diff --git a/src/Cargo.lock b/src/Cargo.lock
index 40c259ba..1e1fcdfc 100644
--- a/src/Cargo.lock
+++ b/src/Cargo.lock
@@ -1524,7 +1524,10 @@ name = "obikdump"
 version = "0.1.0"
 dependencies = [
  "obikfilter",
+ "obikidxcache",
  "obikindex",
+ "obikseq",
+ "rayon",
 ]
 
 [[package]]
@@ -1539,6 +1542,7 @@ name = "obikfilter"
 version = "0.1.0"
 dependencies = [
  "obicompactvec",
+ "obikentropy",
  "obikindex",
  "obikseq",
  "obiskio",
diff --git a/src/obikdump/Cargo.toml b/src/obikdump/Cargo.toml
index b802741d..2d7ecc60 100644
--- a/src/obikdump/Cargo.toml
+++ b/src/obikdump/Cargo.toml
@@ -4,5 +4,8 @@ version = "0.1.0"
 edition = "2024"
 
 [dependencies]
-obikindex  = { path = "../obikindex" }
-obikfilter = { path = "../obikfilter" }
+obikindex   = { path = "../obikindex" }
+obikfilter  = { path = "../obikfilter" }
+obikidxcache = { path = "../obikidxcache" }
+obikseq     = { path = "../obikseq" }
+rayon       = "1"
diff --git a/src/obikdump/src/dump.rs b/src/obikdump/src/dump.rs
index 560ef95f..c3ae6b05 100644
--- a/src/obikdump/src/dump.rs
+++ b/src/obikdump/src/dump.rs
@@ -5,9 +5,14 @@ use rayon::prelude::*;
 
 use obikindex::{OKIError, OKIResult};
 use obikindex::KmerIndex;
+use obikidxcache::index_cache::IndexCache;
 use obikfilter::KmerFilter;
 
-impl KmerIndex {
+use crate::partition_iter::FilteredPartitionIter;
+
+/// Raw content export of a `KmerIndex` — `KmerIndex` is a foreign type
+/// (`obikindex`), so this is an extension trait rather than an inherent `impl`.
+pub trait IndexDump {
     /// Write a CSV table of all indexed kmers to `out`.
     ///
     /// Columns: `kmer`, then one column per genome (in index order).
@@ -17,11 +22,25 @@ impl KmerIndex {
     /// the output uses 0/1 presence columns.
     ///
     /// Partitions are scanned in parallel; each partition buffers its output locally
-    /// before the main thread writes the chunks in partition order.
+    /// before the main thread writes the chunks in partition order. Each partition's
+    /// layers are cached (`IndexCache`) only for the scan of that one partition —
+    /// `self` is a complete, read-only source index, never a destination.
     ///
     /// The caller must have set the global kmer length (`obikseq::set_k`) before
     /// calling this method.
-    pub fn dump<W: Write, F: Fn() + Send + Sync>(
+    fn dump<W: Write, F: Fn() + Send + Sync>(
+        &self,
+        out: &mut W,
+        force_presence: bool,
+        debug: bool,
+        head: Option<usize>,
+        filters: &[Box<dyn KmerFilter>],
+        on_partition: F,
+    ) -> OKIResult<()>;
+}
+
+impl IndexDump for KmerIndex {
+    fn dump<W: Write, F: Fn() + Send + Sync>(
         &self,
         out: &mut W,
         force_presence: bool,
@@ -30,8 +49,8 @@ impl KmerIndex {
         filters: &[Box<dyn KmerFilter>],
         on_partition: F,
     ) -> OKIResult<()> {
-        let genomes    = self.meta.genomes().map_err(OKIError::Io)?;
-        let use_counts = self.meta.config.with_counts && !force_presence;
+        let genomes    = self.meta().genomes().map_err(OKIError::Io)?;
+        let use_counts = self.meta().config.with_counts && !force_presence;
         let n_genomes  = genomes.len().max(1);
         let kmer_size  = self.kmer_size();
 
@@ -68,20 +87,17 @@ impl KmerIndex {
                         Ok(_)  => { write_row(buf, row, prefix); true }
                     }
                 };
+                let cache = IndexCache::new(self, Some(vec![i]));
                 if debug {
-                    self
-                        .iter_partition_kmers_located(i, use_counts, n_genomes, filters, |part, layer, kmer, row| {
-                            let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
-                            try_write(&mut buf, &row, &format!("{part},{layer},{seq}"))
-                        })
-                        .map_err(OKIError::Partition)?;
+                    cache.iter_partition_kmers_located(i, use_counts, n_genomes, filters, |part, layer, kmer, row| {
+                        let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
+                        try_write(&mut buf, &row, &format!("{part},{layer},{seq}"))
+                    })?;
                 } else {
-                    self
-                        .iter_partition_kmers(i, use_counts, n_genomes, filters, |kmer, row| {
-                            let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
-                            try_write(&mut buf, &row, &seq)
-                        })
-                        .map_err(OKIError::Partition)?;
+                    cache.iter_partition_kmers(i, use_counts, n_genomes, filters, |kmer, row| {
+                        let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
+                        try_write(&mut buf, &row, &seq)
+                    })?;
                 }
                 on_partition();
                 Ok(buf)
@@ -90,22 +106,19 @@ impl KmerIndex {
             // ── Unbounded: no atomic, no contention ───────────────────────────
             (0..n).into_par_iter().map(|i| {
                 let mut buf = Vec::<u8>::new();
+                let cache = IndexCache::new(self, Some(vec![i]));
                 if debug {
-                    self
-                        .iter_partition_kmers_located(i, use_counts, n_genomes, filters, |part, layer, kmer, row| {
-                            let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
-                            write_row(&mut buf, &row, &format!("{part},{layer},{seq}"));
-                            true
-                        })
-                        .map_err(OKIError::Partition)?;
+                    cache.iter_partition_kmers_located(i, use_counts, n_genomes, filters, |part, layer, kmer, row| {
+                        let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
+                        write_row(&mut buf, &row, &format!("{part},{layer},{seq}"));
+                        true
+                    })?;
                 } else {
-                    self
-                        .iter_partition_kmers(i, use_counts, n_genomes, filters, |kmer, row| {
-                            let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
-                            write_row(&mut buf, &row, &seq);
-                            true
-                        })
-                        .map_err(OKIError::Partition)?;
+                    cache.iter_partition_kmers(i, use_counts, n_genomes, filters, |kmer, row| {
+                        let seq = String::from_utf8(kmer.to_ascii()).unwrap_or_else(|_| "?".repeat(kmer_size));
+                        write_row(&mut buf, &row, &seq);
+                        true
+                    })?;
                 }
                 on_partition();
                 Ok(buf)
diff --git a/src/obikdump/src/lib.rs b/src/obikdump/src/lib.rs
index f953444f..45a7124d 100644
--- a/src/obikdump/src/lib.rs
+++ b/src/obikdump/src/lib.rs
@@ -6,3 +6,6 @@
 //! reverse), same pattern as `obikindexer`/`obikquery`.
 
 mod dump;
+mod partition_iter;
+
+pub use partition_iter::FilteredPartitionIter;
diff --git a/src/obikdump/src/partition_iter.rs b/src/obikdump/src/partition_iter.rs
new file mode 100644
index 00000000..b419c3e9
--- /dev/null
+++ b/src/obikdump/src/partition_iter.rs
@@ -0,0 +1,126 @@
+//! Filtered, batch-oriented iteration over an already-cached index's
+//! partitions/layers — the read side of `obikfilter`'s `KmerFilter`s.
+//! `IndexCache` is a foreign type (`obikidxcache`), so this is an extension
+//! trait rather than an inherent `impl`.
+//!
+//! Only meaningful on a *complete* source index: `IndexCache` panics if a
+//! layer is missing, which a finished index never has. Never use this on a
+//! destination index still being built (see `obikmerge::partition_merge`,
+//! which follows the same source-only-cache rule).
+
+use obikindex::OKIResult;
+use obikindex::layer::{KmerLayer, LayerContent};
+use obikidxcache::index_cache::IndexCache;
+use obikseq::CanonicalKmer;
+
+use obikfilter::{KmerFilter, passes_all};
+
+/// Kmers pulled per batch from a layer before filtering — keeps matrix reads
+/// grouped by (partition, layer) for locality instead of hopping row to row
+/// across the index. Same convention as `obikphylo::siblings::build`.
+const BATCH_SIZE: usize = 32768;
+
+pub trait FilteredPartitionIter {
+    /// Iterate all indexed kmers in partition `part`, calling `cb(kmer, row)` for each
+    /// kmer that passes every filter in `filters`.
+    ///
+    /// `use_counts = true`  → reads count columns (u32 values per genome), only
+    /// meaningful for `Count` layers. `use_counts = false` → reads presence
+    /// columns, converted to 0/1 u32 (works for both `Count` and `Presence`
+    /// layers — counts collapse to presence).
+    ///
+    /// Returns `Ok(true)` if all kmers were visited, `Ok(false)` if the callback halted.
+    fn iter_partition_kmers(
+        &self,
+        part: usize,
+        use_counts: bool,
+        n_genomes: usize,
+        filters: &[Box<dyn KmerFilter>],
+        cb: impl FnMut(CanonicalKmer, Box<[u32]>) -> bool,
+    ) -> OKIResult<bool>;
+
+    /// Like [`iter_partition_kmers`](Self::iter_partition_kmers) but the callback
+    /// also receives `(partition, layer)` indices, enabling debug output that
+    /// identifies where each kmer was stored.
+    fn iter_partition_kmers_located(
+        &self,
+        part: usize,
+        use_counts: bool,
+        n_genomes: usize,
+        filters: &[Box<dyn KmerFilter>],
+        cb: impl FnMut(usize, usize, CanonicalKmer, Box<[u32]>) -> bool,
+    ) -> OKIResult<bool>;
+}
+
+impl FilteredPartitionIter for IndexCache<'_> {
+    fn iter_partition_kmers(
+        &self,
+        part: usize,
+        use_counts: bool,
+        n_genomes: usize,
+        filters: &[Box<dyn KmerFilter>],
+        mut cb: impl FnMut(CanonicalKmer, Box<[u32]>) -> bool,
+    ) -> OKIResult<bool> {
+        for l in 0..self.n_layer(part).unwrap_or(0) {
+            let layer = self.get_layer(part, l).expect("layer within n_layer(part)");
+            if !iter_layer_kmers(layer, use_counts, n_genomes, filters, &mut |kmer, row| cb(kmer, row))? {
+                return Ok(false);
+            }
+        }
+        Ok(true)
+    }
+
+    fn iter_partition_kmers_located(
+        &self,
+        part: usize,
+        use_counts: bool,
+        n_genomes: usize,
+        filters: &[Box<dyn KmerFilter>],
+        mut cb: impl FnMut(usize, usize, CanonicalKmer, Box<[u32]>) -> bool,
+    ) -> OKIResult<bool> {
+        for l in 0..self.n_layer(part).unwrap_or(0) {
+            let layer = self.get_layer(part, l).expect("layer within n_layer(part)");
+            if !iter_layer_kmers(layer, use_counts, n_genomes, filters, &mut |kmer, row| cb(part, l, kmer, row))? {
+                return Ok(false);
+            }
+        }
+        Ok(true)
+    }
+}
+
+/// Batch-and-transpose one layer's kmers into per-kmer filtered rows.
+/// Returns `Ok(false)` if `cb` asked to stop early.
+fn iter_layer_kmers(
+    layer: &KmerLayer,
+    use_counts: bool,
+    n_genomes: usize,
+    filters: &[Box<dyn KmerFilter>],
+    cb: &mut dyn FnMut(CanonicalKmer, Box<[u32]>) -> bool,
+) -> OKIResult<bool> {
+    let read_counts = use_counts && matches!(layer.content(), LayerContent::Count);
+
+    for kmers in layer.iter_kmers_batch(BATCH_SIZE) {
+        // Kmers come straight from this layer's own iterator, so every one
+        // is a guaranteed member — a raw hash is enough, no membership
+        // recheck needed (see `KmerLayer::hash_batch`'s own doc).
+        let slots = layer.hash_batch(&kmers);
+
+        let cols: Vec<Vec<u32>> = if read_counts {
+            let mut cols: Vec<Vec<u32>> = vec![Vec::new(); n_genomes];
+            layer.fill_sub_matrix(&slots, &mut cols);
+            cols
+        } else {
+            let mut bool_cols: Vec<Vec<bool>> = vec![Vec::new(); n_genomes];
+            layer.fill_sub_matrix_carries(&slots, &mut bool_cols);
+            bool_cols.iter().map(|c| c.iter().map(|&b| b as u32).collect()).collect()
+        };
+
+        for (i, kmer) in kmers.into_iter().enumerate() {
+            let row: Box<[u32]> = cols.iter().map(|c| c[i]).collect();
+            if passes_all(filters, kmer, &row, n_genomes) && !cb(kmer, row) {
+                return Ok(false);
+            }
+        }
+    }
+    Ok(true)
+}
diff --git a/src/obikfilter/Cargo.toml b/src/obikfilter/Cargo.toml
index f1bde2f8..750e9f25 100644
--- a/src/obikfilter/Cargo.toml
+++ b/src/obikfilter/Cargo.toml
@@ -9,3 +9,4 @@ obicompactvec = { path = "../obicompactvec" }
 obikseq       = { path = "../obikseq" }
 obiskio       = { path = "../obiskio" }
 obitaxonomy   = { path = "../obitaxonomy" }
+obikentropy   = { path = "../obikentropy" }
diff --git a/src/obikfilter/src/dump_layer.rs b/src/obikfilter/src/dump_layer.rs
deleted file mode 100644
index 47507806..00000000
--- a/src/obikfilter/src/dump_layer.rs
+++ /dev/null
@@ -1,192 +0,0 @@
-use obikindex::layer::MphfLayer;
-use obicompactvec::{PersistentBitMatrix, PersistentCompactIntMatrix};
-use obikseq::CanonicalKmer;
-use obiskio::UnitigFileReader;
-use obikindex::{OKIError, OKIResult};
-
-use crate::filter::{KmerFilter, passes_all};
-use obikindex::KmerIndex;
-
-impl KmerIndex {
-    /// Iterate all indexed kmers in partition `part`, calling `cb(kmer, row)` for each
-    /// kmer that passes every filter in `filters`.
-    ///
-    /// `use_counts = true`  → reads count columns (u32 values per genome).
-    /// `use_counts = false` → reads presence columns, converted to 0/1 u32.
-    ///
-    /// If no data matrix exists for a layer (pure set-membership, single genome),
-    /// a row of `n_genomes` ones is emitted for every kmer in that layer — unless
-    /// the filter rejects it, in which case the whole layer is skipped.
-    /// Like [`iter_partition_kmers`] but the callback returns `false` to stop early.
-    /// Returns `Ok(true)` if all kmers were visited, `Ok(false)` if the callback halted.
-    pub fn iter_partition_kmers(
-        &self,
-        part: usize,
-        use_counts: bool,
-        n_genomes: usize,
-        filters: &[Box<dyn KmerFilter>],
-        mut cb: impl FnMut(CanonicalKmer, Box<[u32]>) -> bool,
-    ) -> OKIResult<bool> {
-        let index_dir = self.index_dir(part);
-        if !index_dir.exists() {
-            return Ok(true);
-        }
-
-        let mut l = 0;
-        loop {
-            let layer_dir = self.layer_dir(part, l)?;
-            if !layer_dir.exists() {
-                break;
-            }
-            l += 1;
-            let mphf = MphfLayer::open(&layer_dir)?;
-            let reader = UnitigFileReader::open_sequential(&layer_dir.join("unitigs.bin"))?;
-
-            let counts_dir = layer_dir.join("counts");
-            let presence_dir = layer_dir.join("presence");
-
-            let cont = if use_counts && counts_dir.exists() {
-                let mat = PersistentCompactIntMatrix::open(&layer_dir).map_err(OKIError::Io)?;
-                let mut cont = true;
-                for (kmer, _, _) in reader.iter_indexed_canonical_kmers() {
-                    if let Some(slot) = mphf.find(kmer) {
-                        let row = mat.row(slot);
-                        if passes_all(filters, kmer, &row, n_genomes) {
-                            cont = cb(kmer, row);
-                            if !cont {
-                                break;
-                            }
-                        }
-                    }
-                }
-                cont
-            } else if !use_counts && presence_dir.exists() {
-                let mat = PersistentBitMatrix::open(&layer_dir).map_err(OKIError::Io)?;
-                let mut cont = true;
-                for (kmer, _, _) in reader.iter_indexed_canonical_kmers() {
-                    if let Some(slot) = mphf.find(kmer) {
-                        let row: Box<[u32]> = mat.row(slot).iter().map(|&b| b as u32).collect();
-                        if passes_all(filters, kmer, &row, n_genomes) {
-                            cont = cb(kmer, row);
-                            if !cont {
-                                break;
-                            }
-                        }
-                    }
-                }
-                cont
-            } else {
-                // No data matrix: implicit presence — all values are 1. `row`
-                // is identical for every kmer, but a filter can still depend
-                // on the kmer's own sequence (e.g. MinComplexity), so this
-                // cannot be evaluated once for the whole layer — filters must
-                // still be tested per kmer.
-                let all_present: Box<[u32]> = vec![1u32; n_genomes].into();
-                let mut cont = true;
-                for (kmer, _, _) in reader.iter_indexed_canonical_kmers() {
-                    if mphf.find(kmer).is_some()
-                        && passes_all(filters, kmer, &all_present, n_genomes)
-                    {
-                        cont = cb(kmer, all_present.clone());
-                        if !cont {
-                            break;
-                        }
-                    }
-                }
-                cont
-            };
-
-            if !cont {
-                return Ok(false);
-            }
-        }
-
-        Ok(true)
-    }
-
-    /// Like [`iter_partition_kmers`] but the callback also receives `(partition, layer)`
-    /// indices, enabling debug output that identifies where each kmer was stored.
-    /// Returns `Ok(true)` if all kmers were visited, `Ok(false)` if the callback halted.
-    pub fn iter_partition_kmers_located(
-        &self,
-        part: usize,
-        use_counts: bool,
-        n_genomes: usize,
-        filters: &[Box<dyn KmerFilter>],
-        mut cb: impl FnMut(usize, usize, CanonicalKmer, Box<[u32]>) -> bool,
-    ) -> OKIResult<bool> {
-        let index_dir = self.index_dir(part);
-        if !index_dir.exists() {
-            return Ok(true);
-        }
-
-        let mut layer = 0;
-        loop {
-            let layer_dir = self.layer_dir(part, layer);
-            if !layer_dir.exists() {
-                break;
-            }
-            let mphf = MphfLayer::open(&layer_dir)?;
-            let reader = UnitigFileReader::open_sequential(&layer_dir.join("unitigs.bin"))?;
-
-            let counts_dir = layer_dir.join("counts");
-            let presence_dir = layer_dir.join("presence");
-
-            let cont = if use_counts && counts_dir.exists() {
-                let mat = PersistentCompactIntMatrix::open(&layer_dir).map_err(OKIError::Io)?;
-                let mut cont = true;
-                for (kmer, _, _) in reader.iter_indexed_canonical_kmers() {
-                    if let Some(slot) = mphf.find(kmer) {
-                        let row = mat.row(slot);
-                        if passes_all(filters, kmer, &row, n_genomes) {
-                            cont = cb(part, layer, kmer, row);
-                            if !cont {
-                                break;
-                            }
-                        }
-                    }
-                }
-                cont
-            } else if !use_counts && presence_dir.exists() {
-                let mat = PersistentBitMatrix::open(&layer_dir).map_err(OKIError::Io)?;
-                let mut cont = true;
-                for (kmer, _, _) in reader.iter_indexed_canonical_kmers() {
-                    if let Some(slot) = mphf.find(kmer) {
-                        let row: Box<[u32]> = mat.row(slot).iter().map(|&b| b as u32).collect();
-                        if passes_all(filters, kmer, &row, n_genomes) {
-                            cont = cb(part, layer, kmer, row);
-                            if !cont {
-                                break;
-                            }
-                        }
-                    }
-                }
-                cont
-            } else {
-                // Same as iter_partition_kmers: row is constant but a filter
-                // may still depend on the kmer's own sequence, so this must
-                // be tested per kmer, not once for the whole layer.
-                let all_present: Box<[u32]> = vec![1u32; n_genomes].into();
-                let mut cont = true;
-                for (kmer, _, _) in reader.iter_indexed_canonical_kmers() {
-                    if mphf.find(kmer).is_some()
-                        && passes_all(filters, kmer, &all_present, n_genomes)
-                    {
-                        cont = cb(part, layer, kmer, all_present.clone());
-                        if !cont {
-                            break;
-                        }
-                    }
-                }
-                cont
-            };
-
-            if !cont {
-                return Ok(false);
-            }
-            layer += 1;
-        }
-
-        Ok(true)
-    }
-}
diff --git a/src/obikfilter/src/filter.rs b/src/obikfilter/src/filter.rs
index 9f54a695..ffb8b745 100644
--- a/src/obikfilter/src/filter.rs
+++ b/src/obikfilter/src/filter.rs
@@ -1,6 +1,8 @@
 use obicompactvec::FilterMask;
 use obikseq::CanonicalKmer;
 
+use crate::predicate::Selection;
+
 /// Trait for kmer filters.
 ///
 /// `kmer` is the k-mer's own canonical sequence, reconstructed from the
@@ -173,14 +175,12 @@ impl KmerFilter for MaxTotalCount {
 
 // ── Group-based quorum filter ─────────────────────────────────────────────────
 
-/// Quorum filter operating on pre-classified genome groups.
+/// Quorum filter operating on a pre-classified genome [`Selection`].
 ///
-/// `ingroup_idx` / `outgroup_idx` are column indices into the per-genome row.
-/// When `ingroup_idx` is empty, no ingroup quorum is checked.
-/// When `outgroup_idx` is empty, no outgroup quorum is checked.
+/// `selection.ingroup_idx` / `selection.outgroup_idx` are column indices into
+/// the per-genome row. When empty, the corresponding quorum is not checked.
 pub struct GroupQuorumFilter {
-    pub ingroup_idx:        Vec<usize>,
-    pub outgroup_idx:       Vec<usize>,
+    pub selection:           Selection,
     pub threshold:          u32,
     pub min_count:          usize,
     pub max_count:          usize,
@@ -225,22 +225,22 @@ impl GroupQuorumFilter {
 
 impl KmerFilter for GroupQuorumFilter {
     fn passes(&self, _kmer: CanonicalKmer, row: &[u32], _n_genomes: usize) -> bool {
-        if !self.ingroup_idx.is_empty() {
-            let n = self.ingroup_idx.iter()
+        if !self.selection.ingroup_idx.is_empty() {
+            let n = self.selection.ingroup_idx.iter()
                 .filter(|&&i| row.get(i).copied().unwrap_or(0) > self.threshold)
                 .count();
-            let denom = self.ingroup_idx.len();
+            let denom = self.selection.ingroup_idx.len();
             if n < self.min_count { return false; }
             if n > self.max_count { return false; }
             let frac = n as f64 / denom as f64;
             if frac < self.min_frac { return false; }
             if frac > self.max_frac { return false; }
         }
-        if !self.outgroup_idx.is_empty() {
-            let n = self.outgroup_idx.iter()
+        if !self.selection.outgroup_idx.is_empty() {
+            let n = self.selection.outgroup_idx.iter()
                 .filter(|&&i| row.get(i).copied().unwrap_or(0) > self.threshold)
                 .count();
-            let denom = self.outgroup_idx.len();
+            let denom = self.selection.outgroup_idx.len();
             if n < self.min_outgroup_count { return false; }
             if n > self.max_outgroup_count { return false; }
             let frac = n as f64 / denom as f64;
@@ -253,17 +253,17 @@ impl KmerFilter for GroupQuorumFilter {
     fn column_mask_expr(&self, _n_genomes: usize) -> Option<FilterMask> {
         let t = self.threshold.checked_add(1)?;
         let mut parts: Vec<FilterMask> = Vec::new();
-        if !self.ingroup_idx.is_empty() {
+        if !self.selection.ingroup_idx.is_empty() {
             Self::group_mask_parts(
-                &self.ingroup_idx, t,
+                &self.selection.ingroup_idx, t,
                 self.min_count, self.max_count,
                 self.min_frac, self.max_frac,
                 &mut parts,
             );
         }
-        if !self.outgroup_idx.is_empty() {
+        if !self.selection.outgroup_idx.is_empty() {
             Self::group_mask_parts(
-                &self.outgroup_idx, t,
+                &self.selection.outgroup_idx, t,
                 self.min_outgroup_count, self.max_outgroup_count,
                 self.min_outgroup_frac, self.max_outgroup_frac,
                 &mut parts,
diff --git a/src/obikfilter/src/lib.rs b/src/obikfilter/src/lib.rs
index 26a95c23..a6b5736f 100644
--- a/src/obikfilter/src/lib.rs
+++ b/src/obikfilter/src/lib.rs
@@ -4,19 +4,18 @@
 //! operates on already-retained k-mers.
 //!
 //! [`filter`] (the `KmerFilter` trait + its implementations) depends only
-//! on `obicompactvec`/`obikseq`, not on `obikindex`. [`dump_layer`] is the
-//! extension over `obikindex::KmerIndex` that actually iterates a
-//! partition's k-mers through those filters (`iter_partition_kmers`,
-//! `iter_partition_kmers_located`) — every k-mer, filtered or not, goes
-//! through this same path (`passes_all` on an empty filter list is always
-//! `true`), so this crate depends one-way on `obikindex`, not the reverse.
+//! on `obicompactvec`/`obikseq`, not on `obikindex`. The partition/layer
+//! iteration that actually runs these filters over an index
+//! (`iter_partition_kmers`, `iter_partition_kmers_located`) lives in
+//! `obikdump` instead (`FilteredPartitionIter`) — it needs `obikidxcache`'s
+//! `IndexCache` to read a *complete* source index, which this crate has no
+//! reason to depend on.
 
 mod filter;
-mod dump_layer;
 mod predicate;
 
 pub use filter::{
     GroupQuorumFilter, KmerFilter, MaxGenomeCount, MaxGenomeFraction, MaxTotalCount,
     MinComplexity, MinGenomeCount, MinGenomeFraction, MinTotalCount, passes_all,
 };
-pub use predicate::{GroupFilterParams, MetaPred};
+pub use predicate::{GenomeSelector, GroupFilterParams, MetaPred, Selection};
diff --git a/src/obikfilter/src/predicate.rs b/src/obikfilter/src/predicate.rs
index db1be2d8..065c7cff 100644
--- a/src/obikfilter/src/predicate.rs
+++ b/src/obikfilter/src/predicate.rs
@@ -68,14 +68,6 @@ impl MetaPred {
     }
 }
 
-impl GenomeInfo {
-    /// Evaluate a single metadata predicate against this genome.
-    /// Returns `None` when the predicate's key is absent (NA propagation).
-    pub fn matches(&self, pred: &MetaPred) -> Option<bool> {
-        pred.eval(&self.meta)
-    }
-}
-
 // ── Path matching ─────────────────────────────────────────────────────────────
 
 /// True if the stored taxonomy `value` matches `pattern`.
@@ -153,19 +145,49 @@ pub struct GroupFilterParams {
     pub max_outgroup_frac:  Option<f64>,
 }
 
-impl IndexMeta {
-    /// Returns indices of genomes matching `pred_str` (single predicate).
-    pub fn matching_genome_indices(&self, pred_str: &str) -> Result<Vec<usize>, String> {
-        let pred = MetaPred::parse(pred_str)?;
-        let genomes = self.genomes().map_err(|e| e.to_string())?;
-        Ok(genomes.iter().enumerate()
-            .filter_map(|(i, g)| {
-                if g.matches(&pred) == Some(true) { Some(i) } else { std::option::Option::None }
-            })
-            .collect())
+// ── Genome selector ──────────────────────────────────────────────────────────
+
+/// Result of running a [`GenomeSelector`] against an index's metadata.
+pub struct Selection {
+    pub ingroup_idx:  Vec<usize>,
+    pub outgroup_idx: Vec<usize>,
+}
+
+pub struct GenomeSelector {
+    pub(crate) ingroup:  Vec<MetaPred>,
+    pub(crate) outgroup: Vec<MetaPred>,
+}
+
+impl GenomeSelector {
+    /// Parse ingroup (AND'd) and outgroup (OR'd) predicate strings.
+    pub fn parse(ingroup: &[String], outgroup: &[String]) -> Result<Self, String> {
+        let ingroup  = ingroup.iter().map(|s| MetaPred::parse(s)).collect::<Result<Vec<_>, _>>()?;
+        let outgroup = outgroup.iter().map(|s| MetaPred::parse(s)).collect::<Result<Vec<_>, _>>()?;
+        Ok(Self { ingroup, outgroup })
     }
 
-    /// Build a `GroupQuorumFilter` from parsed predicates, evaluated against `self.genomes`.
+    /// Classify `meta`'s genomes into ingroup/outgroup indices.
+    ///
+    /// - No predicates at all: every genome is (implicitly) ingroup.
+    /// - Otherwise: ingroup wins on overlap; uncategorized genomes are dropped.
+    pub fn run(&self, meta: &IndexMeta) -> Result<Selection, String> {
+        let genomes = meta.genomes().map_err(|e| e.to_string())?;
+
+        if self.ingroup.is_empty() && self.outgroup.is_empty() {
+            return Ok(Selection { ingroup_idx: (0..genomes.len()).collect(), outgroup_idx: vec![] });
+        }
+
+        let members = classify(&genomes, &self.ingroup, &self.outgroup);
+        let ingroup_idx: Vec<usize> = members.iter().enumerate()
+            .filter(|(_, m)| matches!(m, Membership::Ingroup))
+            .map(|(i, _)| i).collect();
+        let outgroup_idx: Vec<usize> = members.iter().enumerate()
+            .filter(|(_, m)| matches!(m, Membership::Outgroup))
+            .map(|(i, _)| i).collect();
+        Ok(Selection { ingroup_idx, outgroup_idx })
+    }
+
+    /// Build a `GroupQuorumFilter` from this selector's classification of `meta`.
     ///
     /// - No groups defined: `ingroup_idx` = all genomes (implicit ingroup).
     /// - `ingroup` predicates only: outgroup indices are empty.
@@ -173,34 +195,20 @@ impl IndexMeta {
     /// - Both defined: ingroup wins on overlap; uncategorized genomes are ignored.
     pub fn build_group_filter(
         &self,
-        ingroup_preds:  &[MetaPred],
-        outgroup_preds: &[MetaPred],
-        p:              GroupFilterParams,
+        meta: &IndexMeta,
+        p:    GroupFilterParams,
     ) -> Result<GroupQuorumFilter, String> {
-        let genomes = self.genomes().map_err(|e| e.to_string())?;
-        let (ingroup_idx, outgroup_idx) = if ingroup_preds.is_empty() && outgroup_preds.is_empty() {
-            ((0..genomes.len()).collect(), vec![])
-        } else {
-            let members = classify(&genomes, ingroup_preds, outgroup_preds);
-            let in_idx: Vec<usize>  = members.iter().enumerate()
-                .filter(|(_, m)| matches!(m, Membership::Ingroup))
-                .map(|(i, _)| i).collect();
-            let out_idx: Vec<usize> = members.iter().enumerate()
-                .filter(|(_, m)| matches!(m, Membership::Outgroup))
-                .map(|(i, _)| i).collect();
-            (in_idx, out_idx)
-        };
-
-        let in_size  = ingroup_idx.len();
-        let out_size = outgroup_idx.len();
+        let selection = self.run(meta)?;
+        let in_size  = selection.ingroup_idx.len();
+        let out_size = selection.outgroup_idx.len();
 
         let ingroup_quorum_explicit  = p.min_count.is_some() || p.max_count.is_some()
             || p.min_frac.is_some() || p.max_frac.is_some();
         let outgroup_quorum_explicit = p.min_outgroup_count.is_some() || p.max_outgroup_count.is_some()
             || p.min_outgroup_frac.is_some() || p.max_outgroup_frac.is_some();
 
-        let default_min_frac           = if !ingroup_preds.is_empty()  && !ingroup_quorum_explicit  { 1.0      } else { 0.0      };
-        let default_max_outgroup_count = if !outgroup_preds.is_empty() && !outgroup_quorum_explicit { 0        } else { out_size };
+        let default_min_frac           = if !self.ingroup.is_empty()  && !ingroup_quorum_explicit  { 1.0      } else { 0.0      };
+        let default_max_outgroup_count = if !self.outgroup.is_empty() && !outgroup_quorum_explicit { 0        } else { out_size };
 
         // Resolve a signed count: negative means an offset from the group size
         // (e.g. -1 = all but one), floored at 1 so the negative form always keeps
@@ -238,8 +246,7 @@ impl IndexMeta {
         }
 
         Ok(GroupQuorumFilter {
-            ingroup_idx,
-            outgroup_idx,
+            selection,
             threshold: p.threshold,
             min_count,
             max_count,
@@ -252,3 +259,4 @@ impl IndexMeta {
         })
     }
 }
+