Reuse prepared glossary data across stage sections

internal/framework/validators/llm_context.go

@@ -1,15 +1,43 @@
 package validators
 
 import (
+	"slices"
+
 	"gitea.maximumdirect.net/eric/audita/v3/internal/core/chunking"
 	"gitea.maximumdirect.net/eric/audita/v3/internal/core/glossary"
 	"gitea.maximumdirect.net/eric/audita/v3/internal/core/schema"
 )
 
+// GlossaryView holds the read-only matching data shared by a stage's validators
+// and apply-time checks.
+type GlossaryView struct {
+	Protection ProtectedVocabulary
+	context    glossaryContext
+}
+
+func PrepareGlossaryView(entries []glossary.Entry, protected []glossary.Term) *GlossaryView {
+	return &GlossaryView{Protection: NewProtectedVocabulary(protected), context: prepareGlossaryContext(entries)}
+}
+
+type glossaryContext struct {
+	entries    []glossary.Entry
+	positions  map[glossary.Identity][]int
+	vocabulary ProtectedVocabulary
+}
+
+func prepareGlossaryContext(entries []glossary.Entry) glossaryContext {
+	positions := make(map[glossary.Identity][]int, len(entries))
+	for i, entry := range entries {
+		identity := glossary.Identity{Collection: entry.Collection, ID: entry.ID}
+		positions[identity] = append(positions[identity], i)
+	}
+	return glossaryContext{entries: entries, positions: positions, vocabulary: NewProtectedVocabulary(glossary.Vocabulary(entries, nil))}
+}
+
 // validationContext supplies adjacent source segments within the current chunk
 // and glossary definitions mentioned in the candidate or that local context.
 // These fields travel in each item so batching accounts for their size.
-func validationContext(segments []schema.Segment, position int, selected []glossary.Entry, vocabulary ProtectedVocabulary, corrected string) ([]NeighboringSegment, []glossary.Entry) {
+func (view glossaryContext) validationContext(segments []schema.Segment, position int, corrected string) ([]NeighboringSegment, []glossary.Entry) {
 	var neighbors []NeighboringSegment
 	source := segments[position].Text
 	if position > 0 {
@@ -23,19 +51,20 @@ func validationContext(segments []schema.Segment, position int, selected []gloss
 	for _, neighbor := range neighbors {
 		texts = append(texts, neighbor.Text)
 	}
+	mentioned := make(map[glossary.Identity]struct{})
+	for _, text := range texts {
+		for _, occurrence := range view.vocabulary.occurrences(text) {
+			mentioned[occurrence.identity] = struct{}{}
+		}
+	}
+	var matched []int
+	for identity := range mentioned {
+		matched = append(matched, view.positions[identity]...)
+	}
+	slices.Sort(matched)
 	var entries []glossary.Entry
-	if selected != nil {
-		mentioned := make(map[glossary.Identity]bool)
-		for _, text := range texts {
-			for identity := range vocabulary.occurrencesByIdentity(text) {
-				mentioned[identity] = true
-			}
-		}
-		for _, entry := range selected {
-			if mentioned[glossary.Identity{Collection: entry.Collection, ID: entry.ID}] {
-				entries = append(entries, entry)
-			}
-		}
+	for _, index := range matched {
+		entries = append(entries, view.entries[index])
 	}
 	return neighbors, entries
 }