feat: port khive-hnsw — HNSW with INT8 quantized two-phase search

crates/Cargo.toml

@@ -11,6 +11,7 @@ members = [
     "khive-gate-rego",
     "khive-fusion",
     "khive-bm25",
+    "khive-hnsw",
     "khive-runtime",
     "khive-request",
     "khive-pack-kg",

crates/khive-bm25/src/index/mod.rs

@@ -149,7 +149,9 @@ fn default_postings_epoch() -> u64 {
 // TODO(port): was generated by `khive_types::transparent_string_newtype!` macro
 // which does not yet exist in khive-types. Expanded here manually until the macro
 // lands in khive-types and is re-adopted.
-#[derive(Debug, Clone, PartialEq, Eq, Hash, PartialOrd, Ord, serde::Serialize, serde::Deserialize)]
+#[derive(
+    Debug, Clone, PartialEq, Eq, Hash, PartialOrd, Ord, serde::Serialize, serde::Deserialize,
+)]
 #[serde(transparent)]
 pub struct DocumentId(String);
 
@@ -639,7 +641,9 @@ impl Bm25Index {
             return id;
         }
         let id = self.next_internal_id;
-        self.next_internal_id = self.next_internal_id.checked_add(1)
+        self.next_internal_id = self
+            .next_internal_id
+            .checked_add(1)
             .expect("internal document ID space exhausted (u32::MAX)");
         self.id_to_internal.insert(doc_id.clone(), id);
         if id as usize >= self.internal_to_id.len() {

crates/khive-bm25/src/index/search.rs

@@ -1450,11 +1450,11 @@ mod tests_simd_scoring {
         let tfs_4: [u8; 4] = [0, 0, 0, 0];
         let dls_4: [f32; 4] = [10.0, 20.0, 5.0, 1.0];
         let result = score_batch_4(&tfs_4, &dls_4, TEST_IDF, TEST_K1P1, TEST_BASE, TEST_DL_FAC);
-        for i in 0..4 {
+        for val in &result {
             assert!(
-                result[i].abs() < 1e-10,
+                val.abs() < 1e-10,
                 "tf=0 should produce ~0 score, got {}",
-                result[i]
+                val
             );
         }
 
@@ -1467,11 +1467,11 @@ mod tests_simd_scoring {
             let result = unsafe {
                 score_batch_avx2(&tfs_8, &dls_8, TEST_IDF, TEST_K1P1, TEST_BASE, TEST_DL_FAC)
             };
-            for i in 0..8 {
+            for val in &result {
                 assert!(
-                    result[i].abs() < 1e-10,
+                    val.abs() < 1e-10,
                     "avx2 tf=0 should produce ~0 score, got {}",
-                    result[i]
+                    val
                 );
             }
         }

crates/khive-bm25/src/metrics.rs

@@ -63,13 +63,20 @@ pub struct RecordingSink {
 }
 
 #[cfg(test)]
-impl RecordingSink {
-    /// Create an empty recording sink.
-    pub fn new() -> Self {
+impl Default for RecordingSink {
+    fn default() -> Self {
         Self {
             events: Mutex::new(Vec::new()),
         }
     }
+}
+
+#[cfg(test)]
+impl RecordingSink {
+    /// Create an empty recording sink.
+    pub fn new() -> Self {
+        Self::default()
+    }
 
     /// Return a snapshot of all recorded events.
     pub fn events(&self) -> Vec<MetricEvent> {

crates/khive-bm25/src/tests.rs

@@ -813,8 +813,8 @@ mod golden_tests {
 /// Memory budget enforcement tests for BM25.
 #[cfg(test)]
 mod memory_budget_tests {
-    use crate::{Bm25Config, Bm25Index};
     use crate::error::{ErrorKind, RetrievalError};
+    use crate::{Bm25Config, Bm25Index};
 
     #[test]
     fn test_no_budget_allows_unlimited_indexing() {
@@ -844,9 +844,8 @@ mod memory_budget_tests {
                 format!("doc{i}"),
                 &format!("some content words for document number {i} with extra text"),
             );
-            if result.is_err() {
+            if let Err(err) = result {
                 rejected = true;
-                let err = result.unwrap_err();
                 assert!(
                     matches!(err, RetrievalError::BudgetExceeded { .. }),
                     "Expected BudgetExceeded, got: {err:?}"
@@ -1019,8 +1018,8 @@ mod memory_budget_tests {
 
 #[cfg(test)]
 mod metrics_tests {
-    use crate::{Bm25Config, Bm25Index};
     use crate::metrics::{names, MetricValue, RecordingSink};
+    use crate::{Bm25Config, Bm25Index};
     use std::sync::Arc;
 
     #[test]

crates/khive-bm25/src/tokenizer.rs

@@ -247,8 +247,10 @@ mod tests {
 
     #[test]
     fn test_stop_words_disabled() {
-        let mut tokenizer = SimpleTokenizer::default();
-        tokenizer.filter_stop_words = false;
+        let tokenizer = SimpleTokenizer {
+            filter_stop_words: false,
+            ..Default::default()
+        };
         let tokens = tokenizer.tokenize("The Quick, Brown FOX!");
         assert_eq!(tokens, vec!["the", "quick", "brown", "fox"]);
     }

crates/khive-fusion/src/tests.rs

@@ -221,19 +221,11 @@ mod property_tests {
     /// Verifies the `present_gt_absent` property from RRF.lean.
     #[test]
     fn prop_rrf_more_sources_higher_score() {
-        let source1: Vec<(String, DeterministicScore)> = vec![(
-            "doc_common".to_string(),
-            DeterministicScore::from_f64(0.9),
-        )];
+        let source1: Vec<(String, DeterministicScore)> =
+            vec![("doc_common".to_string(), DeterministicScore::from_f64(0.9))];
         let source2: Vec<(String, DeterministicScore)> = vec![
-            (
-                "doc_common".to_string(),
-                DeterministicScore::from_f64(0.9),
-            ),
-            (
-                "doc_single".to_string(),
-                DeterministicScore::from_f64(0.8),
-            ),
+            ("doc_common".to_string(), DeterministicScore::from_f64(0.9)),
+            ("doc_single".to_string(), DeterministicScore::from_f64(0.8)),
         ];
 
         let fused = reciprocal_rank_fusion(vec![source1, source2], 60);

crates/khive-hnsw/Cargo.toml

@@ -0,0 +1,30 @@
+[package]
+name = "khive-hnsw"
+version.workspace = true
+edition.workspace = true
+authors.workspace = true
+license.workspace = true
+repository.workspace = true
+homepage.workspace = true
+keywords.workspace = true
+categories.workspace = true
+description = "HNSW (Hierarchical Navigable Small World) vector index with INT8 quantized two-phase search — formally verified in Lean4"
+
+[dependencies]
+khive-score = { version = "0.2.0", path = "../khive-score" }
+khive-types = { version = "0.2.0", path = "../khive-types" }
+lattice-embed = { workspace = true }
+serde = { workspace = true }
+serde_json = { workspace = true }
+thiserror = { workspace = true }
+parking_lot = { workspace = true }
+tokio = { workspace = true }
+rand = "0.8"
+rayon = "1.10"
+ulid = "1.1"
+
+[dev-dependencies]
+proptest = "1"
+
+[features]
+checkpoint = []