Struct CsrGraph 

pub struct CsrGraph {
    pub offsets: Vec<u64>,
    pub targets: Vec<NodeId>,
    pub weights: Vec<AtomicU32>,
    pub inhibitory: Vec<bool>,
    pub relations: Vec<InternedStr>,
    pub directions: Vec<EdgeDirection>,
    pub causal_strengths: Vec<FiniteF32>,
    pub rev_offsets: Vec<u64>,
    pub rev_sources: Vec<NodeId>,
    pub rev_edge_idx: Vec<EdgeIdx>,
    pub pending_edges: Vec<PendingEdge>,
}

Compressed Sparse Row graph with forward and reverse adjacency. For node i, outgoing edges span offsets[i]..offsets[i+1] into targets, weights, inhibitory, relations, directions, causal_strengths.

Fields

offsets: Vec<u64>

Length: num_nodes + 1. offsets[num_nodes] == total_edges.

targets: Vec<NodeId>

Length: total_edges. Target node for each edge.

weights: Vec<AtomicU32>

Length: total_edges. Edge weight — atomic for lock-free plasticity updates (FM-ACT-021).

inhibitory: Vec<bool>

Length: total_edges. true = inhibitory edge.

relations: Vec<InternedStr>

Length: total_edges. Relation type (interned).

directions: Vec<EdgeDirection>

Length: total_edges. Forward or Bidirectional.

causal_strengths: Vec<FiniteF32>

Length: total_edges. Causal strength in [0.0, 1.0].

rev_offsets: Vec<u64>

Length: num_nodes + 1.

rev_sources: Vec<NodeId>

Length: total_edges. Source node for each reverse edge.

rev_edge_idx: Vec<EdgeIdx>

Length: total_edges. Index into forward arrays for this reverse edge.

pending_edges: Vec<PendingEdge>

Pre-finalize edge staging area.

Implementations

impl CsrGraph

pub fn empty() -> Self

Create an empty CSR with no nodes/edges.

pub fn num_edges(&self) -> usize

Number of edges in the forward CSR.

pub fn out_range(&self, node: NodeId) -> Range<usize>

Outgoing edge range for node.

pub fn in_range(&self, node: NodeId) -> Range<usize>

Incoming edge range for node (reverse CSR).

pub fn read_weight(&self, edge: EdgeIdx) -> FiniteF32

Read weight atomically as FiniteF32 (FM-ACT-021).

pub fn atomic_max_weight( &self, edge: EdgeIdx, new_val: FiniteF32, max_retries: u32, ) -> M1ndResult<()>

Atomic CAS max on edge weight. Returns Ok on success, Err after retry limit (FM-ACT-019). Replaces: engine_fast.py direct weight assignment (now lock-free).

pub fn atomic_write_weight( &self, edge: EdgeIdx, new_val: FiniteF32, max_retries: u32, ) -> M1ndResult<()>

Atomic CAS write on edge weight (for plasticity). Returns Ok on success, Err after retry limit.