diff --git a/src/metrics.rs b/src/metrics.rs
index fac3baa3e..49a1f6ab9 100644
--- a/src/metrics.rs
+++ b/src/metrics.rs
@@ -34,6 +34,12 @@ pub struct Metrics {
     /// Number of blocks that were read from block cache
     pub(crate) data_block_load_cached: AtomicUsize,
 
+    /// Number of range tombstone blocks that were actually read from disk
+    pub(crate) range_tombstone_block_load_io: AtomicUsize,
+
+    /// Number of range tombstone blocks that were read from block cache
+    pub(crate) range_tombstone_block_load_cached: AtomicUsize,
+
     /// Number of filter queries that were performed
     pub(crate) filter_queries: AtomicUsize,
 
@@ -48,6 +54,9 @@ pub struct Metrics {
 
     /// Number of filter block bytes that were requested from OS or disk
     pub(crate) filter_block_io_requested: AtomicU64,
+
+    /// Number of range tombstone block bytes that were requested from OS or disk
+    pub(crate) range_tombstone_block_io_requested: AtomicU64,
 }
 
 #[expect(
@@ -82,11 +91,17 @@ impl Metrics {
         self.filter_block_io_requested.load(Relaxed)
     }
 
+    /// Number of I/O range tombstone block bytes transferred from disk or OS page cache.
+    pub fn range_tombstone_block_io(&self) -> u64 {
+        self.range_tombstone_block_io_requested.load(Relaxed)
+    }
+
     /// Number of I/O block bytes transferred from disk or OS page cache.
     pub fn block_io(&self) -> u64 {
         self.data_block_io_requested.load(Relaxed)
             + self.index_block_io_requested.load(Relaxed)
             + self.filter_block_io_requested.load(Relaxed)
+            + self.range_tombstone_block_io_requested.load(Relaxed)
     }
 
     /// Number of data blocks that were accessed.
@@ -104,33 +119,46 @@ impl Metrics {
         self.filter_block_load_cached.load(Relaxed) + self.filter_block_load_io.load(Relaxed)
     }
 
+    /// Number of range tombstone blocks that were accessed.
+    pub fn range_tombstone_block_load_count(&self) -> usize {
+        self.range_tombstone_block_load_cached.load(Relaxed)
+            + self.range_tombstone_block_load_io.load(Relaxed)
+    }
+
     /// Number of blocks that were loaded from disk or OS page cache.
     pub fn block_load_io_count(&self) -> usize {
         self.data_block_load_io.load(Relaxed)
             + self.index_block_load_io.load(Relaxed)
             + self.filter_block_load_io.load(Relaxed)
+            + self.range_tombstone_block_load_io.load(Relaxed)
     }
 
-    /// Number of data blocks that were loaded from disk or OS page cache.
+    /// Number of data blocks that were served from block cache.
     pub fn data_block_load_cached_count(&self) -> usize {
         self.data_block_load_cached.load(Relaxed)
     }
 
-    /// Number of index blocks that were loaded from disk or OS page cache.
+    /// Number of index blocks that were served from block cache.
     pub fn index_block_load_cached_count(&self) -> usize {
         self.index_block_load_cached.load(Relaxed)
     }
 
-    /// Number of filter blocks that were loaded from disk or OS page cache.
+    /// Number of filter blocks that were served from block cache.
     pub fn filter_block_load_cached_count(&self) -> usize {
         self.filter_block_load_cached.load(Relaxed)
     }
 
-    /// Number of blocks that were loaded from disk or OS page cache.
+    /// Number of range tombstone blocks that were served from block cache.
+    pub fn range_tombstone_block_load_cached_count(&self) -> usize {
+        self.range_tombstone_block_load_cached.load(Relaxed)
+    }
+
+    /// Number of blocks that were served from block cache.
     pub fn block_load_cached_count(&self) -> usize {
         self.data_block_load_cached.load(Relaxed)
             + self.index_block_load_cached.load(Relaxed)
             + self.filter_block_load_cached.load(Relaxed)
+            + self.range_tombstone_block_load_cached.load(Relaxed)
     }
 
     /// Number of blocks that were accessed.
@@ -174,6 +202,18 @@ impl Metrics {
         }
     }
 
+    /// Range tombstone block cache efficiency in percent (0.0 - 1.0).
+    pub fn range_tombstone_block_cache_hit_rate(&self) -> f64 {
+        let queries = self.range_tombstone_block_load_count() as f64;
+        let hits = self.range_tombstone_block_load_cached_count() as f64;
+
+        if queries == 0.0 {
+            1.0
+        } else {
+            hits / queries
+        }
+    }
+
     /// Block cache efficiency in percent (0.0 - 1.0).
     pub fn block_cache_hit_rate(&self) -> f64 {
         let queries = self.block_loads() as f64;
@@ -210,3 +250,69 @@ impl Metrics {
         self.io_skipped_by_filter.load(Relaxed)
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::sync::atomic::Ordering::Relaxed;
+
+    #[test]
+    fn range_tombstone_counters_default_zero() {
+        let m = Metrics::default();
+        assert_eq!(0, m.range_tombstone_block_load_count());
+        assert_eq!(0, m.range_tombstone_block_load_cached_count());
+        assert_eq!(0, m.range_tombstone_block_io());
+    }
+
+    #[test]
+    fn range_tombstone_block_load_count_sums_cached_and_io() {
+        let m = Metrics::default();
+        m.range_tombstone_block_load_cached.store(3, Relaxed);
+        m.range_tombstone_block_load_io.store(7, Relaxed);
+        assert_eq!(10, m.range_tombstone_block_load_count());
+    }
+
+    #[test]
+    fn range_tombstone_cache_hit_rate_no_loads_returns_one() {
+        let m = Metrics::default();
+        assert!((m.range_tombstone_block_cache_hit_rate() - 1.0).abs() < f64::EPSILON);
+    }
+
+    #[test]
+    fn range_tombstone_cache_hit_rate_mixed_loads() {
+        let m = Metrics::default();
+        m.range_tombstone_block_load_cached.store(3, Relaxed);
+        m.range_tombstone_block_load_io.store(1, Relaxed);
+        assert!((m.range_tombstone_block_cache_hit_rate() - 0.75).abs() < f64::EPSILON);
+    }
+
+    #[test]
+    fn block_io_includes_range_tombstone() {
+        let m = Metrics::default();
+        m.data_block_io_requested.store(10, Relaxed);
+        m.index_block_io_requested.store(20, Relaxed);
+        m.filter_block_io_requested.store(30, Relaxed);
+        m.range_tombstone_block_io_requested.store(40, Relaxed);
+        assert_eq!(100, m.block_io());
+    }
+
+    #[test]
+    fn block_load_io_count_includes_range_tombstone() {
+        let m = Metrics::default();
+        m.data_block_load_io.store(1, Relaxed);
+        m.index_block_load_io.store(2, Relaxed);
+        m.filter_block_load_io.store(3, Relaxed);
+        m.range_tombstone_block_load_io.store(4, Relaxed);
+        assert_eq!(10, m.block_load_io_count());
+    }
+
+    #[test]
+    fn block_load_cached_count_includes_range_tombstone() {
+        let m = Metrics::default();
+        m.data_block_load_cached.store(5, Relaxed);
+        m.index_block_load_cached.store(6, Relaxed);
+        m.filter_block_load_cached.store(7, Relaxed);
+        m.range_tombstone_block_load_cached.store(8, Relaxed);
+        assert_eq!(26, m.block_load_cached_count());
+    }
+}
diff --git a/src/table/tests.rs b/src/table/tests.rs
index 6df174d0f..91f82921b 100644
--- a/src/table/tests.rs
+++ b/src/table/tests.rs
@@ -1429,3 +1429,111 @@ fn table_global_seqno() -> crate::Result<()> {
 
     Ok(())
 }
+
+/// Exercises the `load_block` cache-miss and cache-hit paths for
+/// `BlockType::RangeTombstone`, verifying that the dedicated RT metrics
+/// counters are incremented instead of the data-block counters.
+#[test]
+#[cfg(feature = "metrics")]
+fn load_block_range_tombstone_metrics() -> crate::Result<()> {
+    use crate::{
+        cache::Cache,
+        descriptor_table::DescriptorTable,
+        range_tombstone::RangeTombstone,
+        table::{block::BlockType, util::load_block},
+        CompressionType,
+    };
+    use std::sync::atomic::Ordering::Relaxed;
+
+    let dir = tempdir()?;
+    let file = dir.path().join("table");
+
+    // Build a table that contains a range tombstone block.
+    let mut writer = Writer::new(file.clone(), 0, 0)?;
+    writer.write(InternalValue::from_components(
+        b"a",
+        b"v1",
+        1,
+        crate::ValueType::Value,
+    ))?;
+    writer.write(InternalValue::from_components(
+        b"z",
+        b"v2",
+        2,
+        crate::ValueType::Value,
+    ))?;
+    writer.write_range_tombstone(RangeTombstone::new(b"b".into(), b"y".into(), 3));
+    #[expect(
+        clippy::unwrap_used,
+        reason = "finish() returns Some after writing data items"
+    )]
+    let (_, checksum) = writer.finish()?.unwrap();
+
+    let metrics = Arc::new(crate::metrics::Metrics::default());
+
+    let table = Table::recover(
+        file,
+        checksum,
+        0,
+        0,
+        // Recovery bypasses load_block() (reads via Block::from_file() directly),
+        // so it intentionally does NOT increment block-load metrics — consistent
+        // with how filter and index recovery reads are handled.
+        Arc::new(Cache::with_capacity_bytes(10_000_000)),
+        Some(Arc::new(DescriptorTable::new(10))),
+        false,
+        false,
+        #[cfg(feature = "metrics")]
+        metrics.clone(),
+    )?;
+
+    let rt_handle = table
+        .regions
+        .range_tombstones
+        .expect("table should have range tombstone block");
+
+    let table_id = table.global_id();
+
+    // Recovery does NOT increment block-load counters (bypasses load_block).
+    assert_eq!(0, metrics.range_tombstone_block_load_io.load(Relaxed));
+
+    // Use a fresh cache so the first load_block() call is a cache miss.
+    let fresh_cache = Arc::new(Cache::with_capacity_bytes(10_000_000));
+
+    // load_block cache miss → IO path
+    let _block = load_block(
+        table_id,
+        &table.path,
+        &table.file_accessor,
+        &fresh_cache,
+        &rt_handle,
+        BlockType::RangeTombstone,
+        CompressionType::None,
+        #[cfg(feature = "metrics")]
+        &metrics,
+    )?;
+
+    assert_eq!(1, metrics.range_tombstone_block_load_io.load(Relaxed));
+    assert_eq!(0, metrics.range_tombstone_block_load_cached.load(Relaxed));
+    assert!(metrics.range_tombstone_block_io_requested.load(Relaxed) > 0);
+    assert_eq!(0, metrics.data_block_load_io.load(Relaxed));
+
+    // load_block cache hit (block was inserted into fresh_cache by previous call)
+    let _block = load_block(
+        table_id,
+        &table.path,
+        &table.file_accessor,
+        &fresh_cache,
+        &rt_handle,
+        BlockType::RangeTombstone,
+        CompressionType::None,
+        #[cfg(feature = "metrics")]
+        &metrics,
+    )?;
+
+    assert_eq!(1, metrics.range_tombstone_block_load_io.load(Relaxed));
+    assert_eq!(1, metrics.range_tombstone_block_load_cached.load(Relaxed));
+    assert_eq!(0, metrics.data_block_load_cached.load(Relaxed));
+
+    Ok(())
+}
diff --git a/src/table/util.rs b/src/table/util.rs
index c4aa913aa..a558e9c82 100644
--- a/src/table/util.rs
+++ b/src/table/util.rs
@@ -60,9 +60,12 @@ pub fn load_block(
             BlockType::Index => {
                 metrics.index_block_load_cached.fetch_add(1, Relaxed);
             }
-            // TODO(#34): RangeTombstone counted under data_block metrics — add
-            // dedicated range_tombstone_block_load_cached/miss counters
-            BlockType::Data | BlockType::Meta | BlockType::RangeTombstone => {
+            BlockType::RangeTombstone => {
+                metrics
+                    .range_tombstone_block_load_cached
+                    .fetch_add(1, Relaxed);
+            }
+            BlockType::Data | BlockType::Meta => {
                 metrics.data_block_load_cached.fetch_add(1, Relaxed);
             }
         }
@@ -109,8 +112,14 @@ pub fn load_block(
                 .index_block_io_requested
                 .fetch_add(handle.size().into(), Relaxed);
         }
-        // TODO(#34): same as above — RangeTombstone uses data_block IO counters
-        BlockType::Data | BlockType::Meta | BlockType::RangeTombstone => {
+        BlockType::RangeTombstone => {
+            metrics.range_tombstone_block_load_io.fetch_add(1, Relaxed);
+
+            metrics
+                .range_tombstone_block_io_requested
+                .fetch_add(handle.size().into(), Relaxed);
+        }
+        BlockType::Data | BlockType::Meta => {
             metrics.data_block_load_io.fetch_add(1, Relaxed);
 
             metrics