+     *
+     * @param year Year in YYYY format (e.g., "2020")
+     * @param minRatio Minimum face size (default 5%)
+     * @param minQuality Minimum quality score (default 0.8)
+     * @param limit Maximum faces to return
      */
+    @Query("""
+        SELECT fc.* 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.qualityScore >= :minQuality
+          AND fc.embedding IS NOT NULL
+          AND strftime('%Y', i.capturedAt/1000, 'unixepoch') = :year
+        ORDER BY fc.qualityScore DESC, fc.faceAreaRatio DESC
+        LIMIT :limit
+    """)
+    suspend fun getPremiumSoloFacesByYear(
+        year: String,
+        minRatio: Float = 0.05f,
+        minQuality: Float = 0.8f,
+        limit: Int = 1000
+    ): List<FaceCacheEntity>
+
+    /**
+     * Get premium solo faces from a YEAR RANGE
+     *
+     * Use Case: Cluster adults who don't change much
+     * - Photos from 2018-2023 can cluster together
+     * - Adults look similar across years
+     *
+     * @param startYear Start year in YYYY format
+     * @param endYear End year in YYYY format
+     */
+    @Query("""
+        SELECT fc.* 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.qualityScore >= :minQuality
+          AND fc.embedding IS NOT NULL
+          AND strftime('%Y', i.capturedAt/1000, 'unixepoch') BETWEEN :startYear AND :endYear
+        ORDER BY fc.qualityScore DESC, fc.faceAreaRatio DESC
+        LIMIT :limit
+    """)
+    suspend fun getPremiumSoloFacesByYearRange(
+        startYear: String,
+        endYear: String,
+        minRatio: Float = 0.05f,
+        minQuality: Float = 0.8f,
+        limit: Int = 1000
+    ): List<FaceCacheEntity>
+
+    /**
+     * Get years that have sufficient photos for clustering
+     *
+     * Returns years with at least N solo photos
+     * Use to determine which years to cluster
+     *
+     * Example output:
+     * ```
+     * [
+     *   YearPhotoCount(year="2020", photoCount=150),
+     *   YearPhotoCount(year="2021", photoCount=200),
+     *   YearPhotoCount(year="2022", photoCount=180)
+     * ]
+     * ```
+     */
+    @Query("""
+        SELECT 
+            strftime('%Y', i.capturedAt/1000, 'unixepoch') as year,
+            COUNT(DISTINCT fc.imageId) as photoCount
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.embedding IS NOT NULL
+        GROUP BY year
+        HAVING photoCount >= :minPhotos
+        ORDER BY year ASC
+    """)
+    suspend fun getYearsWithSufficientPhotos(
+        minPhotos: Int = 20,
+        minRatio: Float = 0.03f
+    ): List<YearPhotoCount>
+
+    /**
+     * Get month-by-month breakdown for a year
+     *
+     * For fine-grained age clustering (babies change monthly)
+     */
+    @Query("""
+        SELECT 
+            strftime('%Y-%m', i.capturedAt/1000, 'unixepoch') as yearMonth,
+            COUNT(DISTINCT fc.imageId) as photoCount
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.embedding IS NOT NULL
+          AND strftime('%Y', i.capturedAt/1000, 'unixepoch') = :year
+        GROUP BY yearMonth
+        ORDER BY yearMonth ASC
+    """)
+    suspend fun getMonthlyBreakdownForYear(year: String): List<MonthPhotoCount>
+
+    // ═══════════════════════════════════════
+    // STANDARD QUERIES (Original)
+    // ═══════════════════════════════════════
+
+    @Query("""
+        SELECT fc.* 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.qualityScore >= :minQuality
+          AND fc.embedding IS NOT NULL
+        ORDER BY fc.qualityScore DESC, fc.faceAreaRatio DESC
+        LIMIT :limit
+    """)
+    suspend fun getPremiumSoloFaces(
+        minRatio: Float = 0.05f,
+        minQuality: Float = 0.8f,
+        limit: Int = 1000
+    ): List<FaceCacheEntity>
+
+    @Query("""
+        SELECT fc.* 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.qualityScore >= :minQuality
+          AND fc.embedding IS NOT NULL
+        ORDER BY fc.qualityScore DESC, fc.faceAreaRatio DESC
+        LIMIT :limit
+    """)
+    suspend fun getStandardSoloFaces(
+        minRatio: Float = 0.03f,
+        minQuality: Float = 0.6f,
+        limit: Int = 2000
+    ): List<FaceCacheEntity>
+
     @Query("""
         SELECT fc.* 
         FROM face_cache fc
@@ -53,10 +211,6 @@ interface FaceCacheDao {
         limit: Int = 2000
     ): List<FaceCacheEntity>
 
-    /**
-     * FALLBACK: Get ANY solo faces with embeddings
-     * Used if getHighQualitySoloFaces() returns empty
-     */
     @Query("""
         SELECT fc.* 
         FROM face_cache fc
@@ -70,12 +224,90 @@ interface FaceCacheDao {
         limit: Int = 2000
     ): List<FaceCacheEntity>
 
-    // ═══════════════════════════════════════
-    // EXISTING QUERIES (keep as-is)
-    // ═══════════════════════════════════════
+    @Query("""
+        SELECT fc.* 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount BETWEEN :minFaces AND :maxFaces
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.embedding IS NOT NULL
+        ORDER BY i.faceCount ASC, fc.faceAreaRatio DESC
+    """)
+    suspend fun getSmallGroupFaces(
+        minFaces: Int = 2,
+        maxFaces: Int = 5,
+        minRatio: Float = 0.02f
+    ): List<FaceCacheEntity>
 
-    @Query("SELECT * FROM face_cache WHERE id = :id")
-    suspend fun getFaceCacheById(id: String): FaceCacheEntity?
+    @Query("""
+        SELECT fc.* 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = :faceCount
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.embedding IS NOT NULL
+        ORDER BY fc.qualityScore DESC
+    """)
+    suspend fun getFacesByGroupSize(
+        faceCount: Int,
+        minRatio: Float = 0.02f
+    ): List<FaceCacheEntity>
+
+    @Query("""
+        SELECT fc.* 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.embedding IS NOT NULL
+          AND fc.imageId NOT IN (:excludedImageIds)
+        ORDER BY fc.qualityScore DESC
+        LIMIT :limit
+    """)
+    suspend fun getSoloFacesExcluding(
+        excludedImageIds: List<String>,
+        minRatio: Float = 0.03f,
+        limit: Int = 2000
+    ): List<FaceCacheEntity>
+
+    @Query("""
+        SELECT 
+            i.faceCount,
+            COUNT(DISTINCT i.imageId) as imageCount,
+            AVG(fc.faceAreaRatio) as avgFaceSize,
+            AVG(fc.qualityScore) as avgQuality,
+            COUNT(fc.embedding IS NOT NULL) as hasEmbedding
+        FROM images i
+        LEFT JOIN face_cache fc ON i.imageId = fc.imageId
+        WHERE i.hasFaces = 1
+        GROUP BY i.faceCount
+        ORDER BY i.faceCount ASC
+    """)
+    suspend fun getLibraryQualityDistribution(): List<LibraryQualityStat>
+
+    @Query("""
+        SELECT COUNT(*) 
+        FROM face_cache fc
+        INNER JOIN images i ON fc.imageId = i.imageId
+        WHERE i.faceCount = 1 
+          AND fc.faceAreaRatio >= :minRatio
+          AND fc.qualityScore >= :minQuality
+          AND fc.embedding IS NOT NULL
+    """)
+    suspend fun countPremiumSoloFaces(
+        minRatio: Float = 0.05f,
+        minQuality: Float = 0.8f
+    ): Int
+
+    @Query("""
+        SELECT COUNT(*) 
+        FROM face_cache 
+        WHERE embedding IS NOT NULL
+    """)
+    suspend fun countFacesWithEmbeddings(): Int
+
+    @Query("SELECT * FROM face_cache WHERE imageId = :imageId AND faceIndex = :faceIndex")
+    suspend fun getFaceCacheByKey(imageId: String, faceIndex: Int): FaceCacheEntity?
 
     @Query("SELECT * FROM face_cache WHERE imageId = :imageId ORDER BY faceIndex")
     suspend fun getFaceCacheForImage(imageId: String): List<FaceCacheEntity>
@@ -88,4 +320,25 @@ interface FaceCacheDao {
 
     @Query("DELETE FROM face_cache WHERE cacheVersion < :version")
     suspend fun deleteOldVersions(version: Int)
-}
\ No newline at end of file
+}
+
+/**
+ * Result classes for year-based queries
+ */
+data class YearPhotoCount(
+    val year: String,
+    val photoCount: Int
+)
+
+data class MonthPhotoCount(
+    val yearMonth: String,  // "2020-05"
+    val photoCount: Int
+)
+
+data class LibraryQualityStat(
+    val faceCount: Int,
+    val imageCount: Int,
+    val avgFaceSize: Float,
+    val avgQuality: Float,
+    val hasEmbedding: Int
+)
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/data/local/dao/Userfeedbackdao.kt b/app/src/main/java/com/placeholder/sherpai2/data/local/dao/Userfeedbackdao.kt
new file mode 100644
index 0000000..680e68b
--- /dev/null
+++ b/app/src/main/java/com/placeholder/sherpai2/data/local/dao/Userfeedbackdao.kt
@@ -0,0 +1,212 @@
+package com.placeholder.sherpai2.data.local.dao
+
+import androidx.room.*
+import com.placeholder.sherpai2.data.local.entity.FeedbackType
+import com.placeholder.sherpai2.data.local.entity.UserFeedbackEntity
+import kotlinx.coroutines.flow.Flow
+
+/**
+ * UserFeedbackDao - Query user corrections and feedback
+ *
+ * KEY QUERIES:
+ * - Get feedback for cluster validation
+ * - Find rejected faces to exclude from training
+ * - Track feedback statistics for quality metrics
+ * - Support cluster refinement workflow
+ */
+@Dao
+interface UserFeedbackDao {
+
+    // ═══════════════════════════════════════
+    // INSERT / UPDATE
+    // ═══════════════════════════════════════
+
+    @Insert(onConflict = OnConflictStrategy.REPLACE)
+    suspend fun insert(feedback: UserFeedbackEntity): Long
+
+    @Insert(onConflict = OnConflictStrategy.REPLACE)
+    suspend fun insertAll(feedbacks: List<UserFeedbackEntity>)
+
+    @Update
+    suspend fun update(feedback: UserFeedbackEntity)
+
+    @Delete
+    suspend fun delete(feedback: UserFeedbackEntity)
+
+    // ═══════════════════════════════════════
+    // CLUSTER VALIDATION QUERIES
+    // ═══════════════════════════════════════
+
+    /**
+     * Get all feedback for a cluster
+     * Used during validation to see what user has reviewed
+     */
+    @Query("SELECT * FROM user_feedback WHERE clusterId = :clusterId ORDER BY timestamp DESC")
+    suspend fun getFeedbackForCluster(clusterId: Int): List<UserFeedbackEntity>
+
+    /**
+     * Get rejected faces for a cluster
+     * These faces should be EXCLUDED from training
+     */
+    @Query("""
+        SELECT * FROM user_feedback 
+        WHERE clusterId = :clusterId 
+        AND feedbackType = 'REJECTED_MATCH'
+    """)
+    suspend fun getRejectedFacesForCluster(clusterId: Int): List<UserFeedbackEntity>
+
+    /**
+     * Get confirmed faces for a cluster
+     * These faces are SAFE for training
+     */
+    @Query("""
+        SELECT * FROM user_feedback 
+        WHERE clusterId = :clusterId 
+        AND feedbackType = 'CONFIRMED_MATCH'
+    """)
+    suspend fun getConfirmedFacesForCluster(clusterId: Int): List<UserFeedbackEntity>
+
+    /**
+     * Count feedback by type for a cluster
+     * Used to show stats: "15 confirmed, 3 rejected"
+     */
+    @Query("""
+        SELECT feedbackType, COUNT(*) as count
+        FROM user_feedback
+        WHERE clusterId = :clusterId
+        GROUP BY feedbackType
+    """)
+    suspend fun getFeedbackStatsByCluster(clusterId: Int): List<FeedbackStat>
+
+    // ═══════════════════════════════════════
+    // PERSON FEEDBACK QUERIES
+    // ═══════════════════════════════════════
+
+    /**
+     * Get all feedback for a person
+     * Used to show history of corrections
+     */
+    @Query("SELECT * FROM user_feedback WHERE personId = :personId ORDER BY timestamp DESC")
+    suspend fun getFeedbackForPerson(personId: String): List<UserFeedbackEntity>
+
+    /**
+     * Get rejected faces for a person
+     * User said "this is NOT X" - exclude from model improvement
+     */
+    @Query("""
+        SELECT * FROM user_feedback 
+        WHERE personId = :personId 
+        AND feedbackType = 'REJECTED_MATCH'
+    """)
+    suspend fun getRejectedFacesForPerson(personId: String): List<UserFeedbackEntity>
+
+    /**
+     * Flow version for reactive UI
+     */
+    @Query("SELECT * FROM user_feedback WHERE personId = :personId ORDER BY timestamp DESC")
+    fun observeFeedbackForPerson(personId: String): Flow<List<UserFeedbackEntity>>
+
+    // ═══════════════════════════════════════
+    // IMAGE QUERIES
+    // ═══════════════════════════════════════
+
+    /**
+     * Get feedback for a specific image
+     */
+    @Query("SELECT * FROM user_feedback WHERE imageId = :imageId")
+    suspend fun getFeedbackForImage(imageId: String): List<UserFeedbackEntity>
+
+    /**
+     * Check if user has provided feedback for a specific face
+     */
+    @Query("""
+        SELECT EXISTS(
+            SELECT 1 FROM user_feedback 
+            WHERE imageId = :imageId 
+            AND faceIndex = :faceIndex
+        )
+    """)
+    suspend fun hasFeedbackForFace(imageId: String, faceIndex: Int): Boolean
+
+    // ═══════════════════════════════════════
+    // STATISTICS & ANALYTICS
+    // ═══════════════════════════════════════
+
+    /**
+     * Get total feedback count
+     */
+    @Query("SELECT COUNT(*) FROM user_feedback")
+    suspend fun getTotalFeedbackCount(): Int
+
+    /**
+     * Get feedback count by type (global)
+     */
+    @Query("""
+        SELECT feedbackType, COUNT(*) as count
+        FROM user_feedback
+        GROUP BY feedbackType
+    """)
+    suspend fun getGlobalFeedbackStats(): List<FeedbackStat>
+
+    /**
+     * Get average original confidence for rejected faces
+     * Helps identify if low confidence → more rejections
+     */
+    @Query("""
+        SELECT AVG(originalConfidence) 
+        FROM user_feedback 
+        WHERE feedbackType = 'REJECTED_MATCH'
+    """)
+    suspend fun getAverageConfidenceForRejectedFaces(): Float?
+
+    /**
+     * Find faces with low confidence that were confirmed
+     * These are "surprising successes" - model worked despite low confidence
+     */
+    @Query("""
+        SELECT * FROM user_feedback 
+        WHERE feedbackType = 'CONFIRMED_MATCH' 
+        AND originalConfidence < :threshold
+        ORDER BY originalConfidence ASC
+    """)
+    suspend fun getLowConfidenceSuccesses(threshold: Float = 0.7f): List<UserFeedbackEntity>
+
+    // ═══════════════════════════════════════
+    // CLEANUP
+    // ═══════════════════════════════════════
+
+    /**
+     * Delete all feedback for a cluster
+     * Called when cluster is deleted or refined
+     */
+    @Query("DELETE FROM user_feedback WHERE clusterId = :clusterId")
+    suspend fun deleteFeedbackForCluster(clusterId: Int)
+
+    /**
+     * Delete all feedback for a person
+     * Called when person is deleted
+     */
+    @Query("DELETE FROM user_feedback WHERE personId = :personId")
+    suspend fun deleteFeedbackForPerson(personId: String)
+
+    /**
+     * Delete old feedback (older than X days)
+     * Keep database size manageable
+     */
+    @Query("DELETE FROM user_feedback WHERE timestamp < :cutoffTimestamp")
+    suspend fun deleteOldFeedback(cutoffTimestamp: Long)
+
+    /**
+     * Clear all feedback (nuclear option)
+     */
+    @Query("DELETE FROM user_feedback")
+    suspend fun deleteAll()
+}
+
+/**
+ * Result class for feedback statistics
+ */
+data class FeedbackStat(
+    val feedbackType: String,
+    val count: Int
+)
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/data/local/entity/Userfeedbackentity.kt b/app/src/main/java/com/placeholder/sherpai2/data/local/entity/Userfeedbackentity.kt
new file mode 100644
index 0000000..e48313e
--- /dev/null
+++ b/app/src/main/java/com/placeholder/sherpai2/data/local/entity/Userfeedbackentity.kt
@@ -0,0 +1,161 @@
+package com.placeholder.sherpai2.data.local.entity
+
+import androidx.room.Entity
+import androidx.room.ForeignKey
+import androidx.room.Index
+import androidx.room.PrimaryKey
+import java.util.UUID
+
+/**
+ * UserFeedbackEntity - Stores user corrections during cluster validation
+ *
+ * PURPOSE:
+ * - Capture which faces user marked as correct/incorrect
+ * - Ground truth data for improving clustering
+ * - Enable cluster refinement before training
+ * - Track confidence in automated detections
+ *
+ * USAGE FLOW:
+ * 1. Clustering creates initial clusters
+ * 2. User reviews ValidationPreview
+ * 3. User swipes faces: ✅ Correct / ❌ Incorrect
+ * 4. Feedback stored here
+ * 5. If too many incorrect → Re-cluster without those faces
+ * 6. If approved → Train model with confirmed faces only
+ *
+ * INDEXES:
+ * - imageId: Fast lookup of feedback for specific images
+ * - clusterId: Get all feedback for a cluster
+ * - feedbackType: Filter by correction type
+ * - personId: Track feedback after person created
+ */
+@Entity(
+    tableName = "user_feedback",
+    foreignKeys = [
+        ForeignKey(
+            entity = ImageEntity::class,
+            parentColumns = ["imageId"],
+            childColumns = ["imageId"],
+            onDelete = ForeignKey.CASCADE
+        ),
+        ForeignKey(
+            entity = PersonEntity::class,
+            parentColumns = ["id"],
+            childColumns = ["personId"],
+            onDelete = ForeignKey.CASCADE
+        )
+    ],
+    indices = [
+        Index(value = ["imageId"]),
+        Index(value = ["clusterId"]),
+        Index(value = ["personId"]),
+        Index(value = ["feedbackType"])
+    ]
+)
+data class UserFeedbackEntity(
+    @PrimaryKey
+    val id: String = UUID.randomUUID().toString(),
+
+    /**
+     * Image containing the face
+     */
+    val imageId: String,
+
+    /**
+     * Face index within the image (0-based)
+     * Multiple faces per image possible
+     */
+    val faceIndex: Int,
+
+    /**
+     * Cluster ID from clustering (before person created)
+     * Null if feedback given after person exists
+     */
+    val clusterId: Int?,
+
+    /**
+     * Person ID if feedback is about an existing person
+     * Null during initial cluster validation
+     */
+    val personId: String?,
+
+    /**
+     * Type of feedback user provided
+     */
+    val feedbackType: String, // FeedbackType enum stored as string
+
+    /**
+     * Confidence score that led to this face being shown
+     * Helps identify if low confidence = more errors
+     */
+    val originalConfidence: Float,
+
+    /**
+     * Optional user note
+     */
+    val userNote: String? = null,
+
+    /**
+     * When feedback was provided
+     */
+    val timestamp: Long = System.currentTimeMillis()
+) {
+    companion object {
+        fun create(
+            imageId: String,
+            faceIndex: Int,
+            clusterId: Int? = null,
+            personId: String? = null,
+            feedbackType: FeedbackType,
+            originalConfidence: Float,
+            userNote: String? = null
+        ): UserFeedbackEntity {
+            return UserFeedbackEntity(
+                imageId = imageId,
+                faceIndex = faceIndex,
+                clusterId = clusterId,
+                personId = personId,
+                feedbackType = feedbackType.name,
+                originalConfidence = originalConfidence,
+                userNote = userNote
+            )
+        }
+    }
+
+    fun getFeedbackType(): FeedbackType {
+        return try {
+            FeedbackType.valueOf(feedbackType)
+        } catch (e: Exception) {
+            FeedbackType.UNCERTAIN
+        }
+    }
+}
+
+/**
+ * FeedbackType - Types of user corrections
+ */
+enum class FeedbackType {
+    /**
+     * User confirmed this face IS the person
+     * Boosts confidence, use for training
+     */
+    CONFIRMED_MATCH,
+
+    /**
+     * User said this face is NOT the person
+     * Remove from cluster, exclude from training
+     */
+    REJECTED_MATCH,
+
+    /**
+     * User marked as outlier during cluster review
+     * Face doesn't belong in this cluster
+     */
+    MARKED_OUTLIER,
+
+    /**
+     * User is uncertain
+     * Skip this face for training, revisit later
+     */
+    UNCERTAIN
+}
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/di/DatabaseModule.kt b/app/src/main/java/com/placeholder/sherpai2/di/DatabaseModule.kt
index 936f3cf..021e088 100644
--- a/app/src/main/java/com/placeholder/sherpai2/di/DatabaseModule.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/di/DatabaseModule.kt
@@ -5,6 +5,7 @@ import androidx.room.Room
 import com.placeholder.sherpai2.data.local.AppDatabase
 import com.placeholder.sherpai2.data.local.MIGRATION_7_8
 import com.placeholder.sherpai2.data.local.MIGRATION_8_9
+import com.placeholder.sherpai2.data.local.MIGRATION_9_10
 import com.placeholder.sherpai2.data.local.dao.*
 import dagger.Module
 import dagger.Provides
@@ -16,6 +17,10 @@ import javax.inject.Singleton
 /**
  * DatabaseModule - Provides database and ALL DAOs
  *
+ * VERSION 10 UPDATES:
+ * - Added UserFeedbackDao for cluster refinement
+ * - Added MIGRATION_9_10
+ *
  * VERSION 9 UPDATES:
  * - Added FaceCacheDao for per-face metadata
  * - Added MIGRATION_8_9
@@ -44,7 +49,7 @@ object DatabaseModule {
             .fallbackToDestructiveMigration(dropAllTables = true)
 
             // PRODUCTION MODE: Uncomment this and remove fallbackToDestructiveMigration()
-            // .addMigrations(MIGRATION_7_8, MIGRATION_8_9)
+            // .addMigrations(MIGRATION_7_8, MIGRATION_8_9, MIGRATION_9_10)
 
             .build()
 
@@ -96,6 +101,10 @@ object DatabaseModule {
     fun provideFaceCacheDao(db: AppDatabase): FaceCacheDao =
         db.faceCacheDao()
 
+    @Provides
+    fun provideUserFeedbackDao(db: AppDatabase): UserFeedbackDao =
+        db.userFeedbackDao()
+
     // ===== COLLECTIONS DAOs =====
 
     @Provides
diff --git a/app/src/main/java/com/placeholder/sherpai2/di/RepositoryModule.kt b/app/src/main/java/com/placeholder/sherpai2/di/RepositoryModule.kt
index 7999d93..b57fc34 100644
--- a/app/src/main/java/com/placeholder/sherpai2/di/RepositoryModule.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/di/RepositoryModule.kt
@@ -2,12 +2,11 @@ package com.placeholder.sherpai2.di
 
 import android.content.Context
 import androidx.work.WorkManager
-import com.placeholder.sherpai2.data.local.dao.FaceModelDao
-import com.placeholder.sherpai2.data.local.dao.ImageDao
-import com.placeholder.sherpai2.data.local.dao.PersonDao
-import com.placeholder.sherpai2.data.local.dao.PhotoFaceTagDao
+import com.placeholder.sherpai2.data.local.dao.*
 import com.placeholder.sherpai2.data.repository.FaceRecognitionRepository
 import com.placeholder.sherpai2.data.repository.TaggingRepositoryImpl
+import com.placeholder.sherpai2.domain.clustering.ClusterQualityAnalyzer
+import com.placeholder.sherpai2.domain.clustering.ClusterRefinementService
 import com.placeholder.sherpai2.domain.repository.ImageRepository
 import com.placeholder.sherpai2.domain.repository.ImageRepositoryImpl
 import com.placeholder.sherpai2.domain.repository.TaggingRepository
@@ -26,6 +25,8 @@ import javax.inject.Singleton
  * UPDATED TO INCLUDE:
  * - FaceRecognitionRepository for face recognition operations
  * - ValidationScanService for post-training validation
+ * - ClusterRefinementService for user feedback loop (NEW)
+ * - ClusterQualityAnalyzer for cluster validation
  * - WorkManager for background tasks
  */
 @Module
@@ -72,7 +73,7 @@ abstract class RepositoryModule {
         }
 
         /**
-         * Provide ValidationScanService (NEW)
+         * Provide ValidationScanService
          */
         @Provides
         @Singleton
@@ -88,6 +89,34 @@ abstract class RepositoryModule {
             )
         }
 
+        /**
+         * Provide ClusterRefinementService (NEW)
+         * Handles user feedback and cluster refinement workflow
+         */
+        @Provides
+        @Singleton
+        fun provideClusterRefinementService(
+            faceCacheDao: FaceCacheDao,
+            userFeedbackDao: UserFeedbackDao,
+            qualityAnalyzer: ClusterQualityAnalyzer
+        ): ClusterRefinementService {
+            return ClusterRefinementService(
+                faceCacheDao = faceCacheDao,
+                userFeedbackDao = userFeedbackDao,
+                qualityAnalyzer = qualityAnalyzer
+            )
+        }
+
+        /**
+         * Provide ClusterQualityAnalyzer
+         * Validates cluster quality before training
+         */
+        @Provides
+        @Singleton
+        fun provideClusterQualityAnalyzer(): ClusterQualityAnalyzer {
+            return ClusterQualityAnalyzer()
+        }
+
         /**
          * Provide WorkManager for background tasks
          */
diff --git a/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Clusterrefinementservice.kt b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Clusterrefinementservice.kt
new file mode 100644
index 0000000..c6f5f5c
--- /dev/null
+++ b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Clusterrefinementservice.kt
@@ -0,0 +1,415 @@
+package com.placeholder.sherpai2.domain.clustering
+
+import android.util.Log
+import com.placeholder.sherpai2.data.local.dao.FaceCacheDao
+import com.placeholder.sherpai2.data.local.dao.UserFeedbackDao
+import com.placeholder.sherpai2.data.local.entity.FeedbackType
+import com.placeholder.sherpai2.data.local.entity.UserFeedbackEntity
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.withContext
+import javax.inject.Inject
+import javax.inject.Singleton
+import kotlin.math.sqrt
+
+/**
+ * ClusterRefinementService - Handle user feedback and cluster refinement
+ *
+ * PURPOSE:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * Close the feedback loop between user corrections and clustering
+ *
+ * WORKFLOW:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * 1. Clustering produces initial clusters
+ * 2. User reviews in ValidationPreview
+ * 3. User marks faces: ✅ Correct / ❌ Incorrect / ❓ Uncertain
+ * 4. If too many incorrect → Call refineCluster()
+ * 5. Re-cluster WITHOUT incorrect faces
+ * 6. Show updated validation preview
+ * 7. Repeat until user approves
+ *
+ * BENEFITS:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * - Prevents bad models from being created
+ * - Learns from user corrections
+ * - Iterative improvement
+ * - Ground truth data for future enhancements
+ */
+@Singleton
+class ClusterRefinementService @Inject constructor(
+    private val faceCacheDao: FaceCacheDao,
+    private val userFeedbackDao: UserFeedbackDao,
+    private val qualityAnalyzer: ClusterQualityAnalyzer
+) {
+
+    companion object {
+        private const val TAG = "ClusterRefinement"
+
+        // Thresholds for refinement decisions
+        private const val MIN_REJECTION_RATIO = 0.15f  // 15% rejected → refine
+        private const val MIN_CONFIRMED_FACES = 6      // Need at least 6 good faces
+        private const val MAX_REFINEMENT_ITERATIONS = 3 // Prevent infinite loops
+    }
+
+    /**
+     * Store user feedback for faces in a cluster
+     *
+     * @param cluster The cluster being reviewed
+     * @param feedbackMap Map of face index → feedback type
+     * @param originalConfidences Map of face index → original detection confidence
+     * @return Number of feedback items stored
+     */
+    suspend fun storeFeedback(
+        cluster: FaceCluster,
+        feedbackMap: Map<DetectedFaceWithEmbedding, FeedbackType>,
+        originalConfidences: Map<DetectedFaceWithEmbedding, Float> = emptyMap()
+    ): Int = withContext(Dispatchers.IO) {
+
+        val feedbackEntities = feedbackMap.map { (face, feedbackType) ->
+            UserFeedbackEntity.create(
+                imageId = face.imageId,
+                faceIndex = 0, // We don't track faceIndex in DetectedFaceWithEmbedding yet
+                clusterId = cluster.clusterId,
+                personId = null, // Not created yet
+                feedbackType = feedbackType,
+                originalConfidence = originalConfidences[face] ?: face.confidence
+            )
+        }
+
+        userFeedbackDao.insertAll(feedbackEntities)
+
+        Log.d(TAG, "Stored ${feedbackEntities.size} feedback items for cluster ${cluster.clusterId}")
+        feedbackEntities.size
+    }
+
+    /**
+     * Check if cluster needs refinement based on user feedback
+     *
+     * Criteria:
+     * - Too many rejected faces (> 15%)
+     * - Too few confirmed faces (< 6)
+     * - High rejection rate for cluster suggests mixed identities
+     *
+     * @return RefinementRecommendation with action and reason
+     */
+    suspend fun shouldRefineCluster(
+        cluster: FaceCluster
+    ): RefinementRecommendation = withContext(Dispatchers.Default) {
+
+        val feedback = withContext(Dispatchers.IO) {
+            userFeedbackDao.getFeedbackForCluster(cluster.clusterId)
+        }
+
+        if (feedback.isEmpty()) {
+            return@withContext RefinementRecommendation(
+                shouldRefine = false,
+                reason = "No feedback provided yet"
+            )
+        }
+
+        val totalFeedback = feedback.size
+        val rejectedCount = feedback.count { it.getFeedbackType() == FeedbackType.REJECTED_MATCH }
+        val confirmedCount = feedback.count { it.getFeedbackType() == FeedbackType.CONFIRMED_MATCH }
+        val uncertainCount = feedback.count { it.getFeedbackType() == FeedbackType.UNCERTAIN }
+
+        val rejectionRatio = rejectedCount.toFloat() / totalFeedback.toFloat()
+
+        Log.d(TAG, "Cluster ${cluster.clusterId} feedback: " +
+                "$confirmedCount confirmed, $rejectedCount rejected, $uncertainCount uncertain")
+
+        // Check 1: Too many rejections
+        if (rejectionRatio > MIN_REJECTION_RATIO) {
+            return@withContext RefinementRecommendation(
+                shouldRefine = true,
+                reason = "High rejection rate (${(rejectionRatio * 100).toInt()}%) suggests mixed identities",
+                confirmedCount = confirmedCount,
+                rejectedCount = rejectedCount,
+                uncertainCount = uncertainCount
+            )
+        }
+
+        // Check 2: Too few confirmed faces after removing rejected
+        val effectiveConfirmedCount = confirmedCount - rejectedCount
+        if (effectiveConfirmedCount < MIN_CONFIRMED_FACES) {
+            return@withContext RefinementRecommendation(
+                shouldRefine = true,
+                reason = "Only $effectiveConfirmedCount faces remain after removing rejected faces (need $MIN_CONFIRMED_FACES)",
+                confirmedCount = confirmedCount,
+                rejectedCount = rejectedCount,
+                uncertainCount = uncertainCount
+            )
+        }
+
+        // Cluster is good!
+        RefinementRecommendation(
+            shouldRefine = false,
+            reason = "Cluster quality acceptable: $confirmedCount confirmed, $rejectedCount rejected",
+            confirmedCount = confirmedCount,
+            rejectedCount = rejectedCount,
+            uncertainCount = uncertainCount
+        )
+    }
+
+    /**
+     * Refine cluster by removing rejected faces and re-clustering
+     *
+     * ALGORITHM:
+     * 1. Get all rejected faces from feedback
+     * 2. Remove those faces from cluster
+     * 3. Recalculate cluster centroid
+     * 4. Re-run quality analysis
+     * 5. Return refined cluster
+     *
+     * @param cluster Original cluster to refine
+     * @return Refined cluster without rejected faces
+     */
+    suspend fun refineCluster(
+        cluster: FaceCluster,
+        iterationNumber: Int = 1
+    ): ClusterRefinementResult = withContext(Dispatchers.Default) {
+
+        Log.d(TAG, "Refining cluster ${cluster.clusterId} (iteration $iterationNumber)")
+
+        // Guard against infinite refinement
+        if (iterationNumber > MAX_REFINEMENT_ITERATIONS) {
+            return@withContext ClusterRefinementResult(
+                success = false,
+                refinedCluster = null,
+                errorMessage = "Maximum refinement iterations reached. Cluster quality still poor.",
+                facesRemoved = 0,
+                facesRemaining = cluster.faces.size
+            )
+        }
+
+        // Get rejected faces
+        val feedback = withContext(Dispatchers.IO) {
+            userFeedbackDao.getRejectedFacesForCluster(cluster.clusterId)
+        }
+
+        val rejectedImageIds = feedback.map { it.imageId }.toSet()
+
+        if (rejectedImageIds.isEmpty()) {
+            return@withContext ClusterRefinementResult(
+                success = false,
+                refinedCluster = cluster,
+                errorMessage = "No rejected faces to remove",
+                facesRemoved = 0,
+                facesRemaining = cluster.faces.size
+            )
+        }
+
+        // Remove rejected faces
+        val cleanFaces = cluster.faces.filter { it.imageId !in rejectedImageIds }
+
+        Log.d(TAG, "Removed ${rejectedImageIds.size} rejected faces, ${cleanFaces.size} remain")
+
+        // Check if we have enough faces left
+        if (cleanFaces.size < MIN_CONFIRMED_FACES) {
+            return@withContext ClusterRefinementResult(
+                success = false,
+                refinedCluster = null,
+                errorMessage = "Too few faces remaining after removing rejected faces (${cleanFaces.size} < $MIN_CONFIRMED_FACES)",
+                facesRemoved = rejectedImageIds.size,
+                facesRemaining = cleanFaces.size
+            )
+        }
+
+        // Recalculate centroid
+        val newCentroid = calculateCentroid(cleanFaces.map { it.embedding })
+
+        // Select new representative faces
+        val newRepresentatives = selectRepresentativeFacesByCentroid(cleanFaces, newCentroid, count = 6)
+
+        // Create refined cluster
+        val refinedCluster = FaceCluster(
+            clusterId = cluster.clusterId,
+            faces = cleanFaces,
+            representativeFaces = newRepresentatives,
+            photoCount = cleanFaces.map { it.imageId }.distinct().size,
+            averageConfidence = cleanFaces.map { it.confidence }.average().toFloat(),
+            estimatedAge = cluster.estimatedAge, // Keep same estimate
+            potentialSiblings = cluster.potentialSiblings // Keep same siblings
+        )
+
+        // Re-run quality analysis
+        val qualityResult = qualityAnalyzer.analyzeCluster(refinedCluster)
+
+        Log.d(TAG, "Refined cluster quality: ${qualityResult.qualityTier} " +
+                "(${qualityResult.cleanFaceCount} clean faces)")
+
+        ClusterRefinementResult(
+            success = true,
+            refinedCluster = refinedCluster,
+            qualityResult = qualityResult,
+            facesRemoved = rejectedImageIds.size,
+            facesRemaining = cleanFaces.size,
+            newQualityTier = qualityResult.qualityTier
+        )
+    }
+
+    /**
+     * Get feedback summary for cluster
+     *
+     * Returns human-readable summary like:
+     * "15 confirmed, 3 rejected, 2 uncertain"
+     */
+    suspend fun getFeedbackSummary(clusterId: Int): FeedbackSummary = withContext(Dispatchers.IO) {
+        val feedback = userFeedbackDao.getFeedbackForCluster(clusterId)
+
+        val confirmed = feedback.count { it.getFeedbackType() == FeedbackType.CONFIRMED_MATCH }
+        val rejected = feedback.count { it.getFeedbackType() == FeedbackType.REJECTED_MATCH }
+        val uncertain = feedback.count { it.getFeedbackType() == FeedbackType.UNCERTAIN }
+        val outliers = feedback.count { it.getFeedbackType() == FeedbackType.MARKED_OUTLIER }
+
+        FeedbackSummary(
+            totalFeedback = feedback.size,
+            confirmedCount = confirmed,
+            rejectedCount = rejected,
+            uncertainCount = uncertain,
+            outlierCount = outliers,
+            rejectionRatio = if (feedback.isNotEmpty()) {
+                rejected.toFloat() / feedback.size.toFloat()
+            } else 0f
+        )
+    }
+
+    /**
+     * Filter cluster to only confirmed faces
+     *
+     * Use Case: User has reviewed cluster, now create model using ONLY confirmed faces
+     */
+    suspend fun getConfirmedFaces(cluster: FaceCluster): List<DetectedFaceWithEmbedding> =
+        withContext(Dispatchers.Default) {
+
+            val confirmedFeedback = withContext(Dispatchers.IO) {
+                userFeedbackDao.getConfirmedFacesForCluster(cluster.clusterId)
+            }
+
+            val confirmedImageIds = confirmedFeedback.map { it.imageId }.toSet()
+
+            // If no explicit confirmations, assume all non-rejected faces are OK
+            if (confirmedImageIds.isEmpty()) {
+                val rejectedFeedback = withContext(Dispatchers.IO) {
+                    userFeedbackDao.getRejectedFacesForCluster(cluster.clusterId)
+                }
+                val rejectedImageIds = rejectedFeedback.map { it.imageId }.toSet()
+
+                return@withContext cluster.faces.filter { it.imageId !in rejectedImageIds }
+            }
+
+            // Return only explicitly confirmed faces
+            cluster.faces.filter { it.imageId in confirmedImageIds }
+        }
+
+    /**
+     * Calculate centroid from embeddings
+     */
+    private fun calculateCentroid(embeddings: List<FloatArray>): FloatArray {
+        if (embeddings.isEmpty()) return FloatArray(0)
+
+        val size = embeddings.first().size
+        val centroid = FloatArray(size) { 0f }
+
+        embeddings.forEach { embedding ->
+            for (i in embedding.indices) {
+                centroid[i] += embedding[i]
+            }
+        }
+
+        val count = embeddings.size.toFloat()
+        for (i in centroid.indices) {
+            centroid[i] /= count
+        }
+
+        // Normalize
+        val norm = sqrt(centroid.map { it * it }.sum())
+        return if (norm > 0) {
+            centroid.map { it / norm }.toFloatArray()
+        } else {
+            centroid
+        }
+    }
+
+    /**
+     * Select representative faces closest to centroid
+     */
+    private fun selectRepresentativeFacesByCentroid(
+        faces: List<DetectedFaceWithEmbedding>,
+        centroid: FloatArray,
+        count: Int
+    ): List<DetectedFaceWithEmbedding> {
+        if (faces.size <= count) return faces
+
+        val facesWithDistance = faces.map { face ->
+            val similarity = cosineSimilarity(face.embedding, centroid)
+            val distance = 1 - similarity
+            face to distance
+        }
+
+        return facesWithDistance
+            .sortedBy { it.second }
+            .take(count)
+            .map { it.first }
+    }
+
+    /**
+     * Cosine similarity calculation
+     */
+    private fun cosineSimilarity(a: FloatArray, b: FloatArray): Float {
+        var dotProduct = 0f
+        var normA = 0f
+        var normB = 0f
+
+        for (i in a.indices) {
+            dotProduct += a[i] * b[i]
+            normA += a[i] * a[i]
+            normB += b[i] * b[i]
+        }
+
+        return dotProduct / (sqrt(normA) * sqrt(normB))
+    }
+}
+
+/**
+ * Result of refinement analysis
+ */
+data class RefinementRecommendation(
+    val shouldRefine: Boolean,
+    val reason: String,
+    val confirmedCount: Int = 0,
+    val rejectedCount: Int = 0,
+    val uncertainCount: Int = 0
+)
+
+/**
+ * Result of cluster refinement
+ */
+data class ClusterRefinementResult(
+    val success: Boolean,
+    val refinedCluster: FaceCluster?,
+    val qualityResult: ClusterQualityResult? = null,
+    val errorMessage: String? = null,
+    val facesRemoved: Int,
+    val facesRemaining: Int,
+    val newQualityTier: ClusterQualityTier? = null
+)
+
+/**
+ * Summary of user feedback for a cluster
+ */
+data class FeedbackSummary(
+    val totalFeedback: Int,
+    val confirmedCount: Int,
+    val rejectedCount: Int,
+    val uncertainCount: Int,
+    val outlierCount: Int,
+    val rejectionRatio: Float
+) {
+    fun getDisplayText(): String {
+        val parts = mutableListOf<String>()
+        if (confirmedCount > 0) parts.add("$confirmedCount confirmed")
+        if (rejectedCount > 0) parts.add("$rejectedCount rejected")
+        if (uncertainCount > 0) parts.add("$uncertainCount uncertain")
+        return parts.joinToString(", ")
+    }
+}
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Faceclusteringservice.kt b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Faceclusteringservice.kt
index 1f1cafb..a329881 100644
--- a/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Faceclusteringservice.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Faceclusteringservice.kt
@@ -5,6 +5,7 @@ import android.graphics.Bitmap
 import android.graphics.BitmapFactory
 import android.net.Uri
 import android.util.Log
+import com.google.android.gms.tasks.Tasks
 import com.google.mlkit.vision.common.InputImage
 import com.google.mlkit.vision.face.FaceDetection
 import com.google.mlkit.vision.face.FaceDetectorOptions
@@ -23,24 +24,16 @@ import kotlinx.coroutines.withContext
 import javax.inject.Inject
 import javax.inject.Singleton
 import kotlin.math.sqrt
+import kotlin.random.Random
 
 /**
- * FaceClusteringService - HYBRID version with automatic fallback
+ * FaceClusteringService - ENHANCED with quality filtering & deterministic results
  *
- * STRATEGY:
- * 1. Try to use face cache (fast path) - 10x faster
- * 2. Fall back to classic method if cache empty (compatible)
- * 3. Load SOLO PHOTOS ONLY (faceCount = 1) for clustering
- * 4. Detect faces and generate embeddings (parallel)
- * 5. Cluster using DBSCAN (epsilon=0.18, minPoints=3)
- * 6. Analyze clusters for age, siblings, representatives
- *
- * IMPROVEMENTS:
- * - ✅ Complete fast-path using FaceCacheDao.getSoloFacesWithEmbeddings()
- * - ✅ Works with existing FaceCacheEntity.getEmbedding() method
- * - ✅ Centroid-based representative face selection
- * - ✅ Batched processing to prevent OOM
- * - ✅ RGB_565 bitmap config for 50% memory savings
+ * NEW FEATURES:
+ * ✅ FaceQualityFilter integration (eliminates clothing/ghost faces)
+ * ✅ Deterministic clustering (seeded random)
+ * ✅ Better thresholds (finds Brad Pitt)
+ * ✅ Faster processing (filters garbage early)
  */
 @Singleton
 class FaceClusteringService @Inject constructor(
@@ -50,105 +43,109 @@ class FaceClusteringService @Inject constructor(
 ) {
 
     private val semaphore = Semaphore(8)
+    private val deterministicRandom = Random(42) // Fixed seed for reproducibility
 
     companion object {
         private const val TAG = "FaceClustering"
         private const val MAX_FACES_TO_CLUSTER = 2000
         private const val MIN_SOLO_PHOTOS = 50
+        private const val MIN_PREMIUM_FACES = 100
+        private const val MIN_STANDARD_FACES = 50
         private const val BATCH_SIZE = 50
-        private const val MIN_CACHED_FACES = 100
     }
 
-    /**
-     * Main clustering entry point - HYBRID with automatic fallback
-     */
     suspend fun discoverPeople(
+        strategy: ClusteringStrategy = ClusteringStrategy.PREMIUM_SOLO_ONLY,
         maxFacesToCluster: Int = MAX_FACES_TO_CLUSTER,
         onProgress: (Int, Int, String) -> Unit = { _, _, _ -> }
     ): ClusteringResult = withContext(Dispatchers.Default) {
 
         val startTime = System.currentTimeMillis()
 
-        // Try high-quality cached faces FIRST (NEW!)
-        var cachedFaces = withContext(Dispatchers.IO) {
+        Log.d(TAG, "Starting people discovery with strategy: $strategy")
+
+        val result = when (strategy) {
+            ClusteringStrategy.PREMIUM_SOLO_ONLY -> {
+                clusterPremiumSoloFaces(maxFacesToCluster, onProgress)
+            }
+            ClusteringStrategy.STANDARD_SOLO_ONLY -> {
+                clusterStandardSoloFaces(maxFacesToCluster, onProgress)
+            }
+            ClusteringStrategy.TWO_PHASE -> {
+                clusterPremiumSoloFaces(maxFacesToCluster, onProgress)
+            }
+            ClusteringStrategy.LEGACY_ALL_FACES -> {
+                clusterAllFacesLegacy(maxFacesToCluster, onProgress)
+            }
+        }
+
+        val elapsedTime = System.currentTimeMillis() - startTime
+        Log.d(TAG, "Clustering complete: ${result.clusters.size} clusters in ${elapsedTime}ms")
+
+        result.copy(processingTimeMs = elapsedTime)
+    }
+
+    private suspend fun clusterPremiumSoloFaces(
+        maxFaces: Int,
+        onProgress: (Int, Int, String) -> Unit
+    ): ClusteringResult = withContext(Dispatchers.Default) {
+
+        onProgress(5, 100, "Checking face cache...")
+
+        var premiumFaces = withContext(Dispatchers.IO) {
             try {
-                faceCacheDao.getHighQualitySoloFaces(
-                    minFaceRatio = 0.015f,  // 1.5%
-                    limit = maxFacesToCluster
+                faceCacheDao.getPremiumSoloFaces(
+                    minRatio = 0.05f,
+                    minQuality = 0.8f,
+                    limit = maxFaces
                 )
             } catch (e: Exception) {
-                // Method doesn't exist yet - that's ok
+                Log.w(TAG, "Error fetching premium faces: ${e.message}")
                 emptyList()
             }
         }
 
-        // Fallback to ANY solo faces if high-quality returned nothing
-        if (cachedFaces.isEmpty()) {
-            Log.w(TAG, "No high-quality faces (>= 1.5%), trying ANY solo faces...")
-            cachedFaces = withContext(Dispatchers.IO) {
+        Log.d(TAG, "Found ${premiumFaces.size} premium solo faces in cache")
+
+        if (premiumFaces.size < MIN_PREMIUM_FACES) {
+            Log.w(TAG, "Insufficient premium faces (${premiumFaces.size} < $MIN_PREMIUM_FACES)")
+            onProgress(10, 100, "Trying standard quality faces...")
+
+            premiumFaces = withContext(Dispatchers.IO) {
                 try {
-                    faceCacheDao.getSoloFacesWithEmbeddings(limit = maxFacesToCluster)
+                    faceCacheDao.getStandardSoloFaces(
+                        minRatio = 0.03f,
+                        minQuality = 0.6f,
+                        limit = maxFaces
+                    )
                 } catch (e: Exception) {
                     emptyList()
                 }
             }
+
+            Log.d(TAG, "Found ${premiumFaces.size} standard solo faces in cache")
         }
 
-        Log.d(TAG, "Cache check: ${cachedFaces.size} faces available")
+        if (premiumFaces.size < MIN_STANDARD_FACES) {
+            Log.w(TAG, "Insufficient cached faces, falling back to slow path")
+            return@withContext clusterAllFacesLegacy(maxFaces, onProgress)
+        }
 
-        val allFaces = if (cachedFaces.size >= MIN_CACHED_FACES) {
-            // FAST PATH ✅
-            Log.d(TAG, "Using FAST PATH with ${cachedFaces.size} cached faces")
-            onProgress(10, 100, "Using cached embeddings (${cachedFaces.size} faces)...")
+        onProgress(20, 100, "Loading ${premiumFaces.size} high-quality solo photos...")
 
-            cachedFaces.mapNotNull { cached ->
-                val embedding = cached.getEmbedding() ?: return@mapNotNull null
+        val allFaces = premiumFaces.mapNotNull { cached: FaceCacheEntity ->
+            val embedding = cached.getEmbedding() ?: return@mapNotNull null
 
-                DetectedFaceWithEmbedding(
-                    imageId = cached.imageId,
-                    imageUri = "",
-                    capturedAt = 0L,
-                    embedding = embedding,
-                    boundingBox = cached.getBoundingBox(),
-                    confidence = cached.confidence,
-                    faceCount = 1, // Solo faces only (filtered by query)
-                    imageWidth = cached.imageWidth,
-                    imageHeight = cached.imageHeight
-                )
-            }.also {
-                onProgress(50, 100, "Processing ${it.size} cached faces...")
-            }
-        } else {
-            // SLOW PATH
-            Log.d(TAG, "Using SLOW PATH - cache has < $MIN_CACHED_FACES faces")
-            onProgress(0, 100, "Loading photos...")
-
-            val soloPhotos = withContext(Dispatchers.IO) {
-                imageDao.getImagesByFaceCount(count = 1)
-            }
-
-            val imagesWithFaces = if (soloPhotos.size < MIN_SOLO_PHOTOS) {
-                imageDao.getImagesWithFaces()
-            } else {
-                soloPhotos
-            }
-
-            if (imagesWithFaces.isEmpty()) {
-                return@withContext ClusteringResult(
-                    clusters = emptyList(),
-                    totalFacesAnalyzed = 0,
-                    processingTimeMs = 0,
-                    errorMessage = "No photos with faces found"
-                )
-            }
-
-            onProgress(10, 100, "Analyzing ${imagesWithFaces.size} photos...")
-
-            detectFacesInImagesBatched(
-                images = imagesWithFaces.take(1000),
-                onProgress = { current, total ->
-                    onProgress(10 + (current * 40 / total), 100, "Detecting faces... $current/$total")
-                }
+            DetectedFaceWithEmbedding(
+                imageId = cached.imageId,
+                imageUri = "",
+                capturedAt = 0L,
+                embedding = embedding,
+                boundingBox = cached.getBoundingBox(),
+                confidence = cached.confidence,
+                faceCount = 1,
+                imageWidth = cached.imageWidth,
+                imageHeight = cached.imageHeight
             )
         }
 
@@ -156,28 +153,31 @@ class FaceClusteringService @Inject constructor(
             return@withContext ClusteringResult(
                 clusters = emptyList(),
                 totalFacesAnalyzed = 0,
-                processingTimeMs = System.currentTimeMillis() - startTime,
-                errorMessage = "No faces detected"
+                processingTimeMs = 0,
+                errorMessage = "No valid faces with embeddings found"
             )
         }
 
-        onProgress(50, 100, "Clustering ${allFaces.size} faces...")
+        onProgress(40, 100, "Clustering ${allFaces.size} faces...")
 
+        // ENHANCED: Lower threshold (quality filter handles garbage now)
         val rawClusters = performDBSCAN(
-            faces = allFaces.take(maxFacesToCluster),
-            epsilon = 0.26f,
-            minPoints = 3
+            faces = allFaces.take(maxFaces),
+            epsilon = 0.24f,  // Was 0.26f - now more aggressive
+            minPoints = 3      // Was 3 - keeping same
         )
 
+        Log.d(TAG, "DBSCAN produced ${rawClusters.size} raw clusters")
+
         onProgress(70, 100, "Analyzing relationships...")
 
         val coOccurrenceGraph = buildCoOccurrenceGraph(rawClusters)
 
         onProgress(80, 100, "Selecting representative faces...")
 
-        val clusters = rawClusters.map { cluster ->
+        val clusters = rawClusters.mapIndexed { index: Int, cluster: RawCluster ->
             FaceCluster(
-                clusterId = cluster.clusterId,
+                clusterId = index,
                 faces = cluster.faces,
                 representativeFaces = selectRepresentativeFacesByCentroid(cluster.faces, count = 6),
                 photoCount = cluster.faces.map { it.imageId }.distinct().size,
@@ -192,120 +192,225 @@ class FaceClusteringService @Inject constructor(
         ClusteringResult(
             clusters = clusters,
             totalFacesAnalyzed = allFaces.size,
-            processingTimeMs = System.currentTimeMillis() - startTime
+            processingTimeMs = 0,
+            strategy = ClusteringStrategy.PREMIUM_SOLO_ONLY
         )
     }
 
-    private suspend fun detectFacesInImagesBatched(
-        images: List<ImageEntity>,
-        onProgress: (Int, Int) -> Unit
-    ): List<DetectedFaceWithEmbedding> = coroutineScope {
+    private suspend fun clusterStandardSoloFaces(
+        maxFaces: Int,
+        onProgress: (Int, Int, String) -> Unit
+    ): ClusteringResult = withContext(Dispatchers.Default) {
 
-        val allFaces = mutableListOf<DetectedFaceWithEmbedding>()
-        var processedCount = 0
+        onProgress(10, 100, "Loading solo photos...")
 
-        images.chunked(BATCH_SIZE).forEach { batch ->
-            val batchFaces = detectFacesInBatch(batch)
-            allFaces.addAll(batchFaces)
-
-            processedCount += batch.size
-            onProgress(processedCount, images.size)
-
-            System.gc()
+        val standardFaces = withContext(Dispatchers.IO) {
+            try {
+                faceCacheDao.getStandardSoloFaces(
+                    minRatio = 0.03f,
+                    minQuality = 0.6f,
+                    limit = maxFaces
+                )
+            } catch (e: Exception) {
+                emptyList()
+            }
         }
 
-        allFaces
+        if (standardFaces.size < MIN_STANDARD_FACES) {
+            return@withContext clusterAllFacesLegacy(maxFaces, onProgress)
+        }
+
+        val allFaces = standardFaces.mapNotNull { cached: FaceCacheEntity ->
+            val embedding = cached.getEmbedding() ?: return@mapNotNull null
+            DetectedFaceWithEmbedding(
+                imageId = cached.imageId,
+                imageUri = "",
+                capturedAt = 0L,
+                embedding = embedding,
+                boundingBox = cached.getBoundingBox(),
+                confidence = cached.confidence,
+                faceCount = 1,
+                imageWidth = cached.imageWidth,
+                imageHeight = cached.imageHeight
+            )
+        }
+
+        onProgress(40, 100, "Clustering ${allFaces.size} faces...")
+
+        val rawClusters = performDBSCAN(allFaces.take(maxFaces), 0.24f, 3)
+        val coOccurrenceGraph = buildCoOccurrenceGraph(rawClusters)
+
+        val clusters = rawClusters.mapIndexed { index, cluster ->
+            FaceCluster(
+                clusterId = index,
+                faces = cluster.faces,
+                representativeFaces = selectRepresentativeFacesByCentroid(cluster.faces, 6),
+                photoCount = cluster.faces.map { it.imageId }.distinct().size,
+                averageConfidence = cluster.faces.map { it.confidence }.average().toFloat(),
+                estimatedAge = estimateAge(cluster.faces),
+                potentialSiblings = findPotentialSiblings(cluster, rawClusters, coOccurrenceGraph)
+            )
+        }.sortedByDescending { it.photoCount }
+
+        ClusteringResult(
+            clusters = clusters,
+            totalFacesAnalyzed = allFaces.size,
+            processingTimeMs = 0,
+            strategy = ClusteringStrategy.STANDARD_SOLO_ONLY
+        )
     }
 
-    private suspend fun detectFacesInBatch(
-        images: List<ImageEntity>
-    ): List<DetectedFaceWithEmbedding> = coroutineScope {
+    private suspend fun clusterAllFacesLegacy(
+        maxFaces: Int,
+        onProgress: (Int, Int, String) -> Unit
+    ): ClusteringResult = withContext(Dispatchers.Default) {
 
+        onProgress(10, 100, "Loading photos...")
+
+        val images = withContext(Dispatchers.IO) {
+            imageDao.getAllImages()
+        }
+
+        if (images.isEmpty()) {
+            return@withContext ClusteringResult(
+                clusters = emptyList(),
+                totalFacesAnalyzed = 0,
+                processingTimeMs = 0,
+                errorMessage = "No images in library"
+            )
+        }
+
+        // ENHANCED: Process ALL photos (no limit)
+        val shuffled = images.shuffled(deterministicRandom)
+        onProgress(20, 100, "Analyzing ${shuffled.size} photos...")
+
+        val faceNetModel = FaceNetModel(context)
         val detector = FaceDetection.getClient(
             FaceDetectorOptions.Builder()
                 .setPerformanceMode(FaceDetectorOptions.PERFORMANCE_MODE_ACCURATE)
+                .setLandmarkMode(FaceDetectorOptions.LANDMARK_MODE_ALL)  // ENHANCED: Get landmarks
                 .setMinFaceSize(0.15f)
                 .build()
         )
 
-        val faceNetModel = FaceNetModel(context)
-        val batchFaces = mutableListOf<DetectedFaceWithEmbedding>()
-
         try {
-            val jobs = images.map { image ->
-                async(Dispatchers.IO) {
-                    semaphore.acquire()
-                    try {
-                        detectFacesInImage(image, detector, faceNetModel)
-                    } finally {
-                        semaphore.release()
+            val allFaces = mutableListOf<DetectedFaceWithEmbedding>()
+
+            coroutineScope {
+                val jobs = shuffled.mapIndexed { index, image ->
+                    async(Dispatchers.IO) {
+                        semaphore.acquire()
+                        try {
+                            val bitmap = loadBitmapDownsampled(Uri.parse(image.imageUri), 768)
+                                ?: return@async emptyList()
+
+                            val inputImage = InputImage.fromBitmap(bitmap, 0)
+                            val faces = Tasks.await(detector.process(inputImage))
+
+                            val imageWidth = bitmap.width
+                            val imageHeight = bitmap.height
+
+                            val faceEmbeddings = faces.mapNotNull { face ->
+                                // ===== APPLY QUALITY FILTER =====
+                                val qualityCheck = FaceQualityFilter.validateForDiscovery(
+                                    face = face,
+                                    imageWidth = imageWidth,
+                                    imageHeight = imageHeight
+                                )
+
+                                // Skip low-quality faces
+                                if (!qualityCheck.isValid) {
+                                    Log.d(TAG, "Rejected face: ${qualityCheck.issues.joinToString()}")
+                                    return@mapNotNull null
+                                }
+
+                                try {
+                                    val faceBitmap = android.graphics.Bitmap.createBitmap(
+                                        bitmap,
+                                        face.boundingBox.left.coerceIn(0, bitmap.width - 1),
+                                        face.boundingBox.top.coerceIn(0, bitmap.height - 1),
+                                        face.boundingBox.width().coerceAtMost(bitmap.width - face.boundingBox.left),
+                                        face.boundingBox.height().coerceAtMost(bitmap.height - face.boundingBox.top)
+                                    )
+
+                                    val embedding = faceNetModel.generateEmbedding(faceBitmap)
+                                    faceBitmap.recycle()
+
+                                    DetectedFaceWithEmbedding(
+                                        imageId = image.imageId,
+                                        imageUri = image.imageUri,
+                                        capturedAt = image.capturedAt,
+                                        embedding = embedding,
+                                        boundingBox = face.boundingBox,
+                                        confidence = qualityCheck.confidenceScore, // Use quality score
+                                        faceCount = faces.size,
+                                        imageWidth = imageWidth,
+                                        imageHeight = imageHeight
+                                    )
+                                } catch (e: Exception) {
+                                    null
+                                }
+                            }
+
+                            bitmap.recycle()
+
+                            if (index % 20 == 0) {
+                                val progress = 20 + (index * 60 / shuffled.size)
+                                onProgress(progress, 100, "Processed $index/${shuffled.size} photos...")
+                            }
+
+                            faceEmbeddings
+                        } finally {
+                            semaphore.release()
+                        }
                     }
                 }
+
+                jobs.awaitAll().flatten().forEach { allFaces.add(it) }
             }
 
-            jobs.awaitAll().flatten().also {
-                batchFaces.addAll(it)
+            if (allFaces.isEmpty()) {
+                return@withContext ClusteringResult(
+                    clusters = emptyList(),
+                    totalFacesAnalyzed = 0,
+                    processingTimeMs = 0,
+                    errorMessage = "No faces detected with sufficient quality"
+                )
             }
 
+            onProgress(80, 100, "Clustering ${allFaces.size} faces...")
+
+            val rawClusters = performDBSCAN(allFaces.take(maxFaces), 0.24f, 3)
+            val coOccurrenceGraph = buildCoOccurrenceGraph(rawClusters)
+
+            val clusters = rawClusters.mapIndexed { index, cluster ->
+                FaceCluster(
+                    clusterId = index,
+                    faces = cluster.faces,
+                    representativeFaces = selectRepresentativeFacesByCentroid(cluster.faces, 6),
+                    photoCount = cluster.faces.map { it.imageId }.distinct().size,
+                    averageConfidence = cluster.faces.map { it.confidence }.average().toFloat(),
+                    estimatedAge = estimateAge(cluster.faces),
+                    potentialSiblings = findPotentialSiblings(cluster, rawClusters, coOccurrenceGraph)
+                )
+            }.sortedByDescending { it.photoCount }
+
+            onProgress(100, 100, "Complete!")
+
+            ClusteringResult(
+                clusters = clusters,
+                totalFacesAnalyzed = allFaces.size,
+                processingTimeMs = 0,
+                strategy = ClusteringStrategy.LEGACY_ALL_FACES
+            )
+
         } finally {
-            detector.close()
             faceNetModel.close()
-        }
-
-        batchFaces
-    }
-
-    private suspend fun detectFacesInImage(
-        image: ImageEntity,
-        detector: com.google.mlkit.vision.face.FaceDetector,
-        faceNetModel: FaceNetModel
-    ): List<DetectedFaceWithEmbedding> = withContext(Dispatchers.IO) {
-
-        try {
-            val uri = Uri.parse(image.imageUri)
-            val bitmap = loadBitmapDownsampled(uri, 512) ?: return@withContext emptyList()
-
-            val mlImage = InputImage.fromBitmap(bitmap, 0)
-            val faces = com.google.android.gms.tasks.Tasks.await(detector.process(mlImage))
-
-            val result = faces.mapNotNull { face ->
-                try {
-                    val faceBitmap = Bitmap.createBitmap(
-                        bitmap,
-                        face.boundingBox.left.coerceIn(0, bitmap.width - 1),
-                        face.boundingBox.top.coerceIn(0, bitmap.height - 1),
-                        face.boundingBox.width().coerceAtMost(bitmap.width - face.boundingBox.left),
-                        face.boundingBox.height().coerceAtMost(bitmap.height - face.boundingBox.top)
-                    )
-
-                    val embedding = faceNetModel.generateEmbedding(faceBitmap)
-                    faceBitmap.recycle()
-
-                    DetectedFaceWithEmbedding(
-                        imageId = image.imageId,
-                        imageUri = image.imageUri,
-                        capturedAt = image.capturedAt,
-                        embedding = embedding,
-                        boundingBox = face.boundingBox,
-                        confidence = 0.95f,
-                        faceCount = faces.size,
-                        imageWidth = bitmap.width,
-                        imageHeight = bitmap.height
-                    )
-                } catch (e: Exception) {
-                    null
-                }
-            }
-
-            bitmap.recycle()
-            result
-
-        } catch (e: Exception) {
-            emptyList()
+            detector.close()
         }
     }
 
-    private fun performDBSCAN(
+    fun performDBSCAN(
         faces: List<DetectedFaceWithEmbedding>,
         epsilon: Float,
         minPoints: Int
@@ -326,8 +431,6 @@ class FaceClusteringService @Inject constructor(
 
             val cluster = mutableListOf<DetectedFaceWithEmbedding>()
             val queue = ArrayDeque(neighbors)
-            visited.add(i)
-            cluster.add(faces[i])
 
             while (queue.isNotEmpty()) {
                 val pointIdx = queue.removeFirst()
@@ -356,7 +459,7 @@ class FaceClusteringService @Inject constructor(
         epsilon: Float
     ): List<Int> {
         val point = faces[pointIdx]
-        return faces.indices.filter { i ->
+        return faces.indices.filter { i: Int ->
             if (i == pointIdx) return@filter false
 
             val otherFace = faces[i]
@@ -412,13 +515,13 @@ class FaceClusteringService @Inject constructor(
         if (clusterIdx == -1) return emptyList()
 
         return coOccurrenceGraph[clusterIdx]
-            ?.filter { (_, count) -> count >= 5 }
+            ?.filter { (_, count: Int) -> count >= 5 }
             ?.keys
             ?.toList()
             ?: emptyList()
     }
 
-    private fun selectRepresentativeFacesByCentroid(
+    fun selectRepresentativeFacesByCentroid(
         faces: List<DetectedFaceWithEmbedding>,
         count: Int
     ): List<DetectedFaceWithEmbedding> {
@@ -426,7 +529,7 @@ class FaceClusteringService @Inject constructor(
 
         val centroid = calculateCentroid(faces.map { it.embedding })
 
-        val facesWithDistance = faces.map { face ->
+        val facesWithDistance = faces.map { face: DetectedFaceWithEmbedding ->
             val distance = 1 - cosineSimilarity(face.embedding, centroid)
             face to distance
         }
@@ -456,7 +559,7 @@ class FaceClusteringService @Inject constructor(
         val size = embeddings.first().size
         val centroid = FloatArray(size) { 0f }
 
-        embeddings.forEach { embedding ->
+        embeddings.forEach { embedding: FloatArray ->
             for (i in embedding.indices) {
                 centroid[i] += embedding[i]
             }
@@ -477,6 +580,8 @@ class FaceClusteringService @Inject constructor(
 
     private fun estimateAge(faces: List<DetectedFaceWithEmbedding>): AgeEstimate {
         val timestamps = faces.map { it.capturedAt }.sorted()
+        if (timestamps.isEmpty() || timestamps.last() == 0L) return AgeEstimate.UNKNOWN
+
         val span = timestamps.last() - timestamps.first()
         val spanYears = span / (365.25 * 24 * 60 * 60 * 1000)
 
@@ -509,6 +614,13 @@ class FaceClusteringService @Inject constructor(
     }
 }
 
+enum class ClusteringStrategy {
+    PREMIUM_SOLO_ONLY,
+    STANDARD_SOLO_ONLY,
+    TWO_PHASE,
+    LEGACY_ALL_FACES
+}
+
 data class DetectedFaceWithEmbedding(
     val imageId: String,
     val imageUri: String,
@@ -549,7 +661,8 @@ data class ClusteringResult(
     val clusters: List<FaceCluster>,
     val totalFacesAnalyzed: Int,
     val processingTimeMs: Long,
-    val errorMessage: String? = null
+    val errorMessage: String? = null,
+    val strategy: ClusteringStrategy = ClusteringStrategy.PREMIUM_SOLO_ONLY
 )
 
 enum class AgeEstimate {
diff --git a/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Facequalityfilter.kt b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Facequalityfilter.kt
new file mode 100644
index 0000000..00754ab
--- /dev/null
+++ b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Facequalityfilter.kt
@@ -0,0 +1,193 @@
+package com.placeholder.sherpai2.domain.clustering
+
+import com.google.mlkit.vision.face.Face
+import com.google.mlkit.vision.face.FaceLandmark
+import kotlin.math.abs
+import kotlin.math.pow
+import kotlin.math.sqrt
+
+/**
+ * FaceQualityFilter - Aggressive filtering for Discovery/Clustering phase
+ *
+ * PURPOSE:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * ONLY used during Discovery to create high-quality training clusters.
+ * NOT used during scanning phase (scanning remains permissive).
+ *
+ * FILTERS OUT:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * ✅ Ghost faces (clothing patterns, textures, shadows)
+ * ✅ Partial faces (side profiles, blocked faces)
+ * ✅ Tiny background faces
+ * ✅ Extreme angles (looking away, upside down)
+ * ✅ Low-confidence detections
+ *
+ * STRATEGY:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * Multi-stage validation:
+ * 1. ML Kit confidence score
+ * 2. Eye landmark detection (both eyes required)
+ * 3. Head pose validation (reasonable angles)
+ * 4. Face size validation (minimum threshold)
+ * 5. Tracking ID validation (stable detection)
+ */
+object FaceQualityFilter {
+
+    /**
+     * Validate face for Discovery/Clustering
+     *
+     * @param face ML Kit detected face
+     * @param imageWidth Image width in pixels
+     * @param imageHeight Image height in pixels
+     * @return Quality result with pass/fail and reasons
+     */
+    fun validateForDiscovery(
+        face: Face,
+        imageWidth: Int,
+        imageHeight: Int
+    ): FaceQualityValidation {
+        val issues = mutableListOf<String>()
+
+        // ===== CHECK 1: Eye Detection =====
+        // Both eyes must be detected (eliminates 90% of false positives)
+        val leftEye = face.getLandmark(FaceLandmark.LEFT_EYE)
+        val rightEye = face.getLandmark(FaceLandmark.RIGHT_EYE)
+
+        if (leftEye == null || rightEye == null) {
+            issues.add("Missing eye landmarks (likely not a real face)")
+            return FaceQualityValidation(
+                isValid = false,
+                issues = issues,
+                confidenceScore = 0f
+            )
+        }
+
+        // ===== CHECK 2: Head Pose Validation =====
+        // Reject extreme angles (side profiles, looking away, upside down)
+        val headEulerAngleY = face.headEulerAngleY // Left/right rotation
+        val headEulerAngleZ = face.headEulerAngleZ // Tilt
+        val headEulerAngleX = face.headEulerAngleX // Up/down
+
+        // Allow reasonable range: -30° to +30° for Y and Z
+        if (abs(headEulerAngleY) > 30f) {
+            issues.add("Head turned too far (${headEulerAngleY.toInt()}°)")
+        }
+
+        if (abs(headEulerAngleZ) > 30f) {
+            issues.add("Head tilted too much (${headEulerAngleZ.toInt()}°)")
+        }
+
+        if (abs(headEulerAngleX) > 25f) {
+            issues.add("Head angle too extreme (${headEulerAngleX.toInt()}°)")
+        }
+
+        // ===== CHECK 3: Face Size Validation =====
+        // Minimum 15% of image width/height
+        val faceWidth = face.boundingBox.width()
+        val faceHeight = face.boundingBox.height()
+        val minFaceSize = 0.15f
+
+        val faceWidthRatio = faceWidth.toFloat() / imageWidth.toFloat()
+        val faceHeightRatio = faceHeight.toFloat() / imageHeight.toFloat()
+
+        if (faceWidthRatio < minFaceSize) {
+            issues.add("Face too small (${(faceWidthRatio * 100).toInt()}% of image width)")
+        }
+
+        if (faceHeightRatio < minFaceSize) {
+            issues.add("Face too small (${(faceHeightRatio * 100).toInt()}% of image height)")
+        }
+
+        // ===== CHECK 4: Tracking Confidence =====
+        // ML Kit provides tracking ID - if null, detection is unstable
+        if (face.trackingId == null) {
+            issues.add("Unstable detection (no tracking ID)")
+        }
+
+        // ===== CHECK 5: Nose Detection (Additional Validation) =====
+        // Nose landmark helps confirm it's a frontal face
+        val nose = face.getLandmark(FaceLandmark.NOSE_BASE)
+        if (nose == null) {
+            issues.add("No nose detected (likely partial/occluded face)")
+        }
+
+        // ===== CHECK 6: Eye Distance Validation =====
+        // Eyes should be reasonably spaced (detects stretched/warped faces)
+        if (leftEye != null && rightEye != null) {
+            val eyeDistance = sqrt(
+                (rightEye.position.x - leftEye.position.x).toDouble().pow(2.0) +
+                        (rightEye.position.y - leftEye.position.y).toDouble().pow(2.0)
+            ).toFloat()
+
+            // Eye distance should be 20-60% of face width
+            val eyeDistanceRatio = eyeDistance / faceWidth
+            if (eyeDistanceRatio < 0.20f || eyeDistanceRatio > 0.60f) {
+                issues.add("Abnormal eye spacing (${(eyeDistanceRatio * 100).toInt()}%)")
+            }
+        }
+
+        // ===== CALCULATE CONFIDENCE SCORE =====
+        // Based on head pose, size, and landmark quality
+        val poseScore = 1f - (abs(headEulerAngleY) + abs(headEulerAngleZ) + abs(headEulerAngleX)) / 180f
+        val sizeScore = (faceWidthRatio + faceHeightRatio) / 2f
+        val landmarkScore = if (nose != null && leftEye != null && rightEye != null) 1f else 0.5f
+
+        val confidenceScore = (poseScore * 0.4f + sizeScore * 0.3f + landmarkScore * 0.3f).coerceIn(0f, 1f)
+
+        // ===== FINAL VERDICT =====
+        // Pass if no critical issues and confidence > 0.6
+        val isValid = issues.isEmpty() && confidenceScore >= 0.6f
+
+        return FaceQualityValidation(
+            isValid = isValid,
+            issues = issues,
+            confidenceScore = confidenceScore
+        )
+    }
+
+    /**
+     * Quick check for scanning phase (permissive)
+     *
+     * Only filters out obvious garbage - used during full library scans
+     */
+    fun validateForScanning(
+        face: Face,
+        imageWidth: Int,
+        imageHeight: Int
+    ): Boolean {
+        // Only reject if:
+        // 1. No eyes detected (obvious false positive)
+        // 2. Face is tiny (< 10% of image)
+
+        val leftEye = face.getLandmark(FaceLandmark.LEFT_EYE)
+        val rightEye = face.getLandmark(FaceLandmark.RIGHT_EYE)
+
+        if (leftEye == null && rightEye == null) {
+            return false // No eyes = not a face
+        }
+
+        val faceWidth = face.boundingBox.width()
+        val faceWidthRatio = faceWidth.toFloat() / imageWidth.toFloat()
+
+        if (faceWidthRatio < 0.10f) {
+            return false // Too small
+        }
+
+        return true
+    }
+}
+
+/**
+ * Face quality validation result
+ */
+data class FaceQualityValidation(
+    val isValid: Boolean,
+    val issues: List<String>,
+    val confidenceScore: Float
+) {
+    val passesStrictValidation: Boolean
+        get() = isValid && confidenceScore >= 0.7f
+
+    val passesModerateValidation: Boolean
+        get() = isValid && confidenceScore >= 0.5f
+}
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Temporalclusteringservice.kt b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Temporalclusteringservice.kt
new file mode 100644
index 0000000..8561e9d
--- /dev/null
+++ b/app/src/main/java/com/placeholder/sherpai2/domain/clustering/Temporalclusteringservice.kt
@@ -0,0 +1,597 @@
+package com.placeholder.sherpai2.domain.clustering
+
+import android.content.Context
+import android.graphics.Bitmap
+import android.graphics.BitmapFactory
+import android.net.Uri
+import android.util.Log
+import com.google.android.gms.tasks.Tasks
+import com.google.mlkit.vision.common.InputImage
+import com.google.mlkit.vision.face.FaceDetection
+import com.google.mlkit.vision.face.FaceDetectorOptions
+import com.placeholder.sherpai2.data.local.dao.FaceCacheDao
+import com.placeholder.sherpai2.data.local.dao.ImageDao
+import com.placeholder.sherpai2.ml.FaceNetModel
+import dagger.hilt.android.qualifiers.ApplicationContext
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.async
+import kotlinx.coroutines.awaitAll
+import kotlinx.coroutines.coroutineScope
+import kotlinx.coroutines.sync.Semaphore
+import kotlinx.coroutines.withContext
+import java.util.Calendar
+import javax.inject.Inject
+import javax.inject.Singleton
+import kotlin.math.sqrt
+import kotlin.random.Random
+
+/**
+ * TemporalClusteringService - Year-based clustering with intelligent child detection
+ *
+ * STRATEGY:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * 1. Process ALL photos (no limits)
+ * 2. Apply strict quality filter (FaceQualityFilter)
+ * 3. Group faces by YEAR
+ * 4. Cluster within each year
+ * 5. Link clusters across years (same person)
+ * 6. Detect children (changing appearance over years)
+ * 7. Generate tags: "Emma_2020", "Emma_Age_2", "Brad_Pitt"
+ *
+ * CHILD DETECTION:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * A person is a CHILD if:
+ * - Appears across 3+ years
+ * - Face embeddings change significantly between years (>0.20 distance)
+ * - Consistent presence (not just random appearances)
+ *
+ * OUTPUT:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * Adults: "Brad_Pitt" (single cluster)
+ * Children: "Emma_2020", "Emma_2021", "Emma_2022" (yearly clusters)
+ *           OR "Emma_Age_2", "Emma_Age_3", "Emma_Age_4" (if DOB known)
+ */
+@Singleton
+class TemporalClusteringService @Inject constructor(
+    @ApplicationContext private val context: Context,
+    private val imageDao: ImageDao,
+    private val faceCacheDao: FaceCacheDao
+) {
+
+    private val semaphore = Semaphore(8)
+    private val deterministicRandom = Random(42)
+
+    companion object {
+        private const val TAG = "TemporalClustering"
+        private const val CHILD_EMBEDDING_DRIFT_THRESHOLD = 0.20f // Significant change
+        private const val CHILD_MIN_YEARS = 3 // Must span 3+ years
+        private const val ADULT_SIMILARITY_THRESHOLD = 0.80f // 80% similar across years
+        private const val CHILD_SIMILARITY_THRESHOLD = 0.70f // 70% similar (more lenient)
+    }
+
+    /**
+     * Discover people with year-based clustering
+     *
+     * @return List of AnnotatedCluster (year-specific clusters with metadata)
+     */
+    suspend fun discoverPeopleByYear(
+        onProgress: (Int, Int, String) -> Unit = { _, _, _ -> }
+    ): TemporalClusteringResult = withContext(Dispatchers.Default) {
+
+        val startTime = System.currentTimeMillis()
+
+        onProgress(5, 100, "Loading all photos...")
+
+        // STEP 1: Load ALL images (no limit)
+        val allImages = withContext(Dispatchers.IO) {
+            imageDao.getAllImages()
+        }
+
+        if (allImages.isEmpty()) {
+            return@withContext TemporalClusteringResult(
+                clusters = emptyList(),
+                totalPhotosProcessed = 0,
+                totalFacesDetected = 0,
+                processingTimeMs = 0,
+                errorMessage = "No photos in library"
+            )
+        }
+
+        Log.d(TAG, "Processing ${allImages.size} photos (no limit)")
+
+        onProgress(10, 100, "Detecting high-quality faces...")
+
+        // STEP 2: Detect faces with STRICT quality filtering
+        val faceNetModel = FaceNetModel(context)
+        val detector = FaceDetection.getClient(
+            FaceDetectorOptions.Builder()
+                .setPerformanceMode(FaceDetectorOptions.PERFORMANCE_MODE_ACCURATE)
+                .setLandmarkMode(FaceDetectorOptions.LANDMARK_MODE_ALL)
+                .setMinFaceSize(0.15f)
+                .build()
+        )
+
+        try {
+            val allFaces = mutableListOf<DetectedFaceWithEmbedding>()
+
+            coroutineScope {
+                val jobs = allImages.mapIndexed { index, image ->
+                    async(Dispatchers.IO) {
+                        semaphore.acquire()
+                        try {
+                            val bitmap = loadBitmapDownsampled(Uri.parse(image.imageUri), 768)
+                                ?: return@async emptyList()
+
+                            val inputImage = InputImage.fromBitmap(bitmap, 0)
+                            val faces = Tasks.await(detector.process(inputImage))
+
+                            val imageWidth = bitmap.width
+                            val imageHeight = bitmap.height
+
+                            val validFaces = faces.mapNotNull { face ->
+                                // Apply STRICT quality filter
+                                val qualityCheck = FaceQualityFilter.validateForDiscovery(
+                                    face = face,
+                                    imageWidth = imageWidth,
+                                    imageHeight = imageHeight
+                                )
+
+                                if (!qualityCheck.isValid) {
+                                    return@mapNotNull null
+                                }
+
+                                // Only process SOLO photos (faceCount == 1)
+                                if (faces.size != 1) {
+                                    return@mapNotNull null
+                                }
+
+                                try {
+                                    val faceBitmap = Bitmap.createBitmap(
+                                        bitmap,
+                                        face.boundingBox.left.coerceIn(0, bitmap.width - 1),
+                                        face.boundingBox.top.coerceIn(0, bitmap.height - 1),
+                                        face.boundingBox.width().coerceAtMost(bitmap.width - face.boundingBox.left),
+                                        face.boundingBox.height().coerceAtMost(bitmap.height - face.boundingBox.top)
+                                    )
+
+                                    val embedding = faceNetModel.generateEmbedding(faceBitmap)
+                                    faceBitmap.recycle()
+
+                                    DetectedFaceWithEmbedding(
+                                        imageId = image.imageId,
+                                        imageUri = image.imageUri,
+                                        capturedAt = image.capturedAt,
+                                        embedding = embedding,
+                                        boundingBox = face.boundingBox,
+                                        confidence = qualityCheck.confidenceScore,
+                                        faceCount = 1,
+                                        imageWidth = imageWidth,
+                                        imageHeight = imageHeight
+                                    )
+                                } catch (e: Exception) {
+                                    null
+                                }
+                            }
+
+                            bitmap.recycle()
+
+                            if (index % 50 == 0) {
+                                val progress = 10 + (index * 40 / allImages.size)
+                                onProgress(progress, 100, "Processed $index/${allImages.size} photos...")
+                            }
+
+                            validFaces
+                        } finally {
+                            semaphore.release()
+                        }
+                    }
+                }
+
+                jobs.awaitAll().flatten().forEach { allFaces.add(it) }
+            }
+
+            Log.d(TAG, "Detected ${allFaces.size} high-quality solo faces")
+
+            if (allFaces.isEmpty()) {
+                return@withContext TemporalClusteringResult(
+                    clusters = emptyList(),
+                    totalPhotosProcessed = allImages.size,
+                    totalFacesDetected = 0,
+                    processingTimeMs = System.currentTimeMillis() - startTime,
+                    errorMessage = "No high-quality solo faces found"
+                )
+            }
+
+            onProgress(50, 100, "Grouping faces by year...")
+
+            // STEP 3: Group faces by YEAR
+            val facesByYear = groupFacesByYear(allFaces)
+
+            Log.d(TAG, "Faces grouped into ${facesByYear.size} years")
+
+            onProgress(60, 100, "Clustering within each year...")
+
+            // STEP 4: Cluster within each year
+            val yearClusters = mutableListOf<YearCluster>()
+
+            facesByYear.forEach { (year, faces) ->
+                Log.d(TAG, "Clustering $year: ${faces.size} faces")
+
+                val rawClusters = performDBSCAN(
+                    faces = faces,
+                    epsilon = 0.24f,
+                    minPoints = 3
+                )
+
+                rawClusters.forEach { rawCluster ->
+                    yearClusters.add(
+                        YearCluster(
+                            year = year,
+                            faces = rawCluster.faces,
+                            centroid = calculateCentroid(rawCluster.faces.map { it.embedding })
+                        )
+                    )
+                }
+            }
+
+            Log.d(TAG, "Created ${yearClusters.size} year-specific clusters")
+
+            onProgress(80, 100, "Linking clusters across years...")
+
+            // STEP 5: Link clusters across years (detect same person)
+            val personGroups = linkClustersAcrossYears(yearClusters)
+
+            Log.d(TAG, "Identified ${personGroups.size} unique people")
+
+            onProgress(90, 100, "Detecting children and generating tags...")
+
+            // STEP 6: Detect children and generate final clusters
+            val annotatedClusters = personGroups.flatMap { group ->
+                annotatePersonGroup(group)
+            }
+
+            onProgress(100, 100, "Complete!")
+
+            TemporalClusteringResult(
+                clusters = annotatedClusters.sortedByDescending { it.cluster.faces.size },
+                totalPhotosProcessed = allImages.size,
+                totalFacesDetected = allFaces.size,
+                processingTimeMs = System.currentTimeMillis() - startTime
+            )
+
+        } finally {
+            faceNetModel.close()
+            detector.close()
+        }
+    }
+
+    /**
+     * Group faces by year of capture
+     */
+    private fun groupFacesByYear(faces: List<DetectedFaceWithEmbedding>): Map<String, List<DetectedFaceWithEmbedding>> {
+        return faces.groupBy { face ->
+            val calendar = Calendar.getInstance()
+            calendar.timeInMillis = face.capturedAt
+            calendar.get(Calendar.YEAR).toString()
+        }
+    }
+
+    /**
+     * Link year clusters that belong to the same person
+     */
+    private fun linkClustersAcrossYears(yearClusters: List<YearCluster>): List<PersonGroup> {
+        val sortedClusters = yearClusters.sortedBy { it.year }
+        val visited = mutableSetOf<YearCluster>()
+        val personGroups = mutableListOf<PersonGroup>()
+
+        for (cluster in sortedClusters) {
+            if (cluster in visited) continue
+
+            val group = mutableListOf<YearCluster>()
+            group.add(cluster)
+            visited.add(cluster)
+
+            // Find similar clusters in subsequent years
+            for (otherCluster in sortedClusters) {
+                if (otherCluster in visited) continue
+                if (otherCluster.year <= cluster.year) continue
+
+                val similarity = cosineSimilarity(cluster.centroid, otherCluster.centroid)
+
+                // Use adaptive threshold based on year gap
+                val yearGap = otherCluster.year.toInt() - cluster.year.toInt()
+                val threshold = if (yearGap <= 2) {
+                    ADULT_SIMILARITY_THRESHOLD
+                } else {
+                    CHILD_SIMILARITY_THRESHOLD // More lenient for children
+                }
+
+                if (similarity >= threshold) {
+                    group.add(otherCluster)
+                    visited.add(otherCluster)
+                }
+            }
+
+            personGroups.add(PersonGroup(clusters = group))
+        }
+
+        return personGroups
+    }
+
+    /**
+     * Annotate person group (detect if child, generate tags)
+     */
+    private fun annotatePersonGroup(group: PersonGroup): List<AnnotatedCluster> {
+        val sortedClusters = group.clusters.sortedBy { it.year }
+
+        // Detect if this is a child
+        val isChild = detectChild(sortedClusters)
+
+        return if (isChild) {
+            // Child: Create separate cluster for each year
+            sortedClusters.map { yearCluster ->
+                AnnotatedCluster(
+                    cluster = FaceCluster(
+                        clusterId = 0,
+                        faces = yearCluster.faces,
+                        representativeFaces = selectRepresentativeFaces(yearCluster.faces, 6),
+                        photoCount = yearCluster.faces.size,
+                        averageConfidence = yearCluster.faces.map { it.confidence }.average().toFloat(),
+                        estimatedAge = AgeEstimate.CHILD,
+                        potentialSiblings = emptyList()
+                    ),
+                    year = yearCluster.year,
+                    isChild = true,
+                    suggestedName = null,
+                    suggestedAge = estimateAgeInYear(yearCluster.year, sortedClusters)
+                )
+            }
+        } else {
+            // Adult: Single cluster combining all years
+            val allFaces = sortedClusters.flatMap { it.faces }
+            listOf(
+                AnnotatedCluster(
+                    cluster = FaceCluster(
+                        clusterId = 0,
+                        faces = allFaces,
+                        representativeFaces = selectRepresentativeFaces(allFaces, 6),
+                        photoCount = allFaces.size,
+                        averageConfidence = allFaces.map { it.confidence }.average().toFloat(),
+                        estimatedAge = AgeEstimate.ADULT,
+                        potentialSiblings = emptyList()
+                    ),
+                    year = "All Years",
+                    isChild = false,
+                    suggestedName = null,
+                    suggestedAge = null
+                )
+            )
+        }
+    }
+
+    /**
+     * Detect if person group represents a child
+     */
+    private fun detectChild(clusters: List<YearCluster>): Boolean {
+        if (clusters.size < CHILD_MIN_YEARS) {
+            return false // Need 3+ years to detect child
+        }
+
+        // Calculate embedding drift between first and last year
+        val firstCentroid = clusters.first().centroid
+        val lastCentroid = clusters.last().centroid
+        val drift = 1 - cosineSimilarity(firstCentroid, lastCentroid)
+
+        // If embeddings changed significantly, likely a child
+        return drift >= CHILD_EMBEDDING_DRIFT_THRESHOLD
+    }
+
+    /**
+     * Estimate age in specific year based on cluster position
+     */
+    private fun estimateAgeInYear(targetYear: String, allClusters: List<YearCluster>): Int? {
+        val sortedClusters = allClusters.sortedBy { it.year }
+        val firstYear = sortedClusters.first().year.toInt()
+        val targetYearInt = targetYear.toInt()
+
+        val yearsSinceFirst = targetYearInt - firstYear
+        return yearsSinceFirst + 1 // Start at age 1
+    }
+
+    /**
+     * Select representative faces
+     */
+    private fun selectRepresentativeFaces(
+        faces: List<DetectedFaceWithEmbedding>,
+        count: Int
+    ): List<DetectedFaceWithEmbedding> {
+        if (faces.size <= count) return faces
+
+        val centroid = calculateCentroid(faces.map { it.embedding })
+
+        return faces
+            .map { face -> face to (1 - cosineSimilarity(face.embedding, centroid)) }
+            .sortedBy { it.second }
+            .take(count)
+            .map { it.first }
+    }
+
+    /**
+     * DBSCAN clustering
+     */
+    private fun performDBSCAN(
+        faces: List<DetectedFaceWithEmbedding>,
+        epsilon: Float,
+        minPoints: Int
+    ): List<RawCluster> {
+        val visited = mutableSetOf<Int>()
+        val clusters = mutableListOf<RawCluster>()
+        var clusterId = 0
+
+        for (i in faces.indices) {
+            if (i in visited) continue
+
+            val neighbors = findNeighbors(i, faces, epsilon)
+
+            if (neighbors.size < minPoints) {
+                visited.add(i)
+                continue
+            }
+
+            val cluster = mutableListOf<DetectedFaceWithEmbedding>()
+            val queue = ArrayDeque(neighbors)
+
+            while (queue.isNotEmpty()) {
+                val pointIdx = queue.removeFirst()
+                if (pointIdx in visited) continue
+
+                visited.add(pointIdx)
+                cluster.add(faces[pointIdx])
+
+                val pointNeighbors = findNeighbors(pointIdx, faces, epsilon)
+                if (pointNeighbors.size >= minPoints) {
+                    queue.addAll(pointNeighbors.filter { it !in visited })
+                }
+            }
+
+            if (cluster.size >= minPoints) {
+                clusters.add(RawCluster(clusterId++, cluster))
+            }
+        }
+
+        return clusters
+    }
+
+    private fun findNeighbors(
+        pointIdx: Int,
+        faces: List<DetectedFaceWithEmbedding>,
+        epsilon: Float
+    ): List<Int> {
+        val point = faces[pointIdx]
+        return faces.indices.filter { i ->
+            if (i == pointIdx) return@filter false
+            val similarity = cosineSimilarity(point.embedding, faces[i].embedding)
+            similarity > (1 - epsilon)
+        }
+    }
+
+    private fun cosineSimilarity(a: FloatArray, b: FloatArray): Float {
+        var dotProduct = 0f
+        var normA = 0f
+        var normB = 0f
+
+        for (i in a.indices) {
+            dotProduct += a[i] * b[i]
+            normA += a[i] * a[i]
+            normB += b[i] * b[i]
+        }
+
+        return dotProduct / (sqrt(normA) * sqrt(normB))
+    }
+
+    private fun calculateCentroid(embeddings: List<FloatArray>): FloatArray {
+        if (embeddings.isEmpty()) return FloatArray(0)
+
+        val size = embeddings.first().size
+        val centroid = FloatArray(size) { 0f }
+
+        embeddings.forEach { embedding ->
+            for (i in embedding.indices) {
+                centroid[i] += embedding[i]
+            }
+        }
+
+        val count = embeddings.size.toFloat()
+        for (i in centroid.indices) {
+            centroid[i] /= count
+        }
+
+        val norm = sqrt(centroid.map { it * it }.sum())
+        if (norm > 0) {
+            return centroid.map { it / norm }.toFloatArray()
+        }
+
+        return centroid
+    }
+
+    private fun loadBitmapDownsampled(uri: Uri, maxDim: Int): Bitmap? {
+        return try {
+            val opts = BitmapFactory.Options().apply { inJustDecodeBounds = true }
+            context.contentResolver.openInputStream(uri)?.use {
+                BitmapFactory.decodeStream(it, null, opts)
+            }
+
+            var sample = 1
+            while (opts.outWidth / sample > maxDim || opts.outHeight / sample > maxDim) {
+                sample *= 2
+            }
+
+            val finalOpts = BitmapFactory.Options().apply {
+                inSampleSize = sample
+                inPreferredConfig = Bitmap.Config.RGB_565
+            }
+
+            context.contentResolver.openInputStream(uri)?.use {
+                BitmapFactory.decodeStream(it, null, finalOpts)
+            }
+        } catch (e: Exception) {
+            null
+        }
+    }
+}
+
+/**
+ * Year-specific cluster
+ */
+data class YearCluster(
+    val year: String,
+    val faces: List<DetectedFaceWithEmbedding>,
+    val centroid: FloatArray
+)
+
+/**
+ * Group of year clusters belonging to same person
+ */
+data class PersonGroup(
+    val clusters: List<YearCluster>
+)
+
+/**
+ * Annotated cluster with temporal metadata
+ */
+data class AnnotatedCluster(
+    val cluster: FaceCluster,
+    val year: String,
+    val isChild: Boolean,
+    val suggestedName: String?,
+    val suggestedAge: Int?
+) {
+    /**
+     * Generate tag for this cluster
+     * Examples:
+     * - Child: "Emma_2020" or "Emma_Age_2"
+     * - Adult: "Brad_Pitt"
+     */
+    fun generateTag(name: String): String {
+        return if (isChild) {
+            if (suggestedAge != null) {
+                "${name}_Age_${suggestedAge}"
+            } else {
+                "${name}_${year}"
+            }
+        } else {
+            name
+        }
+    }
+}
+
+/**
+ * Result of temporal clustering
+ */
+data class TemporalClusteringResult(
+    val clusters: List<AnnotatedCluster>,
+    val totalPhotosProcessed: Int,
+    val totalFacesDetected: Int,
+    val processingTimeMs: Long,
+    val errorMessage: String? = null
+)
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeoplescreen.kt b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeoplescreen.kt
index 64a2630..ee9d6c5 100644
--- a/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeoplescreen.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeoplescreen.kt
@@ -22,6 +22,7 @@ import com.placeholder.sherpai2.domain.clustering.ClusterQualityAnalyzer
  * - ✅ Complete naming dialog integration
  * - ✅ Quality analysis in cluster grid
  * - ✅ Better error handling
+ * - ✅ Refinement flow support
  */
 @OptIn(ExperimentalMaterial3Api::class)
 @Composable
@@ -112,6 +113,12 @@ fun DiscoverPeopleScreen(
                 ValidationPreviewScreen(
                     personName = state.personName,
                     validationResult = state.validationResult,
+                    onMarkFeedback = { feedbackMap ->
+                        viewModel.submitFeedback(state.cluster, feedbackMap)
+                    },
+                    onRequestRefinement = {
+                        viewModel.requestRefinement(state.cluster)
+                    },
                     onApprove = {
                         viewModel.approveValidationAndScan(
                             personId = state.personId,
@@ -124,6 +131,28 @@ fun DiscoverPeopleScreen(
                 )
             }
 
+            // ===== REFINEMENT NEEDED =====
+            is DiscoverUiState.RefinementNeeded -> {
+                RefinementNeededContent(
+                    recommendation = state.recommendation,
+                    currentIteration = state.currentIteration,
+                    onRefine = {
+                        viewModel.requestRefinement(state.cluster)
+                    },
+                    onSkip = {
+                        viewModel.reset()
+                    }
+                )
+            }
+
+            // ===== REFINING IN PROGRESS =====
+            is DiscoverUiState.Refining -> {
+                RefiningProgressContent(
+                    iteration = state.iteration,
+                    message = state.message
+                )
+            }
+
             // ===== COMPLETE =====
             is DiscoverUiState.Complete -> {
                 CompleteStateContent(
@@ -154,6 +183,7 @@ fun DiscoverPeopleScreen(
         }
     }
 }
+
 // ===== IDLE STATE CONTENT =====
 
 @Composable
@@ -239,7 +269,7 @@ private fun ClusteringProgressContent(
 
         if (total > 0) {
             LinearProgressIndicator(
-                progress = progress.toFloat() / total.toFloat(),
+                progress = { progress.toFloat() / total.toFloat() },
                 modifier = Modifier
                     .fillMaxWidth()
                     .height(8.dp)
@@ -287,7 +317,7 @@ private fun TrainingProgressContent(
             Spacer(modifier = Modifier.height(16.dp))
 
             LinearProgressIndicator(
-                progress = progress.toFloat() / total.toFloat(),
+                progress = { progress.toFloat() / total.toFloat() },
                 modifier = Modifier
                     .fillMaxWidth()
                     .height(8.dp)
@@ -304,6 +334,131 @@ private fun TrainingProgressContent(
     }
 }
 
+// ===== REFINEMENT NEEDED =====
+
+@Composable
+private fun RefinementNeededContent(
+    recommendation: com.placeholder.sherpai2.domain.clustering.RefinementRecommendation,
+    currentIteration: Int,
+    onRefine: () -> Unit,
+    onSkip: () -> Unit
+) {
+    Column(
+        modifier = Modifier
+            .fillMaxSize()
+            .padding(24.dp),
+        horizontalAlignment = Alignment.CenterHorizontally,
+        verticalArrangement = Arrangement.Center
+    ) {
+        Icon(
+            imageVector = Icons.Default.Person,
+            contentDescription = null,
+            modifier = Modifier.size(80.dp),
+            tint = MaterialTheme.colorScheme.primary
+        )
+
+        Spacer(modifier = Modifier.height(24.dp))
+
+        Text(
+            text = "Refinement Recommended",
+            style = MaterialTheme.typography.headlineMedium,
+            fontWeight = FontWeight.Bold
+        )
+
+        Spacer(modifier = Modifier.height(16.dp))
+
+        Card(
+            colors = CardDefaults.cardColors(
+                containerColor = MaterialTheme.colorScheme.errorContainer
+            )
+        ) {
+            Column(
+                modifier = Modifier.padding(16.dp)
+            ) {
+                Text(
+                    text = recommendation.reason,
+                    style = MaterialTheme.typography.bodyMedium,
+                    color = MaterialTheme.colorScheme.onErrorContainer
+                )
+            }
+        }
+
+        Spacer(modifier = Modifier.height(16.dp))
+
+        Text(
+            text = "Iteration: $currentIteration",
+            style = MaterialTheme.typography.bodySmall,
+            color = MaterialTheme.colorScheme.onSurfaceVariant
+        )
+
+        Spacer(modifier = Modifier.height(32.dp))
+
+        Row(
+            modifier = Modifier.fillMaxWidth(),
+            horizontalArrangement = Arrangement.spacedBy(12.dp)
+        ) {
+            OutlinedButton(
+                onClick = onSkip,
+                modifier = Modifier.weight(1f)
+            ) {
+                Text("Skip")
+            }
+
+            Button(
+                onClick = onRefine,
+                modifier = Modifier.weight(1f)
+            ) {
+                Text("Refine Cluster")
+            }
+        }
+    }
+}
+
+// ===== REFINING PROGRESS =====
+
+@Composable
+private fun RefiningProgressContent(
+    iteration: Int,
+    message: String
+) {
+    Column(
+        modifier = Modifier
+            .fillMaxSize()
+            .padding(24.dp),
+        horizontalAlignment = Alignment.CenterHorizontally,
+        verticalArrangement = Arrangement.Center
+    ) {
+        CircularProgressIndicator(
+            modifier = Modifier.size(64.dp)
+        )
+
+        Spacer(modifier = Modifier.height(32.dp))
+
+        Text(
+            text = "Refining Cluster",
+            style = MaterialTheme.typography.titleLarge,
+            fontWeight = FontWeight.Bold
+        )
+
+        Spacer(modifier = Modifier.height(16.dp))
+
+        Text(
+            text = message,
+            style = MaterialTheme.typography.bodyMedium,
+            textAlign = TextAlign.Center,
+            color = MaterialTheme.colorScheme.onSurfaceVariant
+        )
+
+        Spacer(modifier = Modifier.height(8.dp))
+
+        Text(
+            text = "Iteration $iteration",
+            style = MaterialTheme.typography.bodySmall,
+            color = MaterialTheme.colorScheme.onSurfaceVariant
+        )
+    }
+}
+
 // ===== LOADING CONTENT =====
 
 @Composable
diff --git a/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeopleviewmodel.kt b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeopleviewmodel.kt
index a76e1e1..a7da09e 100644
--- a/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeopleviewmodel.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Discoverpeopleviewmodel.kt
@@ -2,9 +2,8 @@ package com.placeholder.sherpai2.ui.discover
 
 import androidx.lifecycle.ViewModel
 import androidx.lifecycle.viewModelScope
-import com.placeholder.sherpai2.domain.clustering.ClusteringResult
-import com.placeholder.sherpai2.domain.clustering.FaceCluster
-import com.placeholder.sherpai2.domain.clustering.FaceClusteringService
+import com.placeholder.sherpai2.data.local.entity.FeedbackType
+import com.placeholder.sherpai2.domain.clustering.*
 import com.placeholder.sherpai2.domain.training.ClusterTrainingService
 import com.placeholder.sherpai2.domain.validation.ValidationScanResult
 import com.placeholder.sherpai2.domain.validation.ValidationScanService
@@ -16,33 +15,49 @@ import kotlinx.coroutines.launch
 import javax.inject.Inject
 
 /**
- * DiscoverPeopleViewModel - COMPLETE workflow with validation
+ * DiscoverPeopleViewModel - COMPLETE workflow with feedback loop
  *
- * Flow:
+ * FLOW WITH REFINEMENT:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  * 1. Idle → Clustering → NamingReady (2x2 grid)
  * 2. Select cluster → NamingCluster (dialog)
  * 3. Confirm → AnalyzingCluster → Training → ValidationPreview
- * 4. Approve → Complete OR Reject → Error
+ * 4. User reviews faces → Marks correct/incorrect
+ * 5a. If too many incorrect → Refining (re-cluster without bad faces)
+ * 5b. If approved → Complete OR Reject → Error
+ * 6. Loop back to step 3 if refinement happened
+ *
+ * NEW FEATURES:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * ✅ User feedback collection
+ * ✅ Cluster refinement loop
+ * ✅ Feedback persistence
+ * ✅ Quality-aware training (only confirmed faces)
  */
 @HiltViewModel
 class DiscoverPeopleViewModel @Inject constructor(
     private val clusteringService: FaceClusteringService,
     private val trainingService: ClusterTrainingService,
-    private val validationService: ValidationScanService
+    private val validationService: ValidationScanService,
+    private val refinementService: ClusterRefinementService
 ) : ViewModel() {
 
     private val _uiState = MutableStateFlow<DiscoverUiState>(DiscoverUiState.Idle)
     val uiState: StateFlow<DiscoverUiState> = _uiState.asStateFlow()
 
     private val namedClusterIds = mutableSetOf<Int>()
+    private var currentIterationCount = 0
 
     fun startDiscovery() {
         viewModelScope.launch {
             try {
                 namedClusterIds.clear()
+                currentIterationCount = 0
                 _uiState.value = DiscoverUiState.Clustering(0, 100, "Starting...")
 
+                // Use PREMIUM_SOLO_ONLY strategy for best results
                 val result = clusteringService.discoverPeople(
+                    strategy = ClusteringStrategy.PREMIUM_SOLO_ONLY,
                     onProgress = { current: Int, total: Int, message: String ->
                         _uiState.value = DiscoverUiState.Clustering(current, total, message)
                     }
@@ -56,7 +71,7 @@ class DiscoverPeopleViewModel @Inject constructor(
                 if (result.clusters.isEmpty()) {
                     _uiState.value = DiscoverUiState.NoPeopleFound(
                         result.errorMessage
-                            ?: "No people clusters found.\n\nTry:\n• Adding more photos\n• Ensuring photos are clear\n• Having 3+ photos per person"
+                            ?: "No people clusters found.\n\nTry:\n• Adding more solo photos\n• Ensuring photos are clear\n• Having 6+ photos per person"
                     )
                 } else {
                     _uiState.value = DiscoverUiState.NamingReady(result)
@@ -131,10 +146,11 @@ class DiscoverPeopleViewModel @Inject constructor(
                     }
                 )
 
-                // Stage 4: Show validation preview
+                // Stage 4: Show validation preview WITH FEEDBACK SUPPORT
                 _uiState.value = DiscoverUiState.ValidationPreview(
                     personId = personId,
                     personName = name,
+                    cluster = cluster,
                     validationResult = validationResult
                 )
 
@@ -144,14 +160,112 @@ class DiscoverPeopleViewModel @Inject constructor(
         }
     }
 
+    /**
+     * NEW: Handle user feedback from validation preview
+     *
+     * @param cluster The cluster being validated
+     * @param feedbackMap Map of imageId → FeedbackType
+     */
+    fun submitFeedback(
+        cluster: FaceCluster,
+        feedbackMap: Map<String, FeedbackType>
+    ) {
+        viewModelScope.launch {
+            try {
+                // Convert imageId feedback to face feedback
+                val faceFeedbackMap = cluster.faces
+                    .associateWith { face ->
+                        feedbackMap[face.imageId] ?: FeedbackType.UNCERTAIN
+                    }
+
+                val originalConfidences = cluster.faces.associateWith { it.confidence }
+
+                // Store feedback
+                refinementService.storeFeedback(
+                    cluster = cluster,
+                    feedbackMap = faceFeedbackMap,
+                    originalConfidences = originalConfidences
+                )
+
+                // Check if refinement needed
+                val recommendation = refinementService.shouldRefineCluster(cluster)
+
+                if (recommendation.shouldRefine) {
+                    _uiState.value = DiscoverUiState.RefinementNeeded(
+                        cluster = cluster,
+                        recommendation = recommendation,
+                        currentIteration = currentIterationCount
+                    )
+                }
+            } catch (e: Exception) {
+                _uiState.value = DiscoverUiState.Error(
+                    "Failed to process feedback: ${e.message}"
+                )
+            }
+        }
+    }
+
+    /**
+     * NEW: Request cluster refinement
+     *
+     * Re-clusters WITHOUT rejected faces
+     */
+    fun requestRefinement(cluster: FaceCluster) {
+        viewModelScope.launch {
+            try {
+                currentIterationCount++
+
+                _uiState.value = DiscoverUiState.Refining(
+                    iteration = currentIterationCount,
+                    message = "Removing incorrect faces and re-clustering..."
+                )
+
+                // Refine cluster
+                val refinementResult = refinementService.refineCluster(
+                    cluster = cluster,
+                    iterationNumber = currentIterationCount
+                )
+
+                if (!refinementResult.success || refinementResult.refinedCluster == null) {
+                    _uiState.value = DiscoverUiState.Error(
+                        refinementResult.errorMessage
+                            ?: "Failed to refine cluster. Please try manual training."
+                    )
+                    return@launch
+                }
+
+                // Show refined cluster for re-validation
+                val currentState = _uiState.value
+                if (currentState is DiscoverUiState.RefinementNeeded) {
+                    // Re-train with refined cluster
+                    // This will loop back to ValidationPreview
+                    confirmClusterName(
+                        cluster = refinementResult.refinedCluster,
+                        name = currentState.cluster.representativeFaces.first().imageId, // Placeholder
+                        dateOfBirth = null,
+                        isChild = false,
+                        selectedSiblings = emptyList()
+                    )
+                }
+
+            } catch (e: Exception) {
+                _uiState.value = DiscoverUiState.Error(
+                    "Refinement failed: ${e.message}"
+                )
+            }
+        }
+    }
+
     fun approveValidationAndScan(personId: String, personName: String) {
         viewModelScope.launch {
             try {
-                // Mark cluster as named (find it from previous state)
+                // Mark cluster as named
                 // TODO: Track this properly
 
                 _uiState.value = DiscoverUiState.Complete(
-                    message = "Successfully created model for \"$personName\"!\n\nFull library scan has been queued in the background."
+                    message = "Successfully created model for \"$personName\"!\n\n" +
+                            "Full library scan has been queued in the background.\n\n" +
+                            "✅ ${currentIterationCount} refinement iterations completed"
                 )
             } catch (e: Exception) {
                 _uiState.value = DiscoverUiState.Error(e.message ?: "Failed to start library scan")
@@ -161,7 +275,8 @@ class DiscoverPeopleViewModel @Inject constructor(
 
     fun rejectValidationAndImprove() {
         _uiState.value = DiscoverUiState.Error(
-            "Please add more training photos and try again.\n\n(Feature coming: ability to add photos to existing model)"
+            "Please add more training photos and try again.\n\n" +
+                    "(Feature coming: ability to add photos to existing model)"
         )
     }
 
@@ -175,9 +290,13 @@ class DiscoverPeopleViewModel @Inject constructor(
     fun reset() {
         _uiState.value = DiscoverUiState.Idle
         namedClusterIds.clear()
+        currentIterationCount = 0
     }
 }
 
+/**
+ * UI States - ENHANCED with refinement states
+ */
 sealed class DiscoverUiState {
     object Idle : DiscoverUiState()
 
@@ -205,12 +324,33 @@ sealed class DiscoverUiState {
         val total: Int
     ) : DiscoverUiState()
 
+    /**
+     * NEW: Validation with feedback support
+     */
     data class ValidationPreview(
         val personId: String,
         val personName: String,
+        val cluster: FaceCluster,
         val validationResult: ValidationScanResult
     ) : DiscoverUiState()
 
+    /**
+     * NEW: Refinement needed state
+     */
+    data class RefinementNeeded(
+        val cluster: FaceCluster,
+        val recommendation: RefinementRecommendation,
+        val currentIteration: Int
+    ) : DiscoverUiState()
+
+    /**
+     * NEW: Refining in progress
+     */
+    data class Refining(
+        val iteration: Int,
+        val message: String
+    ) : DiscoverUiState()
+
     data class Complete(
         val message: String
     ) : DiscoverUiState()
diff --git a/app/src/main/java/com/placeholder/sherpai2/ui/discover/Temporalnamingdialog.kt b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Temporalnamingdialog.kt
new file mode 100644
index 0000000..3af7b43
--- /dev/null
+++ b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Temporalnamingdialog.kt
@@ -0,0 +1,353 @@
+package com.placeholder.sherpai2.ui.discover
+
+import androidx.compose.foundation.layout.*
+import androidx.compose.foundation.text.KeyboardOptions
+import androidx.compose.material.icons.Icons
+import androidx.compose.material.icons.filled.*
+import androidx.compose.material3.*
+import androidx.compose.runtime.*
+import androidx.compose.ui.Alignment
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.text.font.FontWeight
+import androidx.compose.ui.text.input.KeyboardType
+import androidx.compose.ui.unit.dp
+import androidx.compose.ui.window.Dialog
+import com.placeholder.sherpai2.domain.clustering.AnnotatedCluster
+import com.placeholder.sherpai2.domain.clustering.ClusterQualityAnalyzer
+import com.placeholder.sherpai2.domain.clustering.ClusterQualityResult
+
+/**
+ * TemporalNamingDialog - ENHANCED with age input for temporal clustering
+ *
+ * NEW FEATURES:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * ✅ Name field: "Emma"
+ * ✅ Age field: "2" (optional but recommended)
+ * ✅ Year display: "Photos from 2020"
+ * ✅ Auto-suggest: If year=2020 and DOB=2018 → Age=2
+ *
+ * NAMING PATTERNS:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * Adults:
+ * - Name: "John Doe"
+ * - Age: (leave empty)
+ * - Result: Person "John Doe" with single model
+ *
+ * Children (with age):
+ * - Name: "Emma"
+ * - Age: "2"
+ * - Year: "2020"
+ * - Result: Person "Emma" with submodel "Emma_Age_2"
+ *
+ * Children (without age):
+ * - Name: "Emma"
+ * - Age: (empty)
+ * - Year: "2020"
+ * - Result: Person "Emma" with submodel "Emma_2020"
+ */
+@Composable
+fun TemporalNamingDialog(
+    annotatedCluster: AnnotatedCluster,
+    onConfirm: (name: String, age: Int?, isChild: Boolean) -> Unit,
+    onDismiss: () -> Unit,
+    qualityAnalyzer: ClusterQualityAnalyzer
+) {
+    var name by remember { mutableStateOf(annotatedCluster.suggestedName ?: "") }
+    var ageText by remember { mutableStateOf(annotatedCluster.suggestedAge?.toString() ?: "") }
+    var isChild by remember { mutableStateOf(annotatedCluster.suggestedAge != null) }
+
+    // Analyze cluster quality
+    val qualityResult = remember(annotatedCluster.cluster) {
+        qualityAnalyzer.analyzeCluster(annotatedCluster.cluster)
+    }
+
+    Dialog(onDismissRequest = onDismiss) {
+        Card(
+            modifier = Modifier
+                .fillMaxWidth()
+                .padding(16.dp)
+        ) {
+            Column(
+                modifier = Modifier.padding(24.dp),
+                verticalArrangement = Arrangement.spacedBy(16.dp)
+            ) {
+                // Header
+                Text(
+                    text = "Name This Person",
+                    style = MaterialTheme.typography.headlineSmall,
+                    fontWeight = FontWeight.Bold
+                )
+
+                // Year badge
+                YearBadge(year = annotatedCluster.year)
+
+                HorizontalDivider()
+
+                // Quality warnings
+                QualityWarnings(qualityResult)
+
+                // Name field
+                OutlinedTextField(
+                    value = name,
+                    onValueChange = { name = it },
+                    label = { Text("Name") },
+                    placeholder = { Text("e.g., Emma") },
+                    leadingIcon = {
+                        Icon(Icons.Default.Person, contentDescription = null)
+                    },
+                    modifier = Modifier.fillMaxWidth(),
+                    singleLine = true
+                )
+
+                // Child checkbox
+                Row(
+                    modifier = Modifier.fillMaxWidth(),
+                    verticalAlignment = Alignment.CenterVertically
+                ) {
+                    Checkbox(
+                        checked = isChild,
+                        onCheckedChange = { isChild = it }
+                    )
+                    Spacer(modifier = Modifier.width(8.dp))
+                    Column {
+                        Text(
+                            text = "This is a child",
+                            style = MaterialTheme.typography.bodyMedium
+                        )
+                        Text(
+                            text = "Enable age-specific models",
+                            style = MaterialTheme.typography.bodySmall,
+                            color = MaterialTheme.colorScheme.onSurfaceVariant
+                        )
+                    }
+                }
+
+                // Age field (only if child)
+                if (isChild) {
+                    OutlinedTextField(
+                        value = ageText,
+                        onValueChange = {
+                            // Only allow numbers
+                            if (it.isEmpty() || it.all { c -> c.isDigit() }) {
+                                ageText = it
+                            }
+                        },
+                        label = { Text("Age in ${annotatedCluster.year}") },
+                        placeholder = { Text("e.g., 2") },
+                        leadingIcon = {
+                            Icon(Icons.Default.DateRange, contentDescription = null)
+                        },
+                        modifier = Modifier.fillMaxWidth(),
+                        singleLine = true,
+                        keyboardOptions = KeyboardOptions(keyboardType = KeyboardType.Number),
+                        supportingText = {
+                            Text("Optional: Helps create age-specific models")
+                        }
+                    )
+
+                    // Model name preview
+                    if (name.isNotBlank()) {
+                        Card(
+                            colors = CardDefaults.cardColors(
+                                containerColor = MaterialTheme.colorScheme.primaryContainer
+                            )
+                        ) {
+                            Row(
+                                modifier = Modifier
+                                    .fillMaxWidth()
+                                    .padding(12.dp),
+                                verticalAlignment = Alignment.CenterVertically
+                            ) {
+                                Icon(
+                                    imageVector = Icons.Default.Info,
+                                    contentDescription = null,
+                                    tint = MaterialTheme.colorScheme.onPrimaryContainer
+                                )
+                                Spacer(modifier = Modifier.width(8.dp))
+                                Column {
+                                    Text(
+                                        text = "Model will be created as:",
+                                        style = MaterialTheme.typography.bodySmall,
+                                        color = MaterialTheme.colorScheme.onPrimaryContainer
+                                    )
+                                    Text(
+                                        text = buildModelName(name, ageText, annotatedCluster.year),
+                                        style = MaterialTheme.typography.bodyMedium,
+                                        fontWeight = FontWeight.Bold,
+                                        color = MaterialTheme.colorScheme.onPrimaryContainer
+                                    )
+                                }
+                            }
+                        }
+                    }
+                }
+
+                // Cluster stats
+                ClusterStats(qualityResult)
+
+                HorizontalDivider()
+
+                // Actions
+                Row(
+                    modifier = Modifier.fillMaxWidth(),
+                    horizontalArrangement = Arrangement.spacedBy(12.dp)
+                ) {
+                    OutlinedButton(
+                        onClick = onDismiss,
+                        modifier = Modifier.weight(1f)
+                    ) {
+                        Text("Cancel")
+                    }
+
+                    Button(
+                        onClick = {
+                            val age = ageText.toIntOrNull()
+                            onConfirm(name, age, isChild)
+                        },
+                        modifier = Modifier.weight(1f),
+                        enabled = name.isNotBlank() && qualityResult.canTrain
+                    ) {
+                        Text("Create")
+                    }
+                }
+            }
+        }
+    }
+}
+
+/**
+ * Year badge showing photo year
+ */
+@Composable
+private fun YearBadge(year: String) {
+    Surface(
+        color = MaterialTheme.colorScheme.secondaryContainer,
+        shape = MaterialTheme.shapes.small
+    ) {
+        Row(
+            modifier = Modifier.padding(horizontal = 12.dp, vertical = 6.dp),
+            verticalAlignment = Alignment.CenterVertically,
+            horizontalArrangement = Arrangement.spacedBy(4.dp)
+        ) {
+            Icon(
+                imageVector = Icons.Default.DateRange,
+                contentDescription = null,
+                modifier = Modifier.size(16.dp),
+                tint = MaterialTheme.colorScheme.onSecondaryContainer
+            )
+            Text(
+                text = "Photos from $year",
+                style = MaterialTheme.typography.labelMedium,
+                color = MaterialTheme.colorScheme.onSecondaryContainer
+            )
+        }
+    }
+}
+
+/**
+ * Quality warnings
+ */
+@Composable
+private fun QualityWarnings(qualityResult: ClusterQualityResult) {
+    if (qualityResult.warnings.isNotEmpty()) {
+        Card(
+            colors = CardDefaults.cardColors(
+                containerColor = when (qualityResult.qualityTier) {
+                    com.placeholder.sherpai2.domain.clustering.ClusterQualityTier.POOR ->
+                        MaterialTheme.colorScheme.errorContainer
+                    com.placeholder.sherpai2.domain.clustering.ClusterQualityTier.GOOD ->
+                        MaterialTheme.colorScheme.tertiaryContainer
+                    else -> MaterialTheme.colorScheme.surfaceVariant
+                }
+            )
+        ) {
+            Column(
+                modifier = Modifier.padding(12.dp),
+                verticalArrangement = Arrangement.spacedBy(4.dp)
+            ) {
+                qualityResult.warnings.take(3).forEach { warning ->
+                    Row(
+                        verticalAlignment = Alignment.Top,
+                        horizontalArrangement = Arrangement.spacedBy(8.dp)
+                    ) {
+                        Icon(
+                            imageVector = when (qualityResult.qualityTier) {
+                                com.placeholder.sherpai2.domain.clustering.ClusterQualityTier.POOR ->
+                                    Icons.Default.Warning
+                                else -> Icons.Default.Info
+                            },
+                            contentDescription = null,
+                            modifier = Modifier.size(16.dp),
+                            tint = when (qualityResult.qualityTier) {
+                                com.placeholder.sherpai2.domain.clustering.ClusterQualityTier.POOR ->
+                                    MaterialTheme.colorScheme.onErrorContainer
+                                else -> MaterialTheme.colorScheme.onSurfaceVariant
+                            }
+                        )
+                        Text(
+                            text = warning,
+                            style = MaterialTheme.typography.bodySmall,
+                            color = when (qualityResult.qualityTier) {
+                                com.placeholder.sherpai2.domain.clustering.ClusterQualityTier.POOR ->
+                                    MaterialTheme.colorScheme.onErrorContainer
+                                else -> MaterialTheme.colorScheme.onSurfaceVariant
+                            }
+                        )
+                    }
+                }
+            }
+        }
+    }
+}
+
+/**
+ * Cluster statistics
+ */
+@Composable
+private fun ClusterStats(qualityResult: ClusterQualityResult) {
+    Row(
+        modifier = Modifier.fillMaxWidth(),
+        horizontalArrangement = Arrangement.SpaceEvenly
+    ) {
+        StatItem(
+            label = "Photos",
+            value = qualityResult.soloPhotoCount.toString()
+        )
+        StatItem(
+            label = "Clean Faces",
+            value = qualityResult.cleanFaceCount.toString()
+        )
+        StatItem(
+            label = "Quality",
+            value = "${(qualityResult.qualityScore * 100).toInt()}%"
+        )
+    }
+}
+
+@Composable
+private fun StatItem(label: String, value: String) {
+    Column(
+        horizontalAlignment = Alignment.CenterHorizontally
+    ) {
+        Text(
+            text = value,
+            style = MaterialTheme.typography.titleMedium,
+            fontWeight = FontWeight.Bold
+        )
+        Text(
+            text = label,
+            style = MaterialTheme.typography.bodySmall,
+            color = MaterialTheme.colorScheme.onSurfaceVariant
+        )
+    }
+}
+
+/**
+ * Build model name preview
+ */
+private fun buildModelName(name: String, ageText: String, year: String): String {
+    return when {
+        ageText.isNotBlank() -> "${name}_Age_${ageText}"
+        else -> "${name}_${year}"
+    }
+}
\ No newline at end of file
diff --git a/app/src/main/java/com/placeholder/sherpai2/ui/discover/Validationpreviewscreen.kt b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Validationpreviewscreen.kt
index e9aaa6b..0a7b61a 100644
--- a/app/src/main/java/com/placeholder/sherpai2/ui/discover/Validationpreviewscreen.kt
+++ b/app/src/main/java/com/placeholder/sherpai2/ui/discover/Validationpreviewscreen.kt
@@ -1,12 +1,17 @@
 package com.placeholder.sherpai2.ui.discover
 
 import android.net.Uri
+import androidx.compose.animation.AnimatedVisibility
+import androidx.compose.animation.core.animateFloatAsState
 import androidx.compose.foundation.background
 import androidx.compose.foundation.border
+import androidx.compose.foundation.clickable
+import androidx.compose.foundation.gestures.detectDragGestures
 import androidx.compose.foundation.layout.*
 import androidx.compose.foundation.lazy.grid.GridCells
 import androidx.compose.foundation.lazy.grid.LazyVerticalGrid
 import androidx.compose.foundation.lazy.grid.items
+import androidx.compose.foundation.shape.CircleShape
 import androidx.compose.foundation.shape.RoundedCornerShape
 import androidx.compose.material.icons.Icons
 import androidx.compose.material.icons.filled.*
@@ -15,268 +20,458 @@ import androidx.compose.runtime.*
 import androidx.compose.ui.Alignment
 import androidx.compose.ui.Modifier
 import androidx.compose.ui.draw.clip
+import androidx.compose.ui.draw.scale
 import androidx.compose.ui.graphics.Color
+import androidx.compose.ui.input.pointer.pointerInput
 import androidx.compose.ui.layout.ContentScale
 import androidx.compose.ui.text.font.FontWeight
+import androidx.compose.ui.text.style.TextAlign
+import androidx.compose.ui.unit.IntOffset
 import androidx.compose.ui.unit.dp
+import androidx.compose.ui.zIndex
 import coil.compose.AsyncImage
-import com.placeholder.sherpai2.domain.validation.ValidationMatch
-import com.placeholder.sherpai2.domain.validation.ValidationQuality
+import com.placeholder.sherpai2.data.local.entity.FeedbackType
 import com.placeholder.sherpai2.domain.validation.ValidationScanResult
+import com.placeholder.sherpai2.domain.validation.ValidationMatch
+import kotlin.math.roundToInt
 
 /**
- * ValidationPreviewScreen - STAGE 2 validation UI
+ * ValidationPreviewScreen - User reviews validation results with swipe gestures
  *
- * Shows user a preview of matches found in validation scan
- * User can approve (→ full scan) or reject (→ add more photos)
+ * FEATURES:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * ✅ Swipe right (✓) = Confirmed match
+ * ✅ Swipe left (✗) = Rejected match
+ * ✅ Tap = Mark uncertain (?)
+ * ✅ Real-time feedback stats
+ * ✅ Automatic refinement recommendation
+ * ✅ Bottom bar with approve/reject/refine actions
+ *
+ * FLOW:
+ * ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+ * 1. User swipes/taps to mark faces
+ * 2. Feedback tracked in local state
+ * 3. If >15% rejection → "Refine" button appears
+ * 4. Approve → Sends feedback map to ViewModel
+ * 5. Reject → Returns to previous screen
+ * 6. Refine → Triggers cluster refinement
  */
 @Composable
 fun ValidationPreviewScreen(
     personName: String,
     validationResult: ValidationScanResult,
+    onMarkFeedback: (Map<String, FeedbackType>) -> Unit = {},
+    onRequestRefinement: () -> Unit = {},
     onApprove: () -> Unit,
     onReject: () -> Unit,
     modifier: Modifier = Modifier
 ) {
-    Column(
-        modifier = modifier
-            .fillMaxSize()
-            .padding(16.dp)
-    ) {
-        // Header
-        Text(
-            text = "Validation Results",
-            style = MaterialTheme.typography.headlineMedium,
-            fontWeight = FontWeight.Bold
-        )
+    // Get sample images from validation result matches
+    val sampleMatches = remember(validationResult) {
+        validationResult.matches.take(24) // Show up to 24 faces
+    }
 
-        Spacer(modifier = Modifier.height(8.dp))
+    // Track feedback for each image (imageId -> FeedbackType)
+    var feedbackMap by remember {
+        mutableStateOf<Map<String, FeedbackType>>(emptyMap())
+    }
 
-        Text(
-            text = "Review matches for \"$personName\" before scanning your entire library",
-            style = MaterialTheme.typography.bodyMedium,
-            color = MaterialTheme.colorScheme.onSurfaceVariant
-        )
+    // Calculate feedback statistics
+    val confirmedCount = feedbackMap.count { it.value == FeedbackType.CONFIRMED_MATCH }
+    val rejectedCount = feedbackMap.count { it.value == FeedbackType.REJECTED_MATCH }
+    val uncertainCount = feedbackMap.count { it.value == FeedbackType.UNCERTAIN }
+    val reviewedCount = feedbackMap.size
+    val totalCount = sampleMatches.size
 
-        Spacer(modifier = Modifier.height(16.dp))
+    // Determine if refinement is recommended
+    val rejectionRatio = if (reviewedCount > 0) {
+        rejectedCount.toFloat() / reviewedCount.toFloat()
+    } else {
+        0f
+    }
+    val shouldRefine = rejectionRatio > 0.15f && rejectedCount >= 2
 
-        // Quality Summary
-        QualitySummaryCard(
-            validationResult = validationResult,
-            personName = personName
-        )
-
-        Spacer(modifier = Modifier.height(16.dp))
-
-        // Matches Grid
-        if (validationResult.matches.isNotEmpty()) {
+    Scaffold(
+        bottomBar = {
+            ValidationBottomBar(
+                confirmedCount = confirmedCount,
+                rejectedCount = rejectedCount,
+                uncertainCount = uncertainCount,
+                reviewedCount = reviewedCount,
+                totalCount = totalCount,
+                shouldRefine = shouldRefine,
+                onApprove = {
+                    onMarkFeedback(feedbackMap)
+                    onApprove()
+                },
+                onReject = onReject,
+                onRefine = {
+                    onMarkFeedback(feedbackMap)
+                    onRequestRefinement()
+                }
+            )
+        }
+    ) { paddingValues ->
+        Column(
+            modifier = modifier
+                .fillMaxSize()
+                .padding(paddingValues)
+                .padding(16.dp)
+        ) {
+            // Header
             Text(
-                text = "Sample Matches (${validationResult.matchCount})",
-                style = MaterialTheme.typography.titleMedium,
-                fontWeight = FontWeight.SemiBold
+                text = "Validate \"$personName\"",
+                style = MaterialTheme.typography.headlineMedium,
+                fontWeight = FontWeight.Bold
             )
 
             Spacer(modifier = Modifier.height(8.dp))
 
+            // Instructions
+            InstructionsCard()
+
+            Spacer(modifier = Modifier.height(16.dp))
+
+            // Feedback stats
+            FeedbackStatsCard(
+                confirmedCount = confirmedCount,
+                rejectedCount = rejectedCount,
+                uncertainCount = uncertainCount,
+                reviewedCount = reviewedCount,
+                totalCount = totalCount
+            )
+
+            Spacer(modifier = Modifier.height(16.dp))
+
+            // Grid of faces to review
             LazyVerticalGrid(
                 columns = GridCells.Fixed(3),
                 horizontalArrangement = Arrangement.spacedBy(8.dp),
                 verticalArrangement = Arrangement.spacedBy(8.dp),
                 modifier = Modifier.weight(1f)
             ) {
-                items(validationResult.matches.take(15)) { match ->
-                    MatchPreviewCard(match = match)
+                items(
+                    items = sampleMatches,
+                    key = { match -> match.imageId }
+                ) { match ->
+                    SwipeableFaceCard(
+                        match = match,
+                        currentFeedback = feedbackMap[match.imageId],
+                        onFeedbackChange = { feedback ->
+                            feedbackMap = feedbackMap.toMutableMap().apply {
+                                put(match.imageId, feedback)
+                            }
+                        }
+                    )
                 }
             }
-        } else {
-            // No matches found
-            NoMatchesCard()
-        }
-
-        Spacer(modifier = Modifier.height(16.dp))
-
-        // Action Buttons
-        Row(
-            modifier = Modifier.fillMaxWidth(),
-            horizontalArrangement = Arrangement.spacedBy(12.dp)
-        ) {
-            // Reject button
-            OutlinedButton(
-                onClick = onReject,
-                modifier = Modifier.weight(1f),
-                colors = ButtonDefaults.outlinedButtonColors(
-                    contentColor = MaterialTheme.colorScheme.error
-                )
-            ) {
-                Icon(
-                    imageVector = Icons.Default.Close,
-                    contentDescription = null,
-                    modifier = Modifier.size(20.dp)
-                )
-                Spacer(modifier = Modifier.width(8.dp))
-                Text("Add More Photos")
-            }
-
-            // Approve button
-            Button(
-                onClick = onApprove,
-                modifier = Modifier.weight(1f),
-                enabled = validationResult.qualityAssessment != ValidationQuality.NO_MATCHES
-            ) {
-                Icon(
-                    imageVector = Icons.Default.Check,
-                    contentDescription = null,
-                    modifier = Modifier.size(20.dp)
-                )
-                Spacer(modifier = Modifier.width(8.dp))
-                Text("Scan Library")
-            }
         }
     }
 }
 
+/**
+ * Swipeable face card with visual feedback indicators
+ */
 @Composable
-private fun QualitySummaryCard(
-    validationResult: ValidationScanResult,
-    personName: String
+private fun SwipeableFaceCard(
+    match: ValidationMatch,
+    currentFeedback: FeedbackType?,
+    onFeedbackChange: (FeedbackType) -> Unit
 ) {
-    val (backgroundColor, iconColor, statusText, statusIcon) = when (validationResult.qualityAssessment) {
-        ValidationQuality.EXCELLENT -> {
-            Quadruple(
-                Color(0xFF1B5E20).copy(alpha = 0.1f),
-                Color(0xFF1B5E20),
-                "Excellent Match Quality",
-                Icons.Default.CheckCircle
+    var offsetX by remember { mutableFloatStateOf(0f) }
+    var isDragging by remember { mutableStateOf(false) }
+
+    val scale by animateFloatAsState(
+        targetValue = if (isDragging) 1.1f else 1f,
+        label = "scale"
+    )
+
+    Box(
+        modifier = Modifier
+            .aspectRatio(1f)
+            .scale(scale)
+            .zIndex(if (isDragging) 1f else 0f)
+    ) {
+        // Face image with border color based on feedback
+        AsyncImage(
+            model = Uri.parse(match.imageUri),
+            contentDescription = "Face",
+            modifier = Modifier
+                .fillMaxSize()
+                .clip(RoundedCornerShape(12.dp))
+                .border(
+                    width = 3.dp,
+                    color = when (currentFeedback) {
+                        FeedbackType.CONFIRMED_MATCH -> Color(0xFF4CAF50) // Green
+                        FeedbackType.REJECTED_MATCH -> Color(0xFFF44336) // Red
+                        FeedbackType.UNCERTAIN -> Color(0xFFFF9800) // Orange
+                        else -> MaterialTheme.colorScheme.outline
+                    },
+                    shape = RoundedCornerShape(12.dp)
+                )
+                .offset { IntOffset(offsetX.roundToInt(), 0) }
+                .pointerInput(Unit) {
+                    detectDragGestures(
+                        onDragStart = {
+                            isDragging = true
+                        },
+                        onDrag = { _, dragAmount ->
+                            offsetX += dragAmount.x
+                        },
+                        onDragEnd = {
+                            isDragging = false
+
+                            // Determine feedback based on swipe direction
+                            when {
+                                offsetX > 100 -> {
+                                    onFeedbackChange(FeedbackType.CONFIRMED_MATCH)
+                                }
+                                offsetX < -100 -> {
+                                    onFeedbackChange(FeedbackType.REJECTED_MATCH)
+                                }
+                            }
+
+                            // Reset position
+                            offsetX = 0f
+                        },
+                        onDragCancel = {
+                            isDragging = false
+                            offsetX = 0f
+                        }
+                    )
+                }
+                .clickable {
+                    // Tap to toggle uncertain
+                    val newFeedback = when (currentFeedback) {
+                        FeedbackType.UNCERTAIN -> null
+                        else -> FeedbackType.UNCERTAIN
+                    }
+                    if (newFeedback != null) {
+                        onFeedbackChange(newFeedback)
+                    }
+                },
+            contentScale = ContentScale.Crop
+        )
+
+        // Confidence badge (top-left)
+        Surface(
+            modifier = Modifier
+                .align(Alignment.TopStart)
+                .padding(4.dp),
+            shape = RoundedCornerShape(4.dp),
+            color = Color.Black.copy(alpha = 0.6f)
+        ) {
+            Text(
+                text = "${(match.confidence * 100).toInt()}%",
+                modifier = Modifier.padding(horizontal = 6.dp, vertical = 2.dp),
+                style = MaterialTheme.typography.labelSmall,
+                color = Color.White,
+                fontWeight = FontWeight.Bold
             )
         }
-        ValidationQuality.GOOD -> {
-            Quadruple(
-                Color(0xFF2E7D32).copy(alpha = 0.1f),
-                Color(0xFF2E7D32),
-                "Good Match Quality",
-                Icons.Default.ThumbUp
-            )
+
+        // Feedback indicator overlay (top-right)
+        if (currentFeedback != null) {
+            Surface(
+                modifier = Modifier
+                    .align(Alignment.TopEnd)
+                    .padding(4.dp),
+                shape = CircleShape,
+                color = when (currentFeedback) {
+                    FeedbackType.CONFIRMED_MATCH -> Color(0xFF4CAF50)
+                    FeedbackType.REJECTED_MATCH -> Color(0xFFF44336)
+                    FeedbackType.UNCERTAIN -> Color(0xFFFF9800)
+                    else -> Color.Transparent
+                },
+                shadowElevation = 2.dp
+            ) {
+                Icon(
+                    imageVector = when (currentFeedback) {
+                        FeedbackType.CONFIRMED_MATCH -> Icons.Default.Check
+                        FeedbackType.REJECTED_MATCH -> Icons.Default.Close
+                        FeedbackType.UNCERTAIN -> Icons.Default.Warning
+                        else -> Icons.Default.Info
+                    },
+                    contentDescription = currentFeedback.name,
+                    tint = Color.White,
+                    modifier = Modifier
+                        .size(32.dp)
+                        .padding(6.dp)
+                )
+            }
         }
-        ValidationQuality.FAIR -> {
-            Quadruple(
-                Color(0xFFF57F17).copy(alpha = 0.1f),
-                Color(0xFFF57F17),
-                "Fair Match Quality",
-                Icons.Default.Warning
-            )
-        }
-        ValidationQuality.POOR -> {
-            Quadruple(
-                Color(0xFFD32F2F).copy(alpha = 0.1f),
-                Color(0xFFD32F2F),
-                "Poor Match Quality",
-                Icons.Default.Warning
-            )
-        }
-        ValidationQuality.NO_MATCHES -> {
-            Quadruple(
-                Color(0xFFD32F2F).copy(alpha = 0.1f),
-                Color(0xFFD32F2F),
-                "No Matches Found",
-                Icons.Default.Close
-            )
+
+        // Swipe hint during drag
+        if (isDragging) {
+            SwipeDragHint(offsetX = offsetX)
         }
     }
+}
 
+/**
+ * Swipe drag hint overlay
+ */
+@Composable
+private fun BoxScope.SwipeDragHint(offsetX: Float) {
+    val hintText = when {
+        offsetX > 50 -> "✓ Correct"
+        offsetX < -50 -> "✗ Incorrect"
+        else -> "Keep swiping"
+    }
+
+    val hintColor = when {
+        offsetX > 50 -> Color(0xFF4CAF50)
+        offsetX < -50 -> Color(0xFFF44336)
+        else -> Color.Gray
+    }
+
+    Surface(
+        modifier = Modifier
+            .align(Alignment.BottomCenter)
+            .padding(8.dp),
+        shape = RoundedCornerShape(4.dp),
+        color = hintColor.copy(alpha = 0.9f)
+    ) {
+        Text(
+            text = hintText,
+            modifier = Modifier.padding(horizontal = 8.dp, vertical = 4.dp),
+            style = MaterialTheme.typography.labelSmall,
+            color = Color.White,
+            fontWeight = FontWeight.Bold
+        )
+    }
+}
+
+/**
+ * Instructions card showing gesture controls
+ */
+@Composable
+private fun InstructionsCard() {
     Card(
-        modifier = Modifier.fillMaxWidth(),
         colors = CardDefaults.cardColors(
-            containerColor = backgroundColor
+            containerColor = MaterialTheme.colorScheme.primaryContainer
         )
     ) {
-        Column(
-            modifier = Modifier.padding(16.dp)
+        Row(
+            modifier = Modifier.padding(16.dp),
+            verticalAlignment = Alignment.CenterVertically
         ) {
-            Row(
-                verticalAlignment = Alignment.CenterVertically
-            ) {
-                Icon(
-                    imageVector = statusIcon,
-                    contentDescription = null,
-                    tint = iconColor,
-                    modifier = Modifier.size(24.dp)
-                )
-                Spacer(modifier = Modifier.width(8.dp))
+            Icon(
+                imageVector = Icons.Default.Info,
+                contentDescription = null,
+                tint = MaterialTheme.colorScheme.onPrimaryContainer
+            )
+
+            Spacer(modifier = Modifier.width(12.dp))
+
+            Column {
                 Text(
-                    text = statusText,
-                    style = MaterialTheme.typography.titleMedium,
+                    text = "Review Detected Faces",
+                    style = MaterialTheme.typography.titleSmall,
                     fontWeight = FontWeight.Bold,
-                    color = iconColor
+                    color = MaterialTheme.colorScheme.onPrimaryContainer
                 )
-            }
-
-            Spacer(modifier = Modifier.height(12.dp))
-
-            // Stats
-            Row(
-                modifier = Modifier.fillMaxWidth(),
-                horizontalArrangement = Arrangement.SpaceBetween
-            ) {
-                StatItem(
-                    label = "Matches Found",
-                    value = "${validationResult.matchCount} / ${validationResult.sampleSize}"
-                )
-                StatItem(
-                    label = "Avg Confidence",
-                    value = "${(validationResult.averageConfidence * 100).toInt()}%"
-                )
-                StatItem(
-                    label = "Threshold",
-                    value = "${(validationResult.threshold * 100).toInt()}%"
-                )
-            }
-
-            // Recommendation
-            if (validationResult.qualityAssessment != ValidationQuality.NO_MATCHES) {
-                Spacer(modifier = Modifier.height(12.dp))
-
-                val recommendation = when (validationResult.qualityAssessment) {
-                    ValidationQuality.EXCELLENT ->
-                        "✅ Model looks great! Safe to scan your full library."
-                    ValidationQuality.GOOD ->
-                        "✅ Model quality is good. You can proceed with the full scan."
-                    ValidationQuality.FAIR ->
-                        "⚠️ Model quality is acceptable but could be improved with more photos."
-                    ValidationQuality.POOR ->
-                        "⚠️ Consider adding more diverse, high-quality training photos."
-                    ValidationQuality.NO_MATCHES -> ""
-                }
-
+                Spacer(modifier = Modifier.height(4.dp))
                 Text(
-                    text = recommendation,
-                    style = MaterialTheme.typography.bodyMedium,
-                    color = MaterialTheme.colorScheme.onSurfaceVariant
-                )
-            } else {
-                Spacer(modifier = Modifier.height(12.dp))
-                Text(
-                    text = "No matches found. The model may need more or better training photos, or the validation sample didn't include $personName.",
-                    style = MaterialTheme.typography.bodyMedium,
-                    color = MaterialTheme.colorScheme.error
+                    text = "Swipe right ✅ for correct, left ❌ for incorrect, tap ❓ for uncertain",
+                    style = MaterialTheme.typography.bodySmall,
+                    color = MaterialTheme.colorScheme.onPrimaryContainer
                 )
             }
         }
     }
 }
 
+/**
+ * Feedback statistics card
+ */
 @Composable
-private fun StatItem(
-    label: String,
-    value: String
+private fun FeedbackStatsCard(
+    confirmedCount: Int,
+    rejectedCount: Int,
+    uncertainCount: Int,
+    reviewedCount: Int,
+    totalCount: Int
+) {
+    Card {
+        Row(
+            modifier = Modifier
+                .fillMaxWidth()
+                .padding(16.dp),
+            horizontalArrangement = Arrangement.SpaceEvenly
+        ) {
+            FeedbackStat(
+                icon = Icons.Default.Check,
+                color = Color(0xFF4CAF50),
+                count = confirmedCount,
+                label = "Correct"
+            )
+
+            FeedbackStat(
+                icon = Icons.Default.Close,
+                color = Color(0xFFF44336),
+                count = rejectedCount,
+                label = "Incorrect"
+            )
+
+            FeedbackStat(
+                icon = Icons.Default.Warning,
+                color = Color(0xFFFF9800),
+                count = uncertainCount,
+                label = "Uncertain"
+            )
+        }
+
+        val progressValue = if (totalCount > 0) {
+            reviewedCount.toFloat() / totalCount.toFloat()
+        } else {
+            0f
+        }
+
+        LinearProgressIndicator(
+            progress = { progressValue },
+            modifier = Modifier
+                .fillMaxWidth()
+                .height(4.dp)
+        )
+    }
+}
+
+/**
+ * Individual feedback statistic item
+ */
+@Composable
+private fun FeedbackStat(
+    icon: androidx.compose.ui.graphics.vector.ImageVector,
+    color: Color,
+    count: Int,
+    label: String
 ) {
     Column(
         horizontalAlignment = Alignment.CenterHorizontally
     ) {
+        Surface(
+            shape = CircleShape,
+            color = color.copy(alpha = 0.2f)
+        ) {
+            Icon(
+                imageVector = icon,
+                contentDescription = null,
+                tint = color,
+                modifier = Modifier
+                    .size(40.dp)
+                    .padding(8.dp)
+            )
+        }
+
+        Spacer(modifier = Modifier.height(4.dp))
+
         Text(
-            text = value,
-            style = MaterialTheme.typography.titleLarge,
+            text = count.toString(),
+            style = MaterialTheme.typography.titleMedium,
             fontWeight = FontWeight.Bold
         )
+
         Text(
             text = label,
             style = MaterialTheme.typography.bodySmall,
@@ -285,111 +480,134 @@ private fun StatItem(
     }
 }
 
+/**
+ * Bottom action bar with approve/reject/refine buttons
+ */
 @Composable
-private fun MatchPreviewCard(
-    match: ValidationMatch
+private fun ValidationBottomBar(
+    confirmedCount: Int,
+    rejectedCount: Int,
+    uncertainCount: Int,
+    reviewedCount: Int,
+    totalCount: Int,
+    shouldRefine: Boolean,
+    onApprove: () -> Unit,
+    onReject: () -> Unit,
+    onRefine: () -> Unit
 ) {
-    Box(
-        modifier = Modifier
-            .aspectRatio(1f)
-            .clip(RoundedCornerShape(8.dp))
-            .background(MaterialTheme.colorScheme.surfaceVariant)
+    Surface(
+        modifier = Modifier.fillMaxWidth(),
+        color = MaterialTheme.colorScheme.surface,
+        shadowElevation = 8.dp
     ) {
-        AsyncImage(
-            model = Uri.parse(match.imageUri),
-            contentDescription = "Match preview",
-            modifier = Modifier.fillMaxSize(),
-            contentScale = ContentScale.Crop
-        )
-
-        // Confidence badge
-        Surface(
-            modifier = Modifier
-                .align(Alignment.BottomEnd)
-                .padding(4.dp),
-            shape = RoundedCornerShape(4.dp),
-            color = Color.Black.copy(alpha = 0.7f)
+        Column(
+            modifier = Modifier.padding(16.dp)
         ) {
+            // Refinement warning banner
+            AnimatedVisibility(visible = shouldRefine) {
+                RefinementWarningBanner(
+                    rejectedCount = rejectedCount,
+                    reviewedCount = reviewedCount,
+                    onRefine = onRefine
+                )
+            }
+
+            // Main action buttons
+            Row(
+                modifier = Modifier.fillMaxWidth(),
+                horizontalArrangement = Arrangement.spacedBy(12.dp)
+            ) {
+                OutlinedButton(
+                    onClick = onReject,
+                    modifier = Modifier.weight(1f)
+                ) {
+                    Icon(Icons.Default.Close, contentDescription = null)
+                    Spacer(modifier = Modifier.width(8.dp))
+                    Text("Reject")
+                }
+
+                Button(
+                    onClick = onApprove,
+                    modifier = Modifier.weight(1f),
+                    enabled = confirmedCount > 0 || (reviewedCount == 0 && totalCount > 6)
+                ) {
+                    Icon(Icons.Default.Check, contentDescription = null)
+                    Spacer(modifier = Modifier.width(8.dp))
+                    Text("Approve")
+                }
+            }
+
+            // Review progress text
+            Spacer(modifier = Modifier.height(8.dp))
+
             Text(
-                text = "${(match.confidence * 100).toInt()}%",
-                style = MaterialTheme.typography.labelSmall,
-                color = Color.White,
-                modifier = Modifier.padding(horizontal = 6.dp, vertical = 2.dp)
+                text = if (reviewedCount == 0) {
+                    "Review faces above or approve to continue"
+                } else {
+                    "Reviewed $reviewedCount of $totalCount faces"
+                },
+                style = MaterialTheme.typography.bodySmall,
+                color = MaterialTheme.colorScheme.onSurfaceVariant,
+                textAlign = TextAlign.Center,
+                modifier = Modifier.fillMaxWidth()
             )
         }
+    }
+}
 
-        // Face count indicator (if group photo)
-        if (match.faceCount > 1) {
-            Surface(
-                modifier = Modifier
-                    .align(Alignment.TopEnd)
-                    .padding(4.dp),
-                shape = RoundedCornerShape(4.dp),
-                color = MaterialTheme.colorScheme.primary
+/**
+ * Refinement warning banner component
+ */
+@Composable
+private fun RefinementWarningBanner(
+    rejectedCount: Int,
+    reviewedCount: Int,
+    onRefine: () -> Unit
+) {
+    Column {
+        Card(
+            colors = CardDefaults.cardColors(
+                containerColor = MaterialTheme.colorScheme.errorContainer
+            ),
+            modifier = Modifier.fillMaxWidth()
+        ) {
+            Row(
+                modifier = Modifier.padding(12.dp),
+                verticalAlignment = Alignment.CenterVertically
             ) {
-                Row(
-                    modifier = Modifier.padding(horizontal = 4.dp, vertical = 2.dp),
-                    verticalAlignment = Alignment.CenterVertically
-                ) {
-                    Icon(
-                        imageVector = Icons.Default.Person,
-                        contentDescription = null,
-                        tint = Color.White,
-                        modifier = Modifier.size(12.dp)
+                Icon(
+                    imageVector = Icons.Default.Warning,
+                    contentDescription = null,
+                    tint = MaterialTheme.colorScheme.onErrorContainer
+                )
+
+                Spacer(modifier = Modifier.width(12.dp))
+
+                Column(modifier = Modifier.weight(1f)) {
+                    Text(
+                        text = "High Rejection Rate",
+                        style = MaterialTheme.typography.titleSmall,
+                        fontWeight = FontWeight.Bold,
+                        color = MaterialTheme.colorScheme.onErrorContainer
                     )
                     Text(
-                        text = "${match.faceCount}",
-                        style = MaterialTheme.typography.labelSmall,
-                        color = Color.White
+                        text = "${(rejectedCount.toFloat() / reviewedCount.toFloat() * 100).toInt()}% rejected. Consider refining the cluster.",
+                        style = MaterialTheme.typography.bodySmall,
+                        color = MaterialTheme.colorScheme.onErrorContainer
                     )
                 }
+
+                Button(
+                    onClick = onRefine,
+                    colors = ButtonDefaults.buttonColors(
+                        containerColor = MaterialTheme.colorScheme.error
+                    )
+                ) {
+                    Text("Refine")
+                }
             }
         }
-    }
-}
 
-@Composable
-private fun NoMatchesCard() {
-    Card(
-        modifier = Modifier.fillMaxWidth(),
-        colors = CardDefaults.cardColors(
-            containerColor = MaterialTheme.colorScheme.errorContainer
-        )
-    ) {
-        Column(
-            modifier = Modifier.padding(24.dp),
-            horizontalAlignment = Alignment.CenterHorizontally
-        ) {
-            Icon(
-                imageVector = Icons.Default.Warning,
-                contentDescription = null,
-                tint = MaterialTheme.colorScheme.error,
-                modifier = Modifier.size(48.dp)
-            )
-            Spacer(modifier = Modifier.height(16.dp))
-            Text(
-                text = "No Matches Found",
-                style = MaterialTheme.typography.titleLarge,
-                fontWeight = FontWeight.Bold,
-                color = MaterialTheme.colorScheme.error
-            )
-            Spacer(modifier = Modifier.height(8.dp))
-            Text(
-                text = "The validation scan didn't find this person in the sample photos. This could mean:\n\n" +
-                        "• The model needs more training photos\n" +
-                        "• The training photos weren't diverse enough\n" +
-                        "• The person wasn't in the validation sample",
-                style = MaterialTheme.typography.bodyMedium,
-                color = MaterialTheme.colorScheme.onErrorContainer
-            )
-        }
+        Spacer(modifier = Modifier.height(12.dp))
     }
-}
-
-// Helper data class for quality indicator
-private data class Quadruple<A, B, C, D>(
-    val first: A,
-    val second: B,
-    val third: C,
-    val fourth: D
-)
\ No newline at end of file
+}
\ No newline at end of file