Separate glossary context from prompt protection

internal/framework/proposal_generation/module_proposal.go

@@ -2,6 +2,7 @@ package proposal_generation
 
 import (
 	"context"
+	"encoding/json"
 	"fmt"
 	"strings"
 
@@ -47,6 +48,16 @@ func ExecuteModuleProposal(ctx context.Context, req ModuleProposalRequest) (cont
 		return contracts.ProposalResult{}, err
 	}
 
+	protected := make([]string, 0, len(req.ProposalRequest.ProtectedTerms))
+	for _, term := range req.ProposalRequest.ProtectedTerms {
+		protected = append(protected, term.Spelling)
+	}
+	payload, err := json.Marshal(protected)
+	if err != nil {
+		return contracts.ProposalResult{}, fmt.Errorf("marshal protected spellings: %w", err)
+	}
+	variables["ProtectedTermsJSON"] = string(payload)
+
 	promptMetadata, ok := prompts.LookupMetadata(req.PromptID)
 	if !ok {
 		return contracts.ProposalResult{}, fmt.Errorf("unknown prompt ID %q", req.PromptID)

internal/framework/proposal_generation/module_proposal_test.go

@@ -40,6 +40,7 @@ func TestExecuteModuleProposalBuildsMessagesFromSectionAndDescription(t *testing
 				Settings:          cfg,
 				WorkingTranscript: transcript,
 				Glossary:          glossary,
+				ProtectedTerms:    []domain.Term{{Spelling: "Separate protection"}},
 				Section:           &section,
 			},
 			RunSpec: contracts.ModuleRunSpec{
@@ -76,6 +77,9 @@ func TestExecuteModuleProposalBuildsMessagesFromSectionAndDescription(t *testing
 	if len(client.calls) != 1 || client.calls[0].PromptID != prompts.PromptIDModuleGrammarProposal || client.calls[0].Context.SectionIndex == nil || *client.calls[0].Context.SectionIndex != 7 {
 		t.Fatalf("unexpected stage name calls: %+v", client.calls)
 	}
+	if got := client.calls[0].Variables["ProtectedTermsJSON"]; got != `["Separate protection"]` {
+		t.Fatalf("protected spellings = %s", got)
+	}
 	if len(out.Proposals) != 1 || out.Proposals[0].CorrectedText != "the" {
 		t.Fatalf("unexpected proposals: %+v", out)
 	}

internal/prompts/assets/prompts/modules.glossary.proposal.yaml

@@ -17,7 +17,7 @@ collections:
               - Inappropriate example: correcting "Lyra" to "Jesters" should be omitted because those words are not similar in spoken English, even if "Jesters" appears in the glossary.
               - Do not replace one clear glossary term, character name, location, or ordinary word with a different glossary term unless it is a plausible mishearing.
               - Preserve correctly identified glossary names and aliases. A recognized glossary term may still be a mishearing of another term; correct that misidentification only when spoken similarity and surrounding context support it.
-              - Apart from that supported mishearing exception, do not replace, Anglicize, normalize, lowercase, or otherwise alter protected glossary names or aliases away from their glossary spelling.
+              - Preserve spellings listed under Protected spellings. Glossary context supplies correction targets and guidance.
               - Preserve canonical glossary capitalization for protected names and aliases, even if they look unusual.
               - If a segment includes categories, treat them as additional transcript context.
               - Plural forms of glossary names and aliases are allowed targets when spoken similarity and context support them, even if the plural is not explicitly listed in the glossary.
@@ -28,9 +28,12 @@ collections:
 
               {{ .HardeningPolicy }}
 
-              {{ .TranscriptDescriptionBlock }}Glossary:
+              {{ .TranscriptDescriptionBlock }}Glossary context:
               {{ .GlossaryJSON }}
 
+              Protected spellings:
+              {{ .ProtectedTermsJSON }}
+
               Transcript section:
               {{ .SectionJSON }}
         output:

internal/prompts/assets/prompts/modules.grammar.proposal.yaml

@@ -18,8 +18,7 @@ collections:
               - Apart from the a/an article exception, do not make word substitutions, spelling fixes, homophone fixes, filler cleanup, repetition cleanup, paraphrases, or other semantic rewrites.
               - Do not change one written word into a different written word, except for capitalization changes to the same letters and the a/an article exception.
               - If a possible correction depends on changing a content word into a different word, omit it here rather than bundling it together with formatting cleanup.
-              - Treat glossary names and aliases as protected spellings and context.
-              - Do not replace, Anglicize, normalize, lowercase, or otherwise alter protected glossary names or aliases away from their glossary spelling.
+              - Do not replace, Anglicize, normalize, lowercase, or otherwise alter spellings listed under Protected spellings away from their glossary spelling.
               - Preserve canonical glossary capitalization for protected names and aliases, even if they look unusual.
               - If a segment includes categories, treat them as additional transcript context.
               - Choose an original_text span that appears exactly once in the current segment text.
@@ -31,9 +30,12 @@ collections:
 
               {{ .HardeningPolicy }}
 
-              {{ .TranscriptDescriptionBlock }}Protected glossary/context:
+              {{ .TranscriptDescriptionBlock }}Glossary context:
               {{ .GlossaryJSON }}
 
+              Protected spellings:
+              {{ .ProtectedTermsJSON }}
+
               Transcript section:
               {{ .SectionJSON }}
         output:

internal/prompts/assets/prompts/modules.homophones.proposal.yaml

@@ -16,9 +16,8 @@ collections:
               - Reject unrelated substitutions like changing "Lyra" to "Jesters".
               - Reject antonyms or reversals such as changing "visible" to "invisible".
               - Do not add or remove punctuation, alter capitalization only, normalize spacing, remove filler words, collapse repetitions, or make general readability edits.
-              - Treat glossary names and aliases as protected spellings and context.
               - You may correct toward glossary names, aliases, or their plural forms when the correction is acoustically plausible and supported by local context.
-              - Do not replace, Anglicize, normalize, lowercase, or otherwise alter protected glossary names or aliases that already appear correctly in the transcript.
+              - Do not replace, Anglicize, normalize, lowercase, or otherwise alter spellings listed under Protected spellings that already appear correctly in the transcript.
               - Preserve canonical glossary capitalization for protected names and aliases, even if they look unusual.
               - If a segment includes categories, treat them as additional transcript context.
               - Choose an original_text span that appears exactly once in the current segment text.
@@ -28,9 +27,12 @@ collections:
 
               {{ .HardeningPolicy }}
 
-              {{ .TranscriptDescriptionBlock }}Protected glossary/context:
+              {{ .TranscriptDescriptionBlock }}Glossary context:
               {{ .GlossaryJSON }}
 
+              Protected spellings:
+              {{ .ProtectedTermsJSON }}
+
               Transcript section:
               {{ .SectionJSON }}
         output:

internal/prompts/assets/prompts/modules.spoken_word.proposal.yaml

@@ -20,8 +20,8 @@ collections:
               - Do not paraphrase, summarize, reorder ideas, replace content with different wording, or make substantive semantic edits.
               - Do not change clear content words just because a different phrasing reads better.
               - Do not convert uncertain statements into certain statements.
-              - Treat glossary names and aliases as protected spellings and context. Accidental repetition of a protected name may be reduced when that identity remains in the segment; preserve substantive references and intentional emphasis.
-              - Do not replace, Anglicize, normalize, lowercase, or otherwise alter protected glossary names or aliases that already appear correctly in the transcript.
+              - Accidental repetition of a protected name may be reduced when that identity remains in the segment; preserve substantive references and intentional emphasis.
+              - Do not replace, Anglicize, normalize, lowercase, or otherwise alter spellings listed under Protected spellings that already appear correctly in the transcript.
               - Preserve canonical glossary capitalization for protected names and aliases, even if they look unusual.
               - If a segment includes categories, treat them as additional transcript context.
               - Choose an original_text span that appears exactly once in the current segment text.
@@ -33,9 +33,12 @@ collections:
 
               {{ .HardeningPolicy }}
 
-              {{ .TranscriptDescriptionBlock }}Protected glossary/context:
+              {{ .TranscriptDescriptionBlock }}Glossary context:
               {{ .GlossaryJSON }}
 
+              Protected spellings:
+              {{ .ProtectedTermsJSON }}
+
               Transcript section:
               {{ .SectionJSON }}
         output:

internal/prompts/registry.go

@@ -93,6 +93,9 @@ func WithSharedVariables(values map[string]string) map[string]string {
 			result[key] = value
 		}
 	}
+	if _, exists := result["ProtectedTermsJSON"]; !exists {
+		result["ProtectedTermsJSON"] = "[]"
+	}
 	return result
 }
 

internal/prompts/registry_test.go

@@ -38,9 +38,7 @@ func TestEmbeddedDefinitionsPrepareThroughPromptKit(t *testing.T) {
 		"SectionJSON":                "{\"text\":\"fixture-section\"}",
 		"PayloadJSON":                "[{\"text\":\"fixture-correction\"}]",
 	}
-	for key, value := range SharedVariables() {
-		vars[key] = value
-	}
+	vars = WithSharedVariables(vars)
 	for _, tc := range []struct{ id, instruction, schema, payload string }{
 		{PromptIDModuleGlossaryProposal, "strongly supported by the glossary", "correction_set.v1.json", "fixture-section"},
 		{PromptIDModuleHomophonesProposal, "homophone correction assistant", "correction_set.v1.json", "fixture-section"},