terraphim_rlm 1.21.0

Recursive Language Model (RLM) orchestration for Terraphim AI
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
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
//! Knowledge Graph Validator for RLM command validation.
//!
//! This module provides validation of commands and text against a knowledge graph
//! using term matching and path connectivity analysis. It supports configurable
//! strictness levels and retry logic with LLM rephrasing.
//!
//! ## Architecture
//!
//! ```text
//! KnowledgeGraphValidator
//!     ├── TermMatcher (Aho-Corasick via terraphim_automata)
//!     ├── PathValidator (DFS connectivity via terraphim_rolegraph)
//!     └── RetryPolicy (LLM rephrasing on failure)
//! ```
//!
//! ## Strictness Levels
//!
//! - **Permissive**: Warns but allows execution (log only)
//! - **Normal**: Requires at least one known term (default)
//! - **Strict**: Requires all terms to be connected by a graph path

use serde::{Deserialize, Serialize};
use std::collections::HashSet;
use terraphim_rolegraph::RoleGraph;
use terraphim_types::Thesaurus;

use crate::config::KgStrictness;
use crate::error::RlmError;
use crate::types::SessionId;

/// Result type for this module.
pub type RlmResult<T> = Result<T, RlmError>;

/// Configuration for the knowledge graph validator.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ValidatorConfig {
    /// Strictness level for validation.
    pub strictness: KgStrictness,
    /// Maximum retries before escalation.
    pub max_retries: u32,
    /// Minimum match ratio for normal mode (0.0 to 1.0).
    pub min_match_ratio: f32,
    /// Whether to require path connectivity.
    pub require_connectivity: bool,
}

impl Default for ValidatorConfig {
    fn default() -> Self {
        Self {
            strictness: KgStrictness::Normal,
            max_retries: 3,
            min_match_ratio: 0.1, // At least 10% of words should match
            require_connectivity: false,
        }
    }
}

impl ValidatorConfig {
    /// Create a permissive configuration (warn only).
    pub fn permissive() -> Self {
        Self {
            strictness: KgStrictness::Permissive,
            max_retries: 0,
            min_match_ratio: 0.0,
            require_connectivity: false,
        }
    }

    /// Create a strict configuration (require connectivity).
    pub fn strict() -> Self {
        Self {
            strictness: KgStrictness::Strict,
            max_retries: 3,
            min_match_ratio: 0.3,
            require_connectivity: true,
        }
    }
}

/// Result of command validation.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ValidationResult {
    /// Whether the command passed validation.
    pub passed: bool,
    /// Matched terms from the knowledge graph.
    pub matched_terms: Vec<String>,
    /// Unmatched words from the command.
    pub unmatched_words: Vec<String>,
    /// Whether matched terms are connected by a graph path.
    pub terms_connected: bool,
    /// Match ratio (matched terms / total words).
    pub match_ratio: f32,
    /// Validation message explaining the result.
    pub message: String,
    /// Suggested rephrasings (if validation failed).
    pub suggestions: Vec<String>,
    /// Number of retries attempted.
    pub retry_count: u32,
    /// Whether escalation to user is required.
    pub escalation_required: bool,
}

impl ValidationResult {
    /// Create a passed validation result.
    pub fn passed(
        matched_terms: Vec<String>,
        unmatched_words: Vec<String>,
        terms_connected: bool,
        match_ratio: f32,
    ) -> Self {
        Self {
            passed: true,
            matched_terms,
            unmatched_words,
            terms_connected,
            match_ratio,
            message: "Validation passed".to_string(),
            suggestions: Vec::new(),
            retry_count: 0,
            escalation_required: false,
        }
    }

    /// Create a failed validation result.
    pub fn failed(
        matched_terms: Vec<String>,
        unmatched_words: Vec<String>,
        terms_connected: bool,
        match_ratio: f32,
        message: String,
    ) -> Self {
        Self {
            passed: false,
            matched_terms,
            unmatched_words,
            terms_connected,
            match_ratio,
            message,
            suggestions: Vec::new(),
            retry_count: 0,
            escalation_required: false,
        }
    }

    /// Mark as requiring escalation to user.
    pub fn with_escalation(mut self) -> Self {
        self.escalation_required = true;
        self
    }

    /// Add suggestions for rephrasing.
    pub fn with_suggestions(mut self, suggestions: Vec<String>) -> Self {
        self.suggestions = suggestions;
        self
    }

    /// Set retry count.
    pub fn with_retry_count(mut self, count: u32) -> Self {
        self.retry_count = count;
        self
    }
}

/// Validation context for tracking retry state.
#[derive(Debug, Clone)]
pub struct ValidationContext {
    /// Session ID for this validation.
    pub session_id: SessionId,
    /// Number of retries attempted.
    pub retry_count: u32,
    /// Previous validation results.
    pub history: Vec<ValidationResult>,
}

impl ValidationContext {
    /// Create a new validation context.
    pub fn new(session_id: SessionId) -> Self {
        Self {
            session_id,
            retry_count: 0,
            history: Vec::new(),
        }
    }

    /// Increment retry count and record result.
    pub fn record_attempt(&mut self, result: ValidationResult) {
        self.retry_count += 1;
        self.history.push(result);
    }

    /// Check if max retries exceeded.
    pub fn max_retries_exceeded(&self, max_retries: u32) -> bool {
        self.retry_count >= max_retries
    }
}

/// Knowledge graph validator for command validation.
pub struct KnowledgeGraphValidator {
    config: ValidatorConfig,
    thesaurus: Option<Thesaurus>,
    role_graph: Option<RoleGraph>,
}

impl KnowledgeGraphValidator {
    /// Create a new validator with the given configuration.
    pub fn new(config: ValidatorConfig) -> Self {
        Self {
            config,
            thesaurus: None,
            role_graph: None,
        }
    }

    /// Create a disabled validator that always passes.
    pub fn disabled() -> Self {
        Self {
            config: ValidatorConfig::permissive(),
            thesaurus: None,
            role_graph: None,
        }
    }

    /// Set the thesaurus for term matching.
    pub fn with_thesaurus(mut self, thesaurus: Thesaurus) -> Self {
        self.thesaurus = Some(thesaurus);
        self
    }

    /// Set the role graph for path connectivity.
    pub fn with_role_graph(mut self, role_graph: RoleGraph) -> Self {
        self.role_graph = Some(role_graph);
        self
    }

    /// Validate a command string.
    ///
    /// Returns a validation result indicating whether the command passes
    /// the configured validation rules.
    pub fn validate(&self, command: &str) -> RlmResult<ValidationResult> {
        // Skip validation in permissive mode with no thesaurus
        if self.config.strictness == KgStrictness::Permissive && self.thesaurus.is_none() {
            return Ok(ValidationResult::passed(vec![], vec![], true, 0.0));
        }

        // Extract words from command
        let words = extract_words(command);
        if words.is_empty() {
            return Ok(ValidationResult::passed(vec![], vec![], true, 0.0));
        }

        // Find matched terms
        let (matched_terms, unmatched_words) = self.find_matches(command, &words)?;

        // Calculate match ratio
        let match_ratio = if words.is_empty() {
            0.0
        } else {
            matched_terms.len() as f32 / words.len() as f32
        };

        // Check path connectivity
        let terms_connected = self.check_connectivity(command);

        // Apply validation rules based on strictness
        match self.config.strictness {
            KgStrictness::Permissive => {
                // Always pass, but log if there are issues
                if matched_terms.is_empty() {
                    log::warn!(
                        "KG validation (permissive): No matched terms in command: {}",
                        truncate_for_log(command)
                    );
                }
                Ok(ValidationResult::passed(
                    matched_terms,
                    unmatched_words,
                    terms_connected,
                    match_ratio,
                ))
            }
            KgStrictness::Normal => {
                // Require at least one matched term or minimum ratio
                if matched_terms.is_empty() && self.thesaurus.is_some() {
                    let msg = format!(
                        "No known terms found. Please use domain-specific terminology. Unrecognized: {:?}",
                        &unmatched_words[..unmatched_words.len().min(5)]
                    );
                    Ok(ValidationResult::failed(
                        matched_terms,
                        unmatched_words,
                        terms_connected,
                        match_ratio,
                        msg,
                    ))
                } else if match_ratio < self.config.min_match_ratio && self.thesaurus.is_some() {
                    let msg = format!(
                        "Match ratio {:.1}% below threshold {:.1}%. Consider using more specific terms.",
                        match_ratio * 100.0,
                        self.config.min_match_ratio * 100.0
                    );
                    Ok(ValidationResult::failed(
                        matched_terms,
                        unmatched_words,
                        terms_connected,
                        match_ratio,
                        msg,
                    ))
                } else {
                    Ok(ValidationResult::passed(
                        matched_terms,
                        unmatched_words,
                        terms_connected,
                        match_ratio,
                    ))
                }
            }
            KgStrictness::Strict => {
                // Require connectivity between all matched terms
                if matched_terms.is_empty() && self.thesaurus.is_some() {
                    let msg = "No known terms found. Strict mode requires domain terminology."
                        .to_string();
                    Ok(ValidationResult::failed(
                        matched_terms,
                        unmatched_words,
                        terms_connected,
                        match_ratio,
                        msg,
                    ))
                } else if self.config.require_connectivity
                    && !terms_connected
                    && matched_terms.len() > 1
                {
                    let msg = format!(
                        "Terms {:?} are not connected in the knowledge graph. Please rephrase for semantic coherence.",
                        &matched_terms[..matched_terms.len().min(5)]
                    );
                    Ok(ValidationResult::failed(
                        matched_terms,
                        unmatched_words,
                        terms_connected,
                        match_ratio,
                        msg,
                    ))
                } else {
                    Ok(ValidationResult::passed(
                        matched_terms,
                        unmatched_words,
                        terms_connected,
                        match_ratio,
                    ))
                }
            }
        }
    }

    /// Validate with retry context.
    ///
    /// Tracks retry attempts and escalates to user after max retries.
    pub fn validate_with_context(
        &self,
        command: &str,
        context: &mut ValidationContext,
    ) -> RlmResult<ValidationResult> {
        let mut result = self.validate(command)?;

        if !result.passed {
            // Check if we need to escalate
            if context.max_retries_exceeded(self.config.max_retries) {
                result = result.with_escalation();
                result.message = format!(
                    "Validation failed after {} attempts. User intervention required: {}",
                    context.retry_count, result.message
                );
            } else {
                // Generate suggestions for rephrasing
                let suggestions = self.generate_suggestions(command, &result);
                result = result.with_suggestions(suggestions);
            }

            result = result.with_retry_count(context.retry_count);
            context.record_attempt(result.clone());
        }

        Ok(result)
    }

    /// Find matched and unmatched terms in the command.
    fn find_matches(
        &self,
        command: &str,
        words: &[String],
    ) -> RlmResult<(Vec<String>, Vec<String>)> {
        let Some(ref thesaurus) = self.thesaurus else {
            // No thesaurus, return empty matches
            return Ok((vec![], words.to_vec()));
        };

        // Use terraphim_automata for term matching
        let matches =
            terraphim_automata::find_matches(command, thesaurus.clone(), true).map_err(|e| {
                RlmError::ConfigError {
                    message: format!("Term matching failed: {}", e),
                }
            })?;

        let matched_terms: Vec<String> = matches.iter().map(|m| m.term.clone()).collect();
        let matched_set: HashSet<_> = matched_terms.iter().map(|s| s.to_lowercase()).collect();

        // Find unmatched words
        let unmatched_words: Vec<String> = words
            .iter()
            .filter(|w| !matched_set.contains(&w.to_lowercase()))
            .cloned()
            .collect();

        Ok((matched_terms, unmatched_words))
    }

    /// Check if all matched terms are connected by a graph path.
    fn check_connectivity(&self, text: &str) -> bool {
        if let Some(ref role_graph) = self.role_graph {
            role_graph.is_all_terms_connected_by_path(text)
        } else {
            // No role graph, assume connected
            true
        }
    }

    /// Generate suggestions for rephrasing a failed command.
    fn generate_suggestions(&self, command: &str, result: &ValidationResult) -> Vec<String> {
        let mut suggestions = Vec::new();

        // Suggest using matched terms more prominently
        if !result.matched_terms.is_empty() {
            suggestions.push(format!(
                "Try rephrasing using these known terms: {}",
                result.matched_terms.join(", ")
            ));
        }

        // Suggest being more specific
        if result.unmatched_words.len() > 3 {
            suggestions.push(
                "Consider using more domain-specific terminology instead of generic terms."
                    .to_string(),
            );
        }

        // Suggest breaking down the command
        if command.len() > 100 {
            suggestions
                .push("Consider breaking this into smaller, more focused commands.".to_string());
        }

        suggestions
    }

    /// Get the current configuration.
    pub fn config(&self) -> &ValidatorConfig {
        &self.config
    }

    /// Check if the validator has a thesaurus.
    pub fn has_thesaurus(&self) -> bool {
        self.thesaurus.is_some()
    }

    /// Check if the validator has a role graph.
    pub fn has_role_graph(&self) -> bool {
        self.role_graph.is_some()
    }

    /// Return the strictness level this validator is configured with.
    pub fn strictness(&self) -> KgStrictness {
        self.config.strictness
    }

    /// Build a validator from RLM config fields.
    ///
    /// Eliminates the `build_validator()` duplication between rlm.rs and
    /// executor/mod.rs (GAP-2).  Optionally loads a thesaurus from disk
    /// so Normal/Strict validation performs real term matching (GAP-1).
    pub fn from_config(
        strictness: KgStrictness,
        max_retries: u32,
        thesaurus_path: Option<&str>,
    ) -> Self {
        let mut vcfg = match strictness {
            KgStrictness::Permissive => ValidatorConfig::permissive(),
            KgStrictness::Normal => ValidatorConfig::default(),
            KgStrictness::Strict => ValidatorConfig::strict(),
        };
        vcfg.max_retries = max_retries;
        let mut validator = Self::new(vcfg);

        match thesaurus_path {
            Some(path) => match std::fs::read_to_string(path) {
                Ok(json) => match serde_json::from_str::<Thesaurus>(&json) {
                    Ok(thesaurus) => {
                        log::info!(
                            "KG validator: loaded {} terms from {}",
                            thesaurus.len(),
                            path
                        );
                        validator = validator.with_thesaurus(thesaurus);
                    }
                    Err(e) => log::warn!(
                        "KG validator: could not parse thesaurus at '{}': {}. \
                         Real term matching is disabled.",
                        path,
                        e
                    ),
                },
                Err(e) => log::warn!(
                    "KG validator: could not read thesaurus at '{}': {}. \
                     Real term matching is disabled.",
                    path,
                    e
                ),
            },
            None if strictness != KgStrictness::Permissive => {
                log::warn!(
                    "KG validator: no thesaurus configured (thesaurus_path is None). \
                     {:?} mode will not perform real term matching. \
                     Set thesaurus_path to a JSON thesaurus file to enable.",
                    strictness
                );
            }
            None => {}
        }

        validator
    }
}

/// Extract words from a command string.
fn extract_words(text: &str) -> Vec<String> {
    // Split on any non-word character (not alphanumeric, underscore, or hyphen)
    text.split(|c: char| !c.is_alphanumeric() && c != '_' && c != '-')
        .map(|s| s.to_string())
        .filter(|s| !s.is_empty() && s.len() > 2) // Skip very short words
        .collect()
}

/// Truncate a string for logging (max 100 chars).
fn truncate_for_log(s: &str) -> String {
    if s.len() > 100 {
        let boundary = {
            let mut i = 97_usize.min(s.len());
            while i > 0 && !s.is_char_boundary(i) {
                i -= 1;
            }
            i
        };
        format!("{}...", &s[..boundary])
    } else {
        s.to_string()
    }
}

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

    #[test]
    fn test_validator_config_default() {
        let config = ValidatorConfig::default();
        assert_eq!(config.strictness, KgStrictness::Normal);
        assert_eq!(config.max_retries, 3);
    }

    #[test]
    fn test_validator_config_permissive() {
        let config = ValidatorConfig::permissive();
        assert_eq!(config.strictness, KgStrictness::Permissive);
        assert_eq!(config.max_retries, 0);
    }

    #[test]
    fn test_validator_config_strict() {
        let config = ValidatorConfig::strict();
        assert_eq!(config.strictness, KgStrictness::Strict);
        assert!(config.require_connectivity);
    }

    #[test]
    fn test_validation_result_passed() {
        let result = ValidationResult::passed(
            vec!["term1".to_string()],
            vec!["unknown".to_string()],
            true,
            0.5,
        );
        assert!(result.passed);
        assert!(!result.escalation_required);
    }

    #[test]
    fn test_validation_result_failed() {
        let result = ValidationResult::failed(
            vec![],
            vec!["unknown".to_string()],
            false,
            0.0,
            "No matches".to_string(),
        );
        assert!(!result.passed);
    }

    #[test]
    fn test_validation_result_with_escalation() {
        let result = ValidationResult::failed(
            vec![],
            vec!["unknown".to_string()],
            false,
            0.0,
            "Failed".to_string(),
        )
        .with_escalation();
        assert!(result.escalation_required);
    }

    #[test]
    fn test_validation_context() {
        let session_id = SessionId::new();
        let mut context = ValidationContext::new(session_id);

        assert_eq!(context.retry_count, 0);
        assert!(!context.max_retries_exceeded(3));

        context.record_attempt(ValidationResult::failed(
            vec![],
            vec![],
            false,
            0.0,
            "Failed".to_string(),
        ));
        assert_eq!(context.retry_count, 1);
        assert_eq!(context.history.len(), 1);
    }

    #[test]
    fn test_disabled_validator() {
        let validator = KnowledgeGraphValidator::disabled();
        let result = validator.validate("any command here").unwrap();
        assert!(result.passed);
    }

    #[test]
    fn test_validator_empty_command() {
        let validator = KnowledgeGraphValidator::new(ValidatorConfig::default());
        let result = validator.validate("").unwrap();
        assert!(result.passed);
    }

    #[test]
    fn test_validator_no_thesaurus_permissive() {
        let validator = KnowledgeGraphValidator::new(ValidatorConfig::permissive());
        let result = validator.validate("print hello world").unwrap();
        assert!(result.passed);
    }

    #[test]
    fn test_validator_no_thesaurus_normal() {
        let validator = KnowledgeGraphValidator::new(ValidatorConfig::default());
        // Without a thesaurus, normal mode should pass (no thesaurus = no validation)
        let result = validator.validate("print hello world").unwrap();
        assert!(result.passed);
    }

    #[test]
    fn test_extract_words() {
        let words = extract_words("print('hello, world!')");
        assert!(words.contains(&"print".to_string()));
        assert!(words.contains(&"hello".to_string()));
        assert!(words.contains(&"world".to_string()));
    }

    #[test]
    fn test_extract_words_filters_short() {
        let words = extract_words("a b cd this_is_longer");
        // Should filter out "a", "b", "cd" (2 chars or less)
        assert!(!words.contains(&"a".to_string()));
        assert!(!words.contains(&"b".to_string()));
        assert!(!words.contains(&"cd".to_string()));
        assert!(words.contains(&"this_is_longer".to_string()));
    }

    #[test]
    fn test_truncate_for_log() {
        let short = "short string";
        assert_eq!(truncate_for_log(short), short);

        let long = "a".repeat(150);
        let truncated = truncate_for_log(&long);
        assert!(truncated.len() < 150);
        assert!(truncated.ends_with("..."));
    }

    #[test]
    fn test_truncate_for_log_multibyte() {
        // Each emoji is 4 bytes; 26 of them = 104 bytes, triggering truncation.
        // The byte boundary at 97 falls mid-emoji; floor_char_boundary aligns to a valid char.
        let emoji_str = "😀".repeat(26);
        let result = truncate_for_log(&emoji_str);
        assert!(result.ends_with("..."));
        assert!(result.is_char_boundary(result.len() - 3)); // valid UTF-8
    }

    #[test]
    fn test_validation_context_max_retries() {
        let session_id = SessionId::new();
        let mut context = ValidationContext::new(session_id);

        // Add 3 retry attempts
        for _ in 0..3 {
            context.record_attempt(ValidationResult::failed(
                vec![],
                vec![],
                false,
                0.0,
                "Failed".to_string(),
            ));
        }

        assert!(context.max_retries_exceeded(3));
        assert!(!context.max_retries_exceeded(4));
    }

    #[test]
    fn test_generate_suggestions() {
        let validator = KnowledgeGraphValidator::new(ValidatorConfig::default());
        let result = ValidationResult::failed(
            vec!["term1".to_string(), "term2".to_string()],
            vec![
                "unknown1".to_string(),
                "unknown2".to_string(),
                "unknown3".to_string(),
                "unknown4".to_string(),
            ],
            false,
            0.3,
            "Failed".to_string(),
        );

        let suggestions =
            validator.generate_suggestions("some long command that needs rephrasing", &result);
        assert!(!suggestions.is_empty());
        // Should suggest using known terms
        assert!(
            suggestions
                .iter()
                .any(|s| s.contains("known terms") || s.contains("term1"))
        );
    }

    // --- Thesaurus fixture tests covering all three KgStrictness variants ---
    // These tests use a minimal in-memory Thesaurus to exercise the code paths
    // where a thesaurus is present, verifying the distinction between
    // Permissive (always pass), Normal (fail on zero matches), and Strict
    // (fail on zero matches, require connectivity).

    #[cfg(feature = "kg-validation")]
    #[test]
    fn test_validator_permissive_with_empty_thesaurus_passes() {
        // Permissive mode must always pass even when a thesaurus exists but
        // no terms in the command match any thesaurus entry.
        let thesaurus = Thesaurus::new("test".to_string());
        let validator =
            KnowledgeGraphValidator::new(ValidatorConfig::permissive()).with_thesaurus(thesaurus);
        let result = validator
            .validate("some command with words that do not match")
            .unwrap();
        assert!(
            result.passed,
            "Permissive mode must pass regardless of matches"
        );
    }

    #[cfg(feature = "kg-validation")]
    #[test]
    fn test_validator_normal_with_empty_thesaurus_fails() {
        // Normal mode must fail when a thesaurus is present but zero terms
        // from the command match any thesaurus entry.
        let thesaurus = Thesaurus::new("test".to_string());
        let validator =
            KnowledgeGraphValidator::new(ValidatorConfig::default()).with_thesaurus(thesaurus);
        // All words are unknown — no matches possible from empty thesaurus.
        let result = validator
            .validate("unknown domain words that should not match")
            .unwrap();
        assert!(!result.passed, "Normal mode must fail when no terms match");
        assert!(result.matched_terms.is_empty());
    }

    #[cfg(feature = "kg-validation")]
    #[test]
    fn test_validator_strict_with_empty_thesaurus_fails() {
        // Strict mode must fail when a thesaurus is present but zero terms match.
        let thesaurus = Thesaurus::new("test".to_string());
        let validator =
            KnowledgeGraphValidator::new(ValidatorConfig::strict()).with_thesaurus(thesaurus);
        let result = validator
            .validate("unknown domain words that should not match")
            .unwrap();
        assert!(!result.passed, "Strict mode must fail when no terms match");
        assert!(result.matched_terms.is_empty());
    }

    #[cfg(feature = "kg-validation")]
    #[test]
    fn test_validator_normal_with_matching_thesaurus_passes() {
        // Normal mode passes when matched terms exist.
        // Uses NormalizedTermValue + NormalizedTerm to insert a real entry.
        use terraphim_types::{NormalizedTerm, NormalizedTermValue};
        let mut thesaurus = Thesaurus::new("test".to_string());
        thesaurus.insert(
            NormalizedTermValue::from("rust"),
            NormalizedTerm::with_auto_id(NormalizedTermValue::from("rust")),
        );
        let validator =
            KnowledgeGraphValidator::new(ValidatorConfig::default()).with_thesaurus(thesaurus);
        // "rust" is in the thesaurus — should match and pass Normal mode.
        let result = validator.validate("using rust for this task").unwrap();
        assert!(
            result.passed,
            "Normal mode must pass when at least one term matches"
        );
        assert!(!result.matched_terms.is_empty());
    }

    // -- from_config tests (GAP-2: deduplication; GAP-1: thesaurus loading) --

    #[test]
    fn test_from_config_normal_no_path_has_no_thesaurus() {
        let v = KnowledgeGraphValidator::from_config(KgStrictness::Normal, 3, None);
        assert!(!v.has_thesaurus());
        assert_eq!(v.strictness(), KgStrictness::Normal);
        assert_eq!(v.config().max_retries, 3);
    }

    #[test]
    fn test_from_config_permissive_no_path_passes_silently() {
        let v = KnowledgeGraphValidator::from_config(KgStrictness::Permissive, 0, None);
        assert!(!v.has_thesaurus());
        let result = v.validate("any command here").unwrap();
        assert!(result.passed);
    }

    #[test]
    fn test_from_config_strict_respects_max_retries() {
        let v = KnowledgeGraphValidator::from_config(KgStrictness::Strict, 5, None);
        assert_eq!(v.config().max_retries, 5);
        assert_eq!(v.strictness(), KgStrictness::Strict);
    }

    #[test]
    fn test_from_config_bad_path_falls_back_gracefully() {
        let v = KnowledgeGraphValidator::from_config(
            KgStrictness::Normal,
            3,
            Some("/nonexistent/path/thesaurus.json"),
        );
        assert!(!v.has_thesaurus());
        let result = v.validate("some command").unwrap();
        assert!(result.passed);
    }

    #[test]
    fn test_from_config_valid_thesaurus_json() {
        use std::io::Write;
        let dir = tempfile::tempdir().unwrap();
        let path = dir.path().join("thesaurus.json");
        let thesaurus_json = r#"{"name":"test","data":{}}"#;
        let mut f = std::fs::File::create(&path).unwrap();
        f.write_all(thesaurus_json.as_bytes()).unwrap();
        drop(f);

        let v = KnowledgeGraphValidator::from_config(
            KgStrictness::Normal,
            3,
            Some(path.to_str().unwrap()),
        );
        assert!(v.has_thesaurus());
    }
}