Keep match boundaries for exempt glossary names

internal/framework/validators/deterministic.go

@@ -139,17 +139,24 @@ func (v NoEffectValidator) Validate(_ context.Context, req Request) (Result, err
 func ValidateProtectedTerms(req Request) (Result, error) {
 	vocab := NewProtectedVocabulary(req.ProtectedTerms)
 	decisions := make([]Decision, 0, len(req.CandidateProposal))
+	glossaryStage := req.ModuleKey == modulecatalog.KeyGlossary
 	for _, c := range req.CandidateProposal {
 		reason := vocab.violationReason(c.OriginalText, c.CorrectedText)
-		if reason != "" && req.ModuleKey == modulecatalog.KeySpokenWord && req.WorkingTranscript != nil {
+		repetitionCleanup := reason != "" && req.ModuleKey == modulecatalog.KeySpokenWord
+		if vocab.pattern != nil && (glossaryStage || repetitionCleanup) && req.WorkingTranscript != nil {
 			for i := range req.WorkingTranscript.Segments {
 				segment := &req.WorkingTranscript.Segments[i]
 				if segment.ID != c.TargetSegmentID {
 					continue
 				}
 				preview := proposals.PreviewProposalForSegment(segment, c.CorrectionProposal, req.ReplacementPolicy)
-				if preview.Applicable && vocab.preservesRepeatedIdentities(segment.Text, preview.CorrectedSegmentText) {
-					reason = ""
+				if preview.Applicable {
+					if glossaryStage {
+						// A correction span may sit inside a longer exempt name.
+						reason = vocab.violationReason(segment.Text, preview.CorrectedSegmentText)
+					} else if vocab.preservesRepeatedIdentities(segment.Text, preview.CorrectedSegmentText) {
+						reason = ""
+					}
 				}
 				break
 			}

internal/framework/validators/protected_terms.go

@@ -30,13 +30,15 @@ func NewProtectedVocabulary(terms []glossary.Term) ProtectedVocabulary {
 	termsByFolded := make(map[string]protectedTermDef, len(terms))
 	alternatives := make([]string, 0, len(terms))
 	for _, term := range terms {
-		termsByFolded[strings.ToLower(term.Spelling)] = protectedTermDef{identities: term.Identities, canonical: term.Spelling}
+		if !term.Exempt {
+			termsByFolded[strings.ToLower(term.Spelling)] = protectedTermDef{identities: term.Identities, canonical: term.Spelling}
+		}
 		alternatives = append(alternatives, regexp.QuoteMeta(term.Spelling))
 	}
-	sort.SliceStable(alternatives, func(i, j int) bool { return len(alternatives[i]) > len(alternatives[j]) })
-	if len(alternatives) == 0 {
-		return ProtectedVocabulary{termsByFolded: termsByFolded}
+	if len(termsByFolded) == 0 {
+		return ProtectedVocabulary{}
 	}
+	sort.SliceStable(alternatives, func(i, j int) bool { return len(alternatives[i]) > len(alternatives[j]) })
 	pattern := regexp.MustCompile(`(?i)\b(?:` + strings.Join(alternatives, "|") + `)\b`)
 	return ProtectedVocabulary{termsByFolded: termsByFolded, pattern: pattern}
 }

internal/framework/validators/vocabulary_test.go

@@ -61,6 +61,53 @@ func TestApplicationChecksCurrentTextWithSelectedProtection(t *testing.T) {
 	}
 }
 
+func TestExemptGlossarySpellingKeepsLongestMatchBoundary(t *testing.T) {
+	active := []glossary.Entry{
+		{ID: "greater-mage", Name: "Greater Mage Armor"},
+		{ID: "greater-magus", Name: "Greater Magus Armor"},
+	}
+	protected := []glossary.Entry{
+		active[0], active[1],
+		{ID: "short", Name: "Mage Armor"},
+		{ID: "long", Name: "Superior Greater Mage Armor"},
+	}
+	for _, tc := range []struct {
+		name, module, before, original, replacement string
+		exempt, allowed                             bool
+	}{
+		{"full active name", "glossary", "Cast Greater Mage Armor.", "Greater Mage Armor", "Greater Magus Armor", true, true},
+		{"span within active name", "glossary", "Cast Greater Mage Armor.", "Mage Armor", "Magus Armor", true, true},
+		{"standalone protected name", "glossary", "Cast Mage Armor.", "Mage Armor", "Magus Armor", true, false},
+		{"longer protected name", "glossary", "Cast Superior Greater Mage Armor.", "Greater Mage Armor", "Greater Magus Armor", true, false},
+		{"same name in a grammar pass", "grammar", "Cast Greater Mage Armor.", "Greater Mage Armor", "Greater Magus Armor", false, false},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			var exclusions []glossary.Entry
+			if tc.exempt {
+				exclusions = active
+			}
+			terms := glossary.Vocabulary(protected, exclusions)
+			transcript := &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: tc.before}}}
+			candidate := proposals.EnrichedCorrectionProposal{CorrectionProposal: proposals.CorrectionProposal{
+				TargetSegmentID: 1, OriginalText: tc.original, CorrectedText: tc.replacement, Confidence: 1,
+			}}
+			policy := proposals.ReplacementPolicyRequireUnique
+			decision, err := ValidateProtectedTerms(Request{ModuleKey: tc.module, WorkingTranscript: transcript, ReplacementPolicy: policy,
+				ProtectedTerms: terms, CandidateProposal: []proposals.EnrichedCorrectionProposal{candidate}})
+			if err != nil || len(decision.Decisions) != 1 || decision.Decisions[0].Approved != tc.allowed {
+				t.Fatalf("decision=%+v err=%v", decision, err)
+			}
+			vocabulary := NewProtectedVocabulary(terms)
+			result := proposals.ApplyProposals(transcript, []proposals.EnrichedCorrectionProposal{candidate}, policy, func(before, after string) bool {
+				return vocabulary.AllowsChange(before, after, false)
+			})
+			if (len(result.Applied) == 1) != tc.allowed {
+				t.Fatalf("apply-time protection disagrees with validation: %+v", result)
+			}
+		})
+	}
+}
+
 func TestValidationContextIncludesEachSharedDefinition(t *testing.T) {
 	entries := []glossary.Entry{
 		{Collection: "spells", ID: "shield", Name: "Shield", Category: "spell"},