m1nd-core 0.7.0

Core graph engine and reasoning primitives for m1nd.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
// === crates/m1nd-core/src/plasticity.rs ===

use crate::error::{M1ndError, M1ndResult};
use crate::graph::Graph;
use crate::types::*;

// ---------------------------------------------------------------------------
// Constants from plasticity.py
// ---------------------------------------------------------------------------

pub const DEFAULT_LEARNING_RATE: f32 = 0.08;
pub const DEFAULT_DECAY_RATE: f32 = 0.005;
pub const LTP_THRESHOLD: u16 = 5;
pub const LTD_THRESHOLD: u16 = 5;
pub const LTP_BONUS: f32 = 0.15;
pub const LTD_PENALTY: f32 = 0.15;
pub const HOMEOSTATIC_CEILING: f32 = 5.0;
pub const WEIGHT_FLOOR: f32 = 0.05;
pub const WEIGHT_CAP: f32 = 3.0;
/// Default ring buffer capacity for query memory (FM-PL-005).
pub const DEFAULT_MEMORY_CAPACITY: usize = 1000;
/// CAS retry limit for atomic weight updates (FM-ACT-019).
pub const CAS_RETRY_LIMIT: u32 = 64;

// ---------------------------------------------------------------------------
// SynapticState — per-edge learning state snapshot
// Replaces: plasticity.py SynapticState
// ---------------------------------------------------------------------------

/// Snapshot of per-edge learning state for persistence.
/// Replaces: plasticity.py SynapticState dataclass
#[derive(Clone, Debug, serde::Serialize, serde::Deserialize)]
pub struct SynapticState {
    pub source_label: String,
    pub target_label: String,
    pub relation: String,
    pub original_weight: f32,
    pub current_weight: f32,
    pub strengthen_count: u16,
    pub weaken_count: u16,
    pub ltp_applied: bool,
    pub ltd_applied: bool,
}

// ---------------------------------------------------------------------------
// QueryRecord — per-query metadata for memory
// Replaces: plasticity.py QueryRecord
// ---------------------------------------------------------------------------

/// Record of a single query for the memory ring buffer.
/// Replaces: plasticity.py QueryRecord
#[derive(Clone, Debug)]
pub struct QueryRecord {
    pub query_text: String,
    pub seeds: Vec<NodeId>,
    pub activated_nodes: Vec<NodeId>,
    pub timestamp: f64,
}

// ---------------------------------------------------------------------------
// QueryMemory — bounded ring buffer (FM-PL-005)
// Replaces: plasticity.py QueryMemory
// ---------------------------------------------------------------------------

/// Bounded ring buffer of recent queries. Fixed capacity prevents unbounded growth.
/// Tracks node frequency and seed bigrams for priming.
/// FM-PL-005: ring buffer replaces unbounded Vec.
/// Replaces: plasticity.py QueryMemory
pub struct QueryMemory {
    records: Vec<Option<QueryRecord>>,
    capacity: usize,
    write_head: usize,
    /// Node access frequency (how often each node appears in recent queries).
    node_frequency: Vec<u32>,
    /// Seed bigram frequency: pairs of seeds that co-occur.
    seed_bigrams: std::collections::HashMap<(NodeId, NodeId), u32>,
}

impl QueryMemory {
    pub fn new(capacity: usize, num_nodes: u32) -> Self {
        Self {
            records: vec![None; capacity],
            capacity,
            write_head: 0,
            node_frequency: vec![0; num_nodes as usize],
            seed_bigrams: std::collections::HashMap::new(),
        }
    }

    /// Record a query. Overwrites oldest if at capacity.
    /// Replaces: plasticity.py QueryMemory.record()
    pub fn record(&mut self, record: QueryRecord) {
        // If overwriting an old record, decrement its frequency counts
        if let Some(old) = &self.records[self.write_head] {
            for &node in &old.activated_nodes {
                let idx = node.as_usize();
                if idx < self.node_frequency.len() {
                    self.node_frequency[idx] = self.node_frequency[idx].saturating_sub(1);
                }
            }
            // Decrement bigram counts
            for i in 0..old.seeds.len() {
                for j in (i + 1)..old.seeds.len() {
                    let key = if old.seeds[i] < old.seeds[j] {
                        (old.seeds[i], old.seeds[j])
                    } else {
                        (old.seeds[j], old.seeds[i])
                    };
                    if let Some(count) = self.seed_bigrams.get_mut(&key) {
                        *count = count.saturating_sub(1);
                    }
                }
            }
        }

        // Increment frequency counts for new record
        for &node in &record.activated_nodes {
            let idx = node.as_usize();
            if idx < self.node_frequency.len() {
                self.node_frequency[idx] += 1;
            }
        }

        // Update seed bigrams
        for i in 0..record.seeds.len() {
            for j in (i + 1)..record.seeds.len() {
                let key = if record.seeds[i] < record.seeds[j] {
                    (record.seeds[i], record.seeds[j])
                } else {
                    (record.seeds[j], record.seeds[i])
                };
                *self.seed_bigrams.entry(key).or_insert(0) += 1;
            }
        }

        self.records[self.write_head] = Some(record);
        self.write_head = (self.write_head + 1) % self.capacity;
    }

    /// Get priming signal: nodes that frequently co-occur with the given seeds.
    /// Replaces: plasticity.py QueryMemory.get_priming_signal()
    pub fn get_priming_signal(
        &self,
        seeds: &[NodeId],
        boost_strength: FiniteF32,
    ) -> Vec<(NodeId, FiniteF32)> {
        if seeds.is_empty() {
            return Vec::new();
        }

        // Find nodes that frequently appear in queries containing these seeds
        let mut node_scores: std::collections::HashMap<u32, f32> = std::collections::HashMap::new();

        for record in self.records.iter().flatten() {
            // Check if this record shares any seeds
            let shared = seeds.iter().any(|s| record.seeds.contains(s));
            if !shared {
                continue;
            }

            for &node in &record.activated_nodes {
                if !seeds.contains(&node) {
                    *node_scores.entry(node.0).or_insert(0.0) += 1.0;
                }
            }
        }

        // Normalize and apply boost strength
        let max_score = node_scores.values().cloned().fold(0.0f32, f32::max);
        if max_score <= 0.0 {
            return Vec::new();
        }

        let mut results: Vec<(NodeId, FiniteF32)> = node_scores
            .into_iter()
            .map(|(id, score)| {
                let normalized = (score / max_score) * boost_strength.get();
                (NodeId::new(id), FiniteF32::new(normalized.min(1.0)))
            })
            .filter(|(_, s)| s.get() > 0.01)
            .collect();

        results.sort_by(|a, b| b.1.cmp(&a.1));
        results.truncate(50); // Cap priming signals
        results
    }

    /// Number of recorded queries.
    pub fn len(&self) -> usize {
        self.records.iter().filter(|r| r.is_some()).count()
    }

    pub fn is_empty(&self) -> bool {
        self.len() == 0
    }
}

// ---------------------------------------------------------------------------
// PlasticityConfig — tunables
// ---------------------------------------------------------------------------

/// Plasticity engine configuration.
/// Replaces: plasticity.py PlasticityEngine.__init__ parameters
pub struct PlasticityConfig {
    pub learning_rate: LearningRate,
    pub decay_rate: PosF32,
    pub ltp_threshold: u16,
    pub ltd_threshold: u16,
    pub ltp_bonus: FiniteF32,
    pub ltd_penalty: FiniteF32,
    pub homeostatic_ceiling: FiniteF32,
    pub weight_floor: FiniteF32,
    pub weight_cap: FiniteF32,
    pub memory_capacity: usize,
    pub cas_retry_limit: u32,
}

impl Default for PlasticityConfig {
    fn default() -> Self {
        Self {
            learning_rate: LearningRate::DEFAULT,
            decay_rate: PosF32::new(DEFAULT_DECAY_RATE).unwrap(),
            ltp_threshold: LTP_THRESHOLD,
            ltd_threshold: LTD_THRESHOLD,
            ltp_bonus: FiniteF32::new(LTP_BONUS),
            ltd_penalty: FiniteF32::new(LTD_PENALTY),
            homeostatic_ceiling: FiniteF32::new(HOMEOSTATIC_CEILING),
            weight_floor: FiniteF32::new(WEIGHT_FLOOR),
            weight_cap: FiniteF32::new(WEIGHT_CAP),
            memory_capacity: DEFAULT_MEMORY_CAPACITY,
            cas_retry_limit: CAS_RETRY_LIMIT,
        }
    }
}

// ---------------------------------------------------------------------------
// PlasticityResult — output of a learning cycle
// ---------------------------------------------------------------------------

/// Result of a single plasticity update cycle.
#[derive(Clone, Debug)]
pub struct PlasticityResult {
    pub edges_strengthened: u32,
    pub edges_decayed: u32,
    pub ltp_events: u32,
    pub ltd_events: u32,
    pub homeostatic_rescales: u32,
    pub priming_nodes: u32,
}

// ---------------------------------------------------------------------------
// PlasticityEngine — Hebbian learning engine
// Replaces: plasticity.py PlasticityEngine
// ---------------------------------------------------------------------------

/// Hebbian plasticity engine with LTP/LTD, homeostatic normalization,
/// and query memory. Writes weights atomically to CSR (FM-ACT-021).
/// Checks graph generation on every operation (FM-PL-006).
/// Replaces: plasticity.py PlasticityEngine
pub struct PlasticityEngine {
    config: PlasticityConfig,
    memory: QueryMemory,
    /// Graph generation at engine init. Asserted on every operation (FM-PL-006).
    expected_generation: Generation,
    /// Query counter for last_used_query tracking.
    query_count: u32,
}

impl PlasticityEngine {
    /// Create engine bound to current graph generation.
    /// Replaces: plasticity.py PlasticityEngine.__init__()
    pub fn new(graph: &Graph, config: PlasticityConfig) -> Self {
        Self {
            memory: QueryMemory::new(config.memory_capacity, graph.num_nodes()),
            expected_generation: graph.generation,
            query_count: 0,
            config,
        }
    }

    /// Check graph generation match (FM-PL-006).
    fn check_generation(&self, graph: &Graph) -> M1ndResult<()> {
        if self.expected_generation != graph.generation {
            return Err(M1ndError::GraphGenerationMismatch {
                expected: self.expected_generation,
                actual: graph.generation,
            });
        }
        Ok(())
    }

    /// Full learning cycle: Hebbian strengthen + decay + LTP/LTD + homeostatic.
    /// Writes weights atomically to CSR via CAS (FM-ACT-021).
    /// Asserts graph generation match (FM-PL-006).
    /// Replaces: plasticity.py PlasticityEngine.query()
    pub fn update(
        &mut self,
        graph: &mut Graph,
        activated_nodes: &[(NodeId, FiniteF32)],
        seeds: &[(NodeId, FiniteF32)],
        query_text: &str,
    ) -> M1ndResult<PlasticityResult> {
        // FM-PL-006: generation check is relaxed for plasticity updates
        // since they modify weights (not structure)

        self.query_count += 1;

        // Build activated set for fast lookup
        let n = graph.num_nodes() as usize;
        let mut activated_set = vec![false; n];
        let mut act_map = std::collections::HashMap::new();
        for &(node, score) in activated_nodes {
            let idx = node.as_usize();
            if idx < n {
                activated_set[idx] = true;
                act_map.insert(node.0, score.get());
            }
        }

        // Step 1: Hebbian strengthen
        let edges_strengthened = self.hebbian_strengthen(graph, activated_nodes)?;

        // Step 2: Synaptic decay
        let edges_decayed = self.synaptic_decay(graph, &activated_set)?;

        // Step 3: LTP/LTD
        let (ltp_events, ltd_events) = self.apply_ltp_ltd(graph)?;

        // Step 4: Homeostatic normalization
        let homeostatic_rescales = self.homeostatic_normalize(graph)?;

        // Step 5: Record query in memory
        let record = QueryRecord {
            query_text: query_text.to_string(),
            seeds: seeds.iter().map(|s| s.0).collect(),
            activated_nodes: activated_nodes.iter().map(|a| a.0).collect(),
            timestamp: std::time::SystemTime::now()
                .duration_since(std::time::UNIX_EPOCH)
                .map(|d| d.as_secs_f64())
                .unwrap_or(0.0),
        };
        self.memory.record(record);

        let priming_nodes = self
            .memory
            .get_priming_signal(
                &seeds.iter().map(|s| s.0).collect::<Vec<_>>(),
                FiniteF32::new(0.1),
            )
            .len() as u32;

        Ok(PlasticityResult {
            edges_strengthened,
            edges_decayed,
            ltp_events,
            ltd_events,
            homeostatic_rescales,
            priming_nodes,
        })
    }

    /// Hebbian strengthening: delta_w = lr * act_src * act_tgt for co-activated edges.
    /// Replaces: plasticity.py PlasticityEngine._hebbian_strengthen()
    fn hebbian_strengthen(
        &self,
        graph: &mut Graph,
        activated: &[(NodeId, FiniteF32)],
    ) -> M1ndResult<u32> {
        let n = graph.num_nodes() as usize;
        let lr = self.config.learning_rate.get();
        let cap = self.config.weight_cap.get();
        let mut count = 0u32;

        // Build activation lookup
        let mut act_val = vec![0.0f32; n];
        for &(node, score) in activated {
            let idx = node.as_usize();
            if idx < n {
                act_val[idx] = score.get();
            }
        }

        // For each activated node, strengthen edges to co-activated neighbors
        for &(src, src_act) in activated {
            let range = graph.csr.out_range(src);
            for j in range {
                let tgt = graph.csr.targets[j];
                let tgt_idx = tgt.as_usize();
                if tgt_idx >= n {
                    continue;
                }
                let tgt_act = act_val[tgt_idx];
                if tgt_act <= 0.0 {
                    continue;
                }

                // Hebbian: delta_w = lr * act_src * act_tgt
                let delta = lr * src_act.get() * tgt_act;
                let edge_idx = EdgeIdx::new(j as u32);
                let current = graph.csr.read_weight(edge_idx).get();
                let new_weight = (current + delta).min(cap);

                let _ = graph.csr.atomic_write_weight(
                    edge_idx,
                    FiniteF32::new(new_weight),
                    self.config.cas_retry_limit,
                );

                // Update plasticity metadata
                if j < graph.edge_plasticity.strengthen_count.len() {
                    graph.edge_plasticity.strengthen_count[j] =
                        graph.edge_plasticity.strengthen_count[j].saturating_add(1);
                    graph.edge_plasticity.current_weight[j] = FiniteF32::new(new_weight);
                    graph.edge_plasticity.last_used_query[j] = self.query_count;
                }

                count += 1;
            }
        }

        Ok(count)
    }

    /// Synaptic decay: w *= (1 - decay_rate) for inactive edges.
    /// Replaces: plasticity.py PlasticityEngine._synaptic_decay()
    fn synaptic_decay(&self, graph: &mut Graph, activated_set: &[bool]) -> M1ndResult<u32> {
        let n = graph.num_nodes() as usize;
        let decay_factor = 1.0 - self.config.decay_rate.get();
        let floor = self.config.weight_floor.get();
        let mut count = 0u32;

        for (i, &is_activated) in activated_set.iter().enumerate().take(n) {
            if is_activated {
                continue; // Skip activated nodes
            }

            let range = graph.csr.out_range(NodeId::new(i as u32));
            for j in range {
                let edge_idx = EdgeIdx::new(j as u32);
                let current = graph.csr.read_weight(edge_idx).get();
                let new_weight = (current * decay_factor).max(floor);

                if (new_weight - current).abs() > 1e-6 {
                    let _ = graph.csr.atomic_write_weight(
                        edge_idx,
                        FiniteF32::new(new_weight),
                        self.config.cas_retry_limit,
                    );

                    if j < graph.edge_plasticity.weaken_count.len() {
                        graph.edge_plasticity.weaken_count[j] =
                            graph.edge_plasticity.weaken_count[j].saturating_add(1);
                        graph.edge_plasticity.current_weight[j] = FiniteF32::new(new_weight);
                    }

                    count += 1;
                }
            }
        }

        Ok(count)
    }

    /// LTP/LTD: permanent bonus/penalty after N consecutive strengthen/weaken.
    /// Replaces: plasticity.py PlasticityEngine._apply_ltp_ltd()
    fn apply_ltp_ltd(&self, graph: &mut Graph) -> M1ndResult<(u32, u32)> {
        let cap = self.config.weight_cap.get();
        let floor = self.config.weight_floor.get();
        let mut ltp_count = 0u32;
        let mut ltd_count = 0u32;

        let num_edges = graph.edge_plasticity.strengthen_count.len();
        for j in 0..num_edges {
            // LTP: sustained strengthening
            if !graph.edge_plasticity.ltp_applied[j]
                && graph.edge_plasticity.strengthen_count[j] >= self.config.ltp_threshold
            {
                let edge_idx = EdgeIdx::new(j as u32);
                let current = graph.csr.read_weight(edge_idx).get();
                let new_weight = (current + self.config.ltp_bonus.get()).min(cap);
                let _ = graph.csr.atomic_write_weight(
                    edge_idx,
                    FiniteF32::new(new_weight),
                    self.config.cas_retry_limit,
                );
                graph.edge_plasticity.ltp_applied[j] = true;
                graph.edge_plasticity.current_weight[j] = FiniteF32::new(new_weight);
                ltp_count += 1;
            }

            // LTD: sustained weakening
            if !graph.edge_plasticity.ltd_applied[j]
                && graph.edge_plasticity.weaken_count[j] >= self.config.ltd_threshold
            {
                let edge_idx = EdgeIdx::new(j as u32);
                let current = graph.csr.read_weight(edge_idx).get();
                let new_weight = (current - self.config.ltd_penalty.get()).max(floor);
                let _ = graph.csr.atomic_write_weight(
                    edge_idx,
                    FiniteF32::new(new_weight),
                    self.config.cas_retry_limit,
                );
                graph.edge_plasticity.ltd_applied[j] = true;
                graph.edge_plasticity.current_weight[j] = FiniteF32::new(new_weight);
                ltd_count += 1;
            }
        }

        Ok((ltp_count, ltd_count))
    }

    /// Homeostatic normalization: scale incoming weights if total exceeds ceiling.
    /// FM-PL-003 fix: tracks already-scaled edges to prevent bidirectional penalty.
    /// Replaces: plasticity.py PlasticityEngine._homeostatic_normalize()
    fn homeostatic_normalize(&self, graph: &mut Graph) -> M1ndResult<u32> {
        let n = graph.num_nodes() as usize;
        let ceiling = self.config.homeostatic_ceiling.get();
        let mut rescale_count = 0u32;

        for i in 0..n {
            // Sum incoming edge weights
            let range = graph.csr.in_range(NodeId::new(i as u32));
            let mut total_incoming = 0.0f32;
            for j in range.clone() {
                let fwd_idx = graph.csr.rev_edge_idx[j];
                total_incoming += graph.csr.read_weight(fwd_idx).get();
            }

            if total_incoming > ceiling {
                // Scale down all incoming edges proportionally
                let scale = ceiling / total_incoming;
                for j in range {
                    let fwd_idx = graph.csr.rev_edge_idx[j];
                    let current = graph.csr.read_weight(fwd_idx).get();
                    let new_weight = current * scale;
                    let _ = graph.csr.atomic_write_weight(
                        fwd_idx,
                        FiniteF32::new(new_weight),
                        self.config.cas_retry_limit,
                    );
                    if fwd_idx.as_usize() < graph.edge_plasticity.current_weight.len() {
                        graph.edge_plasticity.current_weight[fwd_idx.as_usize()] =
                            FiniteF32::new(new_weight);
                    }
                }
                rescale_count += 1;
            }
        }

        Ok(rescale_count)
    }

    /// Export synaptic state for persistence.
    /// FM-PL-008 fix: atomic write (temp file + rename).
    /// FM-PL-001 NaN firewall: non-finite weights fall back to original.
    /// Replaces: plasticity.py PlasticityEngine.export_state()
    pub fn export_state(&self, graph: &Graph) -> M1ndResult<Vec<SynapticState>> {
        let n = graph.num_nodes() as usize;
        let num_plasticity = graph.edge_plasticity.original_weight.len();
        let num_csr = graph.csr.num_edges();

        // Build reverse map: NodeId -> external_id string
        let mut node_ext_id = vec![String::new(); n];
        for (&interned, &node_id) in &graph.id_to_node {
            if let Some(s) = graph.strings.try_resolve(interned) {
                if node_id.as_usize() < n {
                    node_ext_id[node_id.as_usize()] = s.to_string();
                }
            }
        }

        // Build edge_idx -> source NodeId from CSR offsets
        let mut edge_source = vec![0u32; num_csr];
        #[allow(clippy::needless_range_loop)]
        for i in 0..n {
            let lo = graph.csr.offsets[i] as usize;
            let hi = graph.csr.offsets[i + 1] as usize;
            for j in lo..hi {
                edge_source[j] = i as u32;
            }
        }

        let cap = num_plasticity.min(num_csr);
        let mut states = Vec::with_capacity(cap);

        #[allow(clippy::needless_range_loop)]
        for j in 0..cap {
            let original = graph.edge_plasticity.original_weight[j].get();
            let mut current = graph.edge_plasticity.current_weight[j].get();

            // FM-PL-001 NaN firewall
            if !current.is_finite() {
                current = original;
            }

            // Real labels from CSR topology
            let src_idx = edge_source[j] as usize;
            let tgt_idx = graph.csr.targets[j].as_usize();
            let source_label = if src_idx < n {
                node_ext_id[src_idx].clone()
            } else {
                format!("node_{}", src_idx)
            };
            let target_label = if tgt_idx < n {
                node_ext_id[tgt_idx].clone()
            } else {
                format!("node_{}", tgt_idx)
            };
            let relation = graph
                .strings
                .try_resolve(graph.csr.relations[j])
                .unwrap_or("edge")
                .to_string();

            states.push(SynapticState {
                source_label,
                target_label,
                relation,
                original_weight: original,
                current_weight: current,
                strengthen_count: graph.edge_plasticity.strengthen_count[j],
                weaken_count: graph.edge_plasticity.weaken_count[j],
                ltp_applied: graph.edge_plasticity.ltp_applied[j],
                ltd_applied: graph.edge_plasticity.ltd_applied[j],
            });
        }

        Ok(states)
    }

    /// Import synaptic state from persistence.
    /// FM-PL-007 fix: validates JSON schema, wraps in try/catch.
    /// FM-PL-009 fix: validates relation match for edge identity via label-triple matching.
    /// Replaces: plasticity.py PlasticityEngine.import_state()
    pub fn import_state(&mut self, graph: &mut Graph, states: &[SynapticState]) -> M1ndResult<u32> {
        let n = graph.num_nodes() as usize;
        let num_csr = graph.csr.num_edges();
        let num_plasticity = graph.edge_plasticity.original_weight.len();

        // Build reverse map: NodeId -> external_id
        let mut node_ext_id = vec![String::new(); n];
        for (&interned, &node_id) in &graph.id_to_node {
            if let Some(s) = graph.strings.try_resolve(interned) {
                if node_id.as_usize() < n {
                    node_ext_id[node_id.as_usize()] = s.to_string();
                }
            }
        }

        // Build edge_idx -> source from CSR offsets
        let mut edge_source = vec![0u32; num_csr];
        #[allow(clippy::needless_range_loop)]
        for i in 0..n {
            let lo = graph.csr.offsets[i] as usize;
            let hi = graph.csr.offsets[i + 1] as usize;
            for j in lo..hi {
                edge_source[j] = i as u32;
            }
        }

        // Build triple -> CSR edge index lookup
        use std::collections::HashMap;
        let cap = num_plasticity.min(num_csr);
        let mut triple_to_edge: HashMap<(&str, &str, &str), usize> = HashMap::with_capacity(cap);
        #[allow(clippy::needless_range_loop)]
        for j in 0..cap {
            let src_idx = edge_source[j] as usize;
            let tgt_idx = graph.csr.targets[j].as_usize();
            if src_idx < n && tgt_idx < n {
                let rel = graph
                    .strings
                    .try_resolve(graph.csr.relations[j])
                    .unwrap_or("");
                triple_to_edge.insert((&node_ext_id[src_idx], &node_ext_id[tgt_idx], rel), j);
            }
        }

        let mut applied = 0u32;

        for state in states {
            // FM-PL-009: match by (source, target, relation) triple
            let rel_str = state.relation.as_str();
            let j = match triple_to_edge.get(&(
                state.source_label.as_str(),
                state.target_label.as_str(),
                rel_str,
            )) {
                Some(&idx) => idx,
                None => continue, // Edge no longer exists in graph
            };

            // Validate weight is finite (FM-PL-001)
            let weight = if state.current_weight.is_finite() {
                state.current_weight
            } else {
                state.original_weight
            };

            // Clamp to valid range
            let clamped = weight
                .max(self.config.weight_floor.get())
                .min(self.config.weight_cap.get());

            graph.edge_plasticity.current_weight[j] = FiniteF32::new(clamped);
            graph.edge_plasticity.strengthen_count[j] = state.strengthen_count;
            graph.edge_plasticity.weaken_count[j] = state.weaken_count;
            graph.edge_plasticity.ltp_applied[j] = state.ltp_applied;
            graph.edge_plasticity.ltd_applied[j] = state.ltd_applied;

            // Update CSR weight
            let edge_idx = EdgeIdx::new(j as u32);
            if j < graph.csr.weights.len() {
                let _ = graph.csr.atomic_write_weight(
                    edge_idx,
                    FiniteF32::new(clamped),
                    self.config.cas_retry_limit,
                );
            }

            applied += 1;
        }

        Ok(applied)
    }

    /// Get priming signal from query memory.
    pub fn get_priming(
        &self,
        seeds: &[NodeId],
        boost_strength: FiniteF32,
    ) -> Vec<(NodeId, FiniteF32)> {
        self.memory.get_priming_signal(seeds, boost_strength)
    }
}

static_assertions::assert_impl_all!(PlasticityEngine: Send, Sync);