llm_agent_runtime/

graph.rs

//! # Module: Graph
//!
//! ## Responsibility
//! Provides an in-memory knowledge graph with typed entities and relationships.
//! Mirrors the public API of `mem-graph`.
//!
//! ## Guarantees
//! - Thread-safe: `GraphStore` wraps state in `Arc<Mutex<_>>`
//! - BFS/DFS traversal and shortest-path are correct for directed graphs
//! - Non-panicking: all operations return `Result`
//!
//! ## NOT Responsible For
//! - Persistence to disk or external store
//! - Graph sharding / distributed graphs

use crate::error::AgentRuntimeError;
use serde::{Deserialize, Serialize};
use serde_json::Value;
use std::collections::{BinaryHeap, HashMap, HashSet, VecDeque};
use std::sync::{Arc, Mutex};

// ── Lock recovery helper ───────────────────────────────────────────────────────

/// Recover from a poisoned mutex by logging a warning and returning the inner
/// value. This is safe because we never leave shared data in a partially-written
/// state across an await or panic boundary in this module.
fn recover_lock<'a, T>(
    result: std::sync::LockResult<std::sync::MutexGuard<'a, T>>,
    ctx: &str,
) -> std::sync::MutexGuard<'a, T>
where
    T: ?Sized,
{
    match result {
        Ok(guard) => guard,
        Err(poisoned) => {
            tracing::warn!("mutex poisoned in {ctx}, recovering inner value");
            poisoned.into_inner()
        }
    }
}

// ── OrdF32 newtype ─────────────────────────────────────────────────────────────

/// Newtype wrapper for `f32` that implements `Ord`.
/// NaN is treated as `Greater` for safe use in a `BinaryHeap`.
#[derive(Debug, Clone, Copy, PartialEq)]
struct OrdF32(f32);

impl Eq for OrdF32 {}

impl PartialOrd for OrdF32 {
    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
        Some(self.cmp(other))
    }
}

impl Ord for OrdF32 {
    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
        self.0
            .partial_cmp(&other.0)
            .unwrap_or(std::cmp::Ordering::Greater)
    }
}

// ── EntityId ──────────────────────────────────────────────────────────────────

/// Stable identifier for a graph entity.
#[derive(Debug, Clone, PartialEq, Eq, Hash, PartialOrd, Ord, Serialize, Deserialize)]
pub struct EntityId(pub String);

impl EntityId {
    /// Create a new `EntityId` from any string-like value.
    pub fn new(id: impl Into<String>) -> Self {
        Self(id.into())
    }
}

impl std::fmt::Display for EntityId {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        write!(f, "{}", self.0)
    }
}

// ── Entity ────────────────────────────────────────────────────────────────────

/// A node in the knowledge graph.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct Entity {
    /// Unique identifier.
    pub id: EntityId,
    /// Human-readable label (e.g. "Person", "Concept").
    pub label: String,
    /// Arbitrary key-value properties.
    pub properties: HashMap<String, Value>,
}

impl Entity {
    /// Construct a new entity with no properties.
    pub fn new(id: impl Into<String>, label: impl Into<String>) -> Self {
        Self {
            id: EntityId::new(id),
            label: label.into(),
            properties: HashMap::new(),
        }
    }

    /// Construct a new entity with the given properties.
    pub fn with_properties(
        id: impl Into<String>,
        label: impl Into<String>,
        properties: HashMap<String, Value>,
    ) -> Self {
        Self {
            id: EntityId::new(id),
            label: label.into(),
            properties,
        }
    }
}

// ── Relationship ──────────────────────────────────────────────────────────────

/// A directed, typed edge between two entities.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct Relationship {
    /// Source entity.
    pub from: EntityId,
    /// Target entity.
    pub to: EntityId,
    /// Relationship type label (e.g. "KNOWS", "PART_OF").
    pub kind: String,
    /// Optional weight for weighted-graph use cases.
    pub weight: f32,
}

impl Relationship {
    /// Construct a new relationship with the given kind and weight.
    pub fn new(
        from: impl Into<String>,
        to: impl Into<String>,
        kind: impl Into<String>,
        weight: f32,
    ) -> Self {
        Self {
            from: EntityId::new(from),
            to: EntityId::new(to),
            kind: kind.into(),
            weight,
        }
    }
}

// ── MemGraphError (mirrors upstream) ─────────────────────────────────────────

/// Graph-specific errors, mirrors `mem-graph::MemGraphError`.
#[derive(Debug, thiserror::Error)]
pub enum MemGraphError {
    /// The requested entity was not found.
    #[error("Entity '{0}' not found")]
    EntityNotFound(String),

    /// A relationship between the two entities already exists with the same kind.
    #[error("Relationship '{kind}' from '{from}' to '{to}' already exists")]
    DuplicateRelationship {
        /// Source entity ID.
        from: String,
        /// Target entity ID.
        to: String,
        /// Relationship kind label.
        kind: String,
    },

    /// Generic internal error.
    #[error("Graph internal error: {0}")]
    Internal(String),
}

impl From<MemGraphError> for AgentRuntimeError {
    fn from(e: MemGraphError) -> Self {
        AgentRuntimeError::Graph(e.to_string())
    }
}

// ── GraphStore ────────────────────────────────────────────────────────────────

/// In-memory knowledge graph supporting entities, relationships, BFS/DFS,
/// shortest-path, weighted shortest-path, and graph analytics.
///
/// ## Guarantees
/// - Thread-safe via `Arc<Mutex<_>>`
/// - BFS/DFS are non-recursive (stack-safe)
/// - Shortest-path is hop-count based (BFS)
/// - Weighted shortest-path uses Dijkstra's algorithm
#[derive(Debug, Clone)]
pub struct GraphStore {
    inner: Arc<Mutex<GraphInner>>,
}

#[derive(Debug)]
struct GraphInner {
    entities: HashMap<EntityId, Entity>,
    relationships: Vec<Relationship>,
}

impl GraphStore {
    /// Create a new, empty graph store.
    pub fn new() -> Self {
        Self {
            inner: Arc::new(Mutex::new(GraphInner {
                entities: HashMap::new(),
                relationships: Vec::new(),
            })),
        }
    }

    /// Add an entity to the graph.
    ///
    /// If an entity with the same ID already exists, it is replaced.
    pub fn add_entity(&self, entity: Entity) -> Result<(), AgentRuntimeError> {
        let mut inner = recover_lock(self.inner.lock(), "add_entity");
        inner.entities.insert(entity.id.clone(), entity);
        Ok(())
    }

    /// Retrieve an entity by ID.
    pub fn get_entity(&self, id: &EntityId) -> Result<Entity, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "get_entity");
        inner
            .entities
            .get(id)
            .cloned()
            .ok_or_else(|| AgentRuntimeError::Graph(format!("entity '{}' not found", id.0)))
    }

    /// Add a directed relationship between two existing entities.
    ///
    /// Both source and target entities must already exist in the graph.
    pub fn add_relationship(&self, rel: Relationship) -> Result<(), AgentRuntimeError> {
        let mut inner = recover_lock(self.inner.lock(), "add_relationship");

        if !inner.entities.contains_key(&rel.from) {
            return Err(AgentRuntimeError::Graph(format!(
                "source entity '{}' not found",
                rel.from.0
            )));
        }
        if !inner.entities.contains_key(&rel.to) {
            return Err(AgentRuntimeError::Graph(format!(
                "target entity '{}' not found",
                rel.to.0
            )));
        }

        // Reject duplicate (from, to, kind) triples — the DuplicateRelationship
        // error variant existed but was never raised, silently allowing duplicate
        // edges that corrupt relationship_count() and BFS/DFS result counts.
        let duplicate = inner
            .relationships
            .iter()
            .any(|r| r.from == rel.from && r.to == rel.to && r.kind == rel.kind);
        if duplicate {
            return Err(AgentRuntimeError::Graph(
                MemGraphError::DuplicateRelationship {
                    from: rel.from.0.clone(),
                    to: rel.to.0.clone(),
                    kind: rel.kind.clone(),
                }
                .to_string(),
            ));
        }

        inner.relationships.push(rel);
        Ok(())
    }

    /// Remove an entity and all relationships involving it.
    pub fn remove_entity(&self, id: &EntityId) -> Result<(), AgentRuntimeError> {
        let mut inner = recover_lock(self.inner.lock(), "remove_entity");

        if inner.entities.remove(id).is_none() {
            return Err(AgentRuntimeError::Graph(format!(
                "entity '{}' not found",
                id.0
            )));
        }
        inner.relationships.retain(|r| &r.from != id && &r.to != id);
        Ok(())
    }

    /// Return all direct neighbours of the given entity (BFS layer 1).
    fn neighbours(relationships: &[Relationship], id: &EntityId) -> Vec<EntityId> {
        relationships
            .iter()
            .filter(|r| &r.from == id)
            .map(|r| r.to.clone())
            .collect()
    }

    /// Breadth-first search starting from `start`.
    ///
    /// Returns entity IDs in BFS discovery order (not including the start node).
    #[tracing::instrument(skip(self))]
    pub fn bfs(&self, start: &EntityId) -> Result<Vec<EntityId>, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "bfs");

        if !inner.entities.contains_key(start) {
            return Err(AgentRuntimeError::Graph(format!(
                "start entity '{}' not found",
                start.0
            )));
        }

        let mut visited: HashSet<EntityId> = HashSet::new();
        let mut queue: VecDeque<EntityId> = VecDeque::new();
        let mut result: Vec<EntityId> = Vec::new();

        visited.insert(start.clone());
        queue.push_back(start.clone());

        while let Some(current) = queue.pop_front() {
            let neighbours: Vec<EntityId> = Self::neighbours(&inner.relationships, &current);
            for neighbour in neighbours {
                if visited.insert(neighbour.clone()) {
                    result.push(neighbour.clone());
                    queue.push_back(neighbour);
                }
            }
        }

        tracing::debug!("BFS visited {} nodes", result.len());
        Ok(result)
    }

    /// Depth-first search starting from `start`.
    ///
    /// Returns entity IDs in DFS discovery order (not including the start node).
    #[tracing::instrument(skip(self))]
    pub fn dfs(&self, start: &EntityId) -> Result<Vec<EntityId>, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "dfs");

        if !inner.entities.contains_key(start) {
            return Err(AgentRuntimeError::Graph(format!(
                "start entity '{}' not found",
                start.0
            )));
        }

        let mut visited: HashSet<EntityId> = HashSet::new();
        let mut stack: Vec<EntityId> = Vec::new();
        let mut result: Vec<EntityId> = Vec::new();

        visited.insert(start.clone());
        stack.push(start.clone());

        while let Some(current) = stack.pop() {
            let neighbours: Vec<EntityId> = Self::neighbours(&inner.relationships, &current);
            for neighbour in neighbours {
                if visited.insert(neighbour.clone()) {
                    result.push(neighbour.clone());
                    stack.push(neighbour);
                }
            }
        }

        tracing::debug!("DFS visited {} nodes", result.len());
        Ok(result)
    }

    /// Find the shortest path (by hop count) between `from` and `to`.
    ///
    /// # Returns
    /// - `Some(path)` — ordered list of `EntityId`s from `from` to `to` (inclusive)
    /// - `None` — no path exists
    #[tracing::instrument(skip(self))]
    pub fn shortest_path(
        &self,
        from: &EntityId,
        to: &EntityId,
    ) -> Result<Option<Vec<EntityId>>, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "shortest_path");

        if !inner.entities.contains_key(from) {
            return Err(AgentRuntimeError::Graph(format!(
                "source entity '{}' not found",
                from.0
            )));
        }
        if !inner.entities.contains_key(to) {
            return Err(AgentRuntimeError::Graph(format!(
                "target entity '{}' not found",
                to.0
            )));
        }

        if from == to {
            return Ok(Some(vec![from.clone()]));
        }

        let mut visited: HashSet<EntityId> = HashSet::new();
        let mut queue: VecDeque<Vec<EntityId>> = VecDeque::new();

        visited.insert(from.clone());
        queue.push_back(vec![from.clone()]);

        while let Some(path) = queue.pop_front() {
            let current = match path.last() {
                Some(c) => c.clone(),
                None => continue,
            };

            let neighbours: Vec<EntityId> = Self::neighbours(&inner.relationships, &current);

            for neighbour in neighbours {
                if &neighbour == to {
                    let mut full_path = path.clone();
                    full_path.push(neighbour);
                    return Ok(Some(full_path));
                }
                if visited.insert(neighbour.clone()) {
                    let mut new_path = path.clone();
                    new_path.push(neighbour.clone());
                    queue.push_back(new_path);
                }
            }
        }

        Ok(None)
    }

    /// Find the shortest weighted path between `from` and `to` using Dijkstra's algorithm.
    ///
    /// Uses `Relationship::weight` as edge cost. Negative weights are not supported
    /// and will cause this method to return an error.
    ///
    /// # Returns
    /// - `Ok(Some((path, total_weight)))` — the shortest path and its total weight
    /// - `Ok(None)` — no path exists between `from` and `to`
    /// - `Err(...)` — either entity not found, or a negative edge weight was encountered
    pub fn shortest_path_weighted(
        &self,
        from: &EntityId,
        to: &EntityId,
    ) -> Result<Option<(Vec<EntityId>, f32)>, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "shortest_path_weighted");

        if !inner.entities.contains_key(from) {
            return Err(AgentRuntimeError::Graph(format!(
                "source entity '{}' not found",
                from.0
            )));
        }
        if !inner.entities.contains_key(to) {
            return Err(AgentRuntimeError::Graph(format!(
                "target entity '{}' not found",
                to.0
            )));
        }

        // Validate: no negative weights
        for rel in &inner.relationships {
            if rel.weight < 0.0 {
                return Err(AgentRuntimeError::Graph(format!(
                    "negative weight {:.4} on edge '{}' -> '{}'",
                    rel.weight, rel.from.0, rel.to.0
                )));
            }
        }

        if from == to {
            return Ok(Some((vec![from.clone()], 0.0)));
        }

        // Dijkstra using a max-heap with negated weights to simulate a min-heap.
        // Heap entries: (negated_cost, node_id)
        let mut dist: HashMap<EntityId, f32> = HashMap::new();
        let mut prev: HashMap<EntityId, EntityId> = HashMap::new();
        // BinaryHeap is a max-heap; negate weights to get min-heap behaviour.
        let mut heap: BinaryHeap<(OrdF32, EntityId)> = BinaryHeap::new();

        dist.insert(from.clone(), 0.0);
        heap.push((OrdF32(-0.0), from.clone()));

        while let Some((OrdF32(neg_cost), current)) = heap.pop() {
            let cost = -neg_cost;

            // Skip stale entries.
            if let Some(&best) = dist.get(&current) {
                if cost > best {
                    continue;
                }
            }

            if &current == to {
                // Reconstruct path in reverse.
                let mut path = vec![to.clone()];
                let mut node = to.clone();
                while let Some(p) = prev.get(&node) {
                    path.push(p.clone());
                    node = p.clone();
                }
                path.reverse();
                return Ok(Some((path, cost)));
            }

            for rel in inner.relationships.iter().filter(|r| &r.from == &current) {
                let next_cost = cost + rel.weight;
                let entry = dist.entry(rel.to.clone()).or_insert(f32::INFINITY);
                if next_cost < *entry {
                    *entry = next_cost;
                    prev.insert(rel.to.clone(), current.clone());
                    heap.push((OrdF32(-next_cost), rel.to.clone()));
                }
            }
        }

        Ok(None)
    }

    /// Compute the transitive closure: all entities reachable from `start`.
    pub fn transitive_closure(
        &self,
        start: &EntityId,
    ) -> Result<HashSet<EntityId>, AgentRuntimeError> {
        let reachable = self.bfs(start)?;
        let mut set: HashSet<EntityId> = reachable.into_iter().collect();
        set.insert(start.clone());
        Ok(set)
    }

    /// Return the number of entities in the graph.
    pub fn entity_count(&self) -> Result<usize, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "entity_count");
        Ok(inner.entities.len())
    }

    /// Return the number of relationships in the graph.
    pub fn relationship_count(&self) -> Result<usize, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "relationship_count");
        Ok(inner.relationships.len())
    }

    /// Compute normalized degree centrality for each entity.
    /// Degree = (in_degree + out_degree) / (n - 1), or 0.0 if n <= 1.
    pub fn degree_centrality(&self) -> Result<HashMap<EntityId, f32>, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "degree_centrality");
        let n = inner.entities.len();

        let mut out_degree: HashMap<EntityId, usize> = HashMap::new();
        let mut in_degree: HashMap<EntityId, usize> = HashMap::new();

        for id in inner.entities.keys() {
            out_degree.insert(id.clone(), 0);
            in_degree.insert(id.clone(), 0);
        }

        for rel in &inner.relationships {
            *out_degree.entry(rel.from.clone()).or_insert(0) += 1;
            *in_degree.entry(rel.to.clone()).or_insert(0) += 1;
        }

        let denom = if n <= 1 { 1.0 } else { (n - 1) as f32 };
        let mut result = HashMap::new();
        for id in inner.entities.keys() {
            let od = *out_degree.get(id).unwrap_or(&0);
            let id_ = *in_degree.get(id).unwrap_or(&0);
            let centrality = if n <= 1 {
                0.0
            } else {
                (od + id_) as f32 / denom
            };
            result.insert(id.clone(), centrality);
        }

        Ok(result)
    }

    /// Compute normalized betweenness centrality for each entity.
    /// Uses Brandes' algorithm with hop-count BFS.
    ///
    /// # Complexity
    /// O(V * E) time. Not suitable for very large graphs (>1000 nodes).
    pub fn betweenness_centrality(&self) -> Result<HashMap<EntityId, f32>, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "betweenness_centrality");
        let n = inner.entities.len();
        let nodes: Vec<EntityId> = inner.entities.keys().cloned().collect();

        let mut centrality: HashMap<EntityId, f32> =
            nodes.iter().map(|id| (id.clone(), 0.0f32)).collect();

        for source in &nodes {
            // BFS to find shortest path counts and predecessors.
            let mut stack: Vec<EntityId> = Vec::new();
            let mut predecessors: HashMap<EntityId, Vec<EntityId>> =
                nodes.iter().map(|id| (id.clone(), vec![])).collect();
            let mut sigma: HashMap<EntityId, f32> =
                nodes.iter().map(|id| (id.clone(), 0.0f32)).collect();
            let mut dist: HashMap<EntityId, i64> =
                nodes.iter().map(|id| (id.clone(), -1i64)).collect();

            *sigma.entry(source.clone()).or_insert(0.0) = 1.0;
            *dist.entry(source.clone()).or_insert(-1) = 0;

            let mut queue: VecDeque<EntityId> = VecDeque::new();
            queue.push_back(source.clone());

            while let Some(v) = queue.pop_front() {
                stack.push(v.clone());
                let d_v = *dist.get(&v).unwrap_or(&0);
                let sigma_v = *sigma.get(&v).unwrap_or(&0.0);
                for rel in inner.relationships.iter().filter(|r| &r.from == &v) {
                    let w = &rel.to;
                    let d_w = dist.get(w).copied().unwrap_or(-1);
                    if d_w < 0 {
                        queue.push_back(w.clone());
                        *dist.entry(w.clone()).or_insert(-1) = d_v + 1;
                    }
                    if dist.get(w).copied().unwrap_or(-1) == d_v + 1 {
                        *sigma.entry(w.clone()).or_insert(0.0) += sigma_v;
                        predecessors.entry(w.clone()).or_default().push(v.clone());
                    }
                }
            }

            let mut delta: HashMap<EntityId, f32> =
                nodes.iter().map(|id| (id.clone(), 0.0f32)).collect();

            while let Some(w) = stack.pop() {
                let delta_w = *delta.get(&w).unwrap_or(&0.0);
                let sigma_w = *sigma.get(&w).unwrap_or(&1.0);
                for v in predecessors.get(&w).cloned().unwrap_or_default() {
                    let sigma_v = *sigma.get(&v).unwrap_or(&1.0);
                    let contribution = (sigma_v / sigma_w) * (1.0 + delta_w);
                    *delta.entry(v.clone()).or_insert(0.0) += contribution;
                }
                if &w != source {
                    *centrality.entry(w.clone()).or_insert(0.0) += delta_w;
                }
            }
        }

        // Normalize by 2 / ((n-1) * (n-2)) for directed graphs.
        if n > 2 {
            let norm = 2.0 / (((n - 1) * (n - 2)) as f32);
            for v in centrality.values_mut() {
                *v *= norm;
            }
        } else {
            for v in centrality.values_mut() {
                *v = 0.0;
            }
        }

        Ok(centrality)
    }

    /// Detect communities using label propagation.
    /// Each entity starts as its own community. In each iteration, each entity
    /// adopts the most frequent label among its neighbours.
    /// Returns a map of entity ID → community ID (usize).
    pub fn label_propagation_communities(
        &self,
        max_iterations: usize,
    ) -> Result<HashMap<EntityId, usize>, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "label_propagation_communities");
        let nodes: Vec<EntityId> = inner.entities.keys().cloned().collect();

        // Assign each node a unique initial label (index in nodes vec).
        let mut labels: HashMap<EntityId, usize> = nodes
            .iter()
            .enumerate()
            .map(|(i, id)| (id.clone(), i))
            .collect();

        for _ in 0..max_iterations {
            let mut changed = false;
            // Iterate in a stable order.
            for node in &nodes {
                // Collect neighbour labels.
                let neighbour_labels: Vec<usize> = inner
                    .relationships
                    .iter()
                    .filter(|r| &r.from == node || &r.to == node)
                    .map(|r| {
                        if &r.from == node {
                            labels.get(&r.to).copied().unwrap_or(0)
                        } else {
                            labels.get(&r.from).copied().unwrap_or(0)
                        }
                    })
                    .collect();

                if neighbour_labels.is_empty() {
                    continue;
                }

                // Find the most frequent label.
                let mut freq: HashMap<usize, usize> = HashMap::new();
                for &lbl in &neighbour_labels {
                    *freq.entry(lbl).or_insert(0) += 1;
                }
                let best = freq
                    .into_iter()
                    .max_by_key(|&(_, count)| count)
                    .map(|(lbl, _)| lbl);

                if let Some(new_label) = best {
                    let current = labels.entry(node.clone()).or_insert(0);
                    if *current != new_label {
                        *current = new_label;
                        changed = true;
                    }
                }
            }

            if !changed {
                break;
            }
        }

        Ok(labels)
    }

    /// Extract a subgraph containing only the specified entities and the
    /// relationships between them.
    pub fn subgraph(&self, node_ids: &[EntityId]) -> Result<GraphStore, AgentRuntimeError> {
        let inner = recover_lock(self.inner.lock(), "subgraph");
        let id_set: HashSet<&EntityId> = node_ids.iter().collect();

        let new_store = GraphStore::new();

        for id in node_ids {
            let entity = inner
                .entities
                .get(id)
                .ok_or_else(|| {
                    AgentRuntimeError::Graph(format!("entity '{}' not found", id.0))
                })?
                .clone();
            // We hold inner lock; call directly on the new store's inner.
            let mut new_inner = recover_lock(new_store.inner.lock(), "subgraph:add_entity");
            new_inner.entities.insert(entity.id.clone(), entity);
        }

        for rel in inner.relationships.iter() {
            if id_set.contains(&rel.from) && id_set.contains(&rel.to) {
                let mut new_inner =
                    recover_lock(new_store.inner.lock(), "subgraph:add_relationship");
                new_inner.relationships.push(rel.clone());
            }
        }

        Ok(new_store)
    }
}

impl Default for GraphStore {
    fn default() -> Self {
        Self::new()
    }
}

// ── Tests ─────────────────────────────────────────────────────────────────────

#[cfg(test)]
mod tests {
    use super::*;

    fn make_graph() -> GraphStore {
        GraphStore::new()
    }

    fn add(g: &GraphStore, id: &str) {
        g.add_entity(Entity::new(id, "Node")).unwrap();
    }

    fn link(g: &GraphStore, from: &str, to: &str) {
        g.add_relationship(Relationship::new(from, to, "CONNECTS", 1.0))
            .unwrap();
    }

    fn link_w(g: &GraphStore, from: &str, to: &str, weight: f32) {
        g.add_relationship(Relationship::new(from, to, "CONNECTS", weight))
            .unwrap();
    }

    // ── EntityId ──────────────────────────────────────────────────────────────

    #[test]
    fn test_entity_id_equality() {
        assert_eq!(EntityId::new("a"), EntityId::new("a"));
        assert_ne!(EntityId::new("a"), EntityId::new("b"));
    }

    #[test]
    fn test_entity_id_display() {
        let id = EntityId::new("hello");
        assert_eq!(id.to_string(), "hello");
    }

    // ── Entity ────────────────────────────────────────────────────────────────

    #[test]
    fn test_entity_new_has_empty_properties() {
        let e = Entity::new("e1", "Person");
        assert!(e.properties.is_empty());
    }

    #[test]
    fn test_entity_with_properties_stores_props() {
        let mut props = HashMap::new();
        props.insert("age".into(), Value::Number(42.into()));
        let e = Entity::with_properties("e1", "Person", props);
        assert!(e.properties.contains_key("age"));
    }

    // ── GraphStore basic ops ──────────────────────────────────────────────────

    #[test]
    fn test_graph_add_entity_increments_count() {
        let g = make_graph();
        add(&g, "a");
        assert_eq!(g.entity_count().unwrap(), 1);
    }

    #[test]
    fn test_graph_get_entity_returns_entity() {
        let g = make_graph();
        g.add_entity(Entity::new("e1", "Person")).unwrap();
        let e = g.get_entity(&EntityId::new("e1")).unwrap();
        assert_eq!(e.label, "Person");
    }

    #[test]
    fn test_graph_get_entity_missing_returns_error() {
        let g = make_graph();
        assert!(g.get_entity(&EntityId::new("ghost")).is_err());
    }

    #[test]
    fn test_graph_add_relationship_increments_count() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        link(&g, "a", "b");
        assert_eq!(g.relationship_count().unwrap(), 1);
    }

    #[test]
    fn test_graph_add_relationship_missing_source_fails() {
        let g = make_graph();
        add(&g, "b");
        let result = g.add_relationship(Relationship::new("ghost", "b", "X", 1.0));
        assert!(result.is_err());
    }

    #[test]
    fn test_graph_add_relationship_missing_target_fails() {
        let g = make_graph();
        add(&g, "a");
        let result = g.add_relationship(Relationship::new("a", "ghost", "X", 1.0));
        assert!(result.is_err());
    }

    #[test]
    fn test_graph_remove_entity_removes_relationships() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        link(&g, "a", "b");
        g.remove_entity(&EntityId::new("a")).unwrap();
        assert_eq!(g.entity_count().unwrap(), 1);
        assert_eq!(g.relationship_count().unwrap(), 0);
    }

    #[test]
    fn test_graph_remove_entity_missing_returns_error() {
        let g = make_graph();
        assert!(g.remove_entity(&EntityId::new("ghost")).is_err());
    }

    // ── BFS ───────────────────────────────────────────────────────────────────

    #[test]
    fn test_bfs_finds_direct_neighbours() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        link(&g, "a", "b");
        link(&g, "a", "c");
        let visited = g.bfs(&EntityId::new("a")).unwrap();
        assert_eq!(visited.len(), 2);
    }

    #[test]
    fn test_bfs_traverses_chain() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        add(&g, "d");
        link(&g, "a", "b");
        link(&g, "b", "c");
        link(&g, "c", "d");
        let visited = g.bfs(&EntityId::new("a")).unwrap();
        assert_eq!(visited.len(), 3);
        assert_eq!(visited[0], EntityId::new("b"));
    }

    #[test]
    fn test_bfs_handles_isolated_node() {
        let g = make_graph();
        add(&g, "a");
        let visited = g.bfs(&EntityId::new("a")).unwrap();
        assert!(visited.is_empty());
    }

    #[test]
    fn test_bfs_missing_start_returns_error() {
        let g = make_graph();
        assert!(g.bfs(&EntityId::new("ghost")).is_err());
    }

    // ── DFS ───────────────────────────────────────────────────────────────────

    #[test]
    fn test_dfs_visits_all_reachable_nodes() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        add(&g, "d");
        link(&g, "a", "b");
        link(&g, "a", "c");
        link(&g, "b", "d");
        let visited = g.dfs(&EntityId::new("a")).unwrap();
        assert_eq!(visited.len(), 3);
    }

    #[test]
    fn test_dfs_handles_isolated_node() {
        let g = make_graph();
        add(&g, "a");
        let visited = g.dfs(&EntityId::new("a")).unwrap();
        assert!(visited.is_empty());
    }

    #[test]
    fn test_dfs_missing_start_returns_error() {
        let g = make_graph();
        assert!(g.dfs(&EntityId::new("ghost")).is_err());
    }

    // ── Shortest path ─────────────────────────────────────────────────────────

    #[test]
    fn test_shortest_path_direct_connection() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        link(&g, "a", "b");
        let path = g
            .shortest_path(&EntityId::new("a"), &EntityId::new("b"))
            .unwrap();
        assert_eq!(path, Some(vec![EntityId::new("a"), EntityId::new("b")]));
    }

    #[test]
    fn test_shortest_path_multi_hop() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        link(&g, "a", "b");
        link(&g, "b", "c");
        let path = g
            .shortest_path(&EntityId::new("a"), &EntityId::new("c"))
            .unwrap();
        assert_eq!(path.as_ref().map(|p| p.len()), Some(3));
    }

    #[test]
    fn test_shortest_path_returns_none_for_disconnected() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        let path = g
            .shortest_path(&EntityId::new("a"), &EntityId::new("b"))
            .unwrap();
        assert_eq!(path, None);
    }

    #[test]
    fn test_shortest_path_same_node_returns_single_element() {
        let g = make_graph();
        add(&g, "a");
        let path = g
            .shortest_path(&EntityId::new("a"), &EntityId::new("a"))
            .unwrap();
        assert_eq!(path, Some(vec![EntityId::new("a")]));
    }

    #[test]
    fn test_shortest_path_missing_source_returns_error() {
        let g = make_graph();
        add(&g, "b");
        assert!(g
            .shortest_path(&EntityId::new("ghost"), &EntityId::new("b"))
            .is_err());
    }

    #[test]
    fn test_shortest_path_missing_target_returns_error() {
        let g = make_graph();
        add(&g, "a");
        assert!(g
            .shortest_path(&EntityId::new("a"), &EntityId::new("ghost"))
            .is_err());
    }

    // ── Transitive closure ────────────────────────────────────────────────────

    #[test]
    fn test_transitive_closure_includes_start() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        link(&g, "a", "b");
        let closure = g.transitive_closure(&EntityId::new("a")).unwrap();
        assert!(closure.contains(&EntityId::new("a")));
        assert!(closure.contains(&EntityId::new("b")));
    }

    #[test]
    fn test_transitive_closure_isolated_node_contains_only_self() {
        let g = make_graph();
        add(&g, "a");
        let closure = g.transitive_closure(&EntityId::new("a")).unwrap();
        assert_eq!(closure.len(), 1);
    }

    // ── MemGraphError conversion ──────────────────────────────────────────────

    #[test]
    fn test_mem_graph_error_converts_to_runtime_error() {
        let e = MemGraphError::EntityNotFound("x".into());
        let re: AgentRuntimeError = e.into();
        assert!(matches!(re, AgentRuntimeError::Graph(_)));
    }

    // ── Weighted shortest path ────────────────────────────────────────────────

    #[test]
    fn test_shortest_path_weighted_simple() {
        // a --(1.0)--> b --(2.0)--> c
        // a --(10.0)--> c  (direct but heavier)
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        link_w(&g, "a", "b", 1.0);
        link_w(&g, "b", "c", 2.0);
        g.add_relationship(Relationship::new("a", "c", "DIRECT", 10.0))
            .unwrap();

        let result = g
            .shortest_path_weighted(&EntityId::new("a"), &EntityId::new("c"))
            .unwrap();
        assert!(result.is_some());
        let (path, weight) = result.unwrap();
        // The cheapest path is a -> b -> c with total weight 3.0
        assert_eq!(
            path,
            vec![EntityId::new("a"), EntityId::new("b"), EntityId::new("c")]
        );
        assert!((weight - 3.0).abs() < 1e-5);
    }

    #[test]
    fn test_shortest_path_weighted_returns_none_for_disconnected() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        let result = g
            .shortest_path_weighted(&EntityId::new("a"), &EntityId::new("b"))
            .unwrap();
        assert!(result.is_none());
    }

    #[test]
    fn test_shortest_path_weighted_same_node() {
        let g = make_graph();
        add(&g, "a");
        let result = g
            .shortest_path_weighted(&EntityId::new("a"), &EntityId::new("a"))
            .unwrap();
        assert_eq!(result, Some((vec![EntityId::new("a")], 0.0)));
    }

    #[test]
    fn test_shortest_path_weighted_negative_weight_errors() {
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        g.add_relationship(Relationship::new("a", "b", "NEG", -1.0))
            .unwrap();
        let result = g.shortest_path_weighted(&EntityId::new("a"), &EntityId::new("b"));
        assert!(result.is_err());
    }

    // ── Degree centrality ─────────────────────────────────────────────────────

    #[test]
    fn test_degree_centrality_basic() {
        // Star graph: a -> b, a -> c, a -> d
        // a: out=3, in=0 => (3+0)/(4-1) = 1.0
        // b: out=0, in=1 => (0+1)/3 = 0.333...
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        add(&g, "d");
        link(&g, "a", "b");
        link(&g, "a", "c");
        link(&g, "a", "d");

        let centrality = g.degree_centrality().unwrap();
        let a_cent = *centrality.get(&EntityId::new("a")).unwrap();
        let b_cent = *centrality.get(&EntityId::new("b")).unwrap();

        assert!((a_cent - 1.0).abs() < 1e-5, "a centrality was {a_cent}");
        assert!(
            (b_cent - 1.0 / 3.0).abs() < 1e-5,
            "b centrality was {b_cent}"
        );
    }

    // ── Betweenness centrality ────────────────────────────────────────────────

    #[test]
    fn test_betweenness_centrality_chain() {
        // Chain: a -> b -> c -> d
        // b and c are on all paths from a to c, a to d, b to d
        // Node b and c should have higher centrality than a and d.
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        add(&g, "d");
        link(&g, "a", "b");
        link(&g, "b", "c");
        link(&g, "c", "d");

        let centrality = g.betweenness_centrality().unwrap();
        let a_cent = *centrality.get(&EntityId::new("a")).unwrap();
        let b_cent = *centrality.get(&EntityId::new("b")).unwrap();
        let c_cent = *centrality.get(&EntityId::new("c")).unwrap();
        let d_cent = *centrality.get(&EntityId::new("d")).unwrap();

        // Endpoints should have 0 centrality; intermediate nodes should be > 0.
        assert!(
            (a_cent).abs() < 1e-5,
            "expected a_cent ~ 0, got {a_cent}"
        );
        assert!(b_cent > 0.0, "expected b_cent > 0, got {b_cent}");
        assert!(c_cent > 0.0, "expected c_cent > 0, got {c_cent}");
        assert!(
            (d_cent).abs() < 1e-5,
            "expected d_cent ~ 0, got {d_cent}"
        );
    }

    // ── Label propagation communities ─────────────────────────────────────────

    #[test]
    fn test_label_propagation_communities_two_clusters() {
        // Cluster 1: a <-> b <-> c (fully connected via bidirectional edges)
        // Cluster 2: x <-> y <-> z
        // No edges between clusters.
        let g = make_graph();
        for id in &["a", "b", "c", "x", "y", "z"] {
            add(&g, id);
        }
        // Cluster 1 (bidirectional via two directed edges each)
        link(&g, "a", "b");
        link(&g, "b", "a");
        link(&g, "b", "c");
        link(&g, "c", "b");
        link(&g, "a", "c");
        link(&g, "c", "a");
        // Cluster 2
        link(&g, "x", "y");
        link(&g, "y", "x");
        link(&g, "y", "z");
        link(&g, "z", "y");
        link(&g, "x", "z");
        link(&g, "z", "x");

        let communities = g.label_propagation_communities(100).unwrap();

        let label_a = communities[&EntityId::new("a")];
        let label_b = communities[&EntityId::new("b")];
        let label_c = communities[&EntityId::new("c")];
        let label_x = communities[&EntityId::new("x")];
        let label_y = communities[&EntityId::new("y")];
        let label_z = communities[&EntityId::new("z")];

        // All nodes in cluster 1 share a label, all in cluster 2 share a label,
        // and the two clusters have different labels.
        assert_eq!(label_a, label_b, "a and b should be in same community");
        assert_eq!(label_b, label_c, "b and c should be in same community");
        assert_eq!(label_x, label_y, "x and y should be in same community");
        assert_eq!(label_y, label_z, "y and z should be in same community");
        assert_ne!(
            label_a, label_x,
            "cluster 1 and cluster 2 should be different communities"
        );
    }

    // ── Subgraph extraction ───────────────────────────────────────────────────

    #[test]
    fn test_subgraph_extracts_correct_nodes_and_edges() {
        // Full graph: a -> b -> c -> d
        // Subgraph of {a, b, c} should contain edges a->b and b->c but not c->d.
        let g = make_graph();
        add(&g, "a");
        add(&g, "b");
        add(&g, "c");
        add(&g, "d");
        link(&g, "a", "b");
        link(&g, "b", "c");
        link(&g, "c", "d");

        let sub = g
            .subgraph(&[
                EntityId::new("a"),
                EntityId::new("b"),
                EntityId::new("c"),
            ])
            .unwrap();

        assert_eq!(sub.entity_count().unwrap(), 3);
        assert_eq!(sub.relationship_count().unwrap(), 2);

        // d should not be present in the subgraph.
        assert!(sub.get_entity(&EntityId::new("d")).is_err());

        // a -> b and b -> c should be present; c -> d should not.
        let path = sub
            .shortest_path(&EntityId::new("a"), &EntityId::new("c"))
            .unwrap();
        assert!(path.is_some());
        assert_eq!(path.unwrap().len(), 3);
    }
}