feat(ai-sdk): vocab->tag proposer (P2 slice 5, type 3)

Extends Method C: for each unknown narrative token that pattern text names, suggest
the keyword_dictionary tag = the RequiredComponentTags shared by the naming
patterns (ranked by frequency, kept only when shared by >=40% of them, top 3).
Surfaces real dictionary gaps like "zwischenkreis" -> stored_energy and
"updates" -> has_software, which close coverage without hand-editing the dict.

Two precision fixes to Method C while here:
- patternsMentioning now matches WHOLE WORDS, not substrings — substring matching
  flagged fragments like "stehen" inside "entstehen" and produced nonsensical
  tag suggestions.
- a token is only proposed with a tag if one is shared by >=40% of its naming
  patterns, so diffuse common verbs (spread across categories) drop out.

Wired into iace-audit propose -> audit-reports/vocab.{md,json}. Residual
common-verb noise is left to the human/LLM filter rather than a hand-grown
stopword list. Type 4 (coverage blind spots) + P3 (pin accepted proposals into a
GT case) remain for slice 6.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

ai-compliance-sdk/cmd/iace-audit/propose.go

@@ -6,8 +6,10 @@ import (
 	"fmt"
 	"os"
 	"strconv"
+	"strings"
 
 	"github.com/breakpilot/ai-compliance-sdk/internal/iace"
+	"github.com/breakpilot/ai-compliance-sdk/internal/iace/audit"
 	"github.com/breakpilot/ai-compliance-sdk/internal/llm"
 )
 
@@ -89,12 +91,25 @@ func cmdPropose(args []string) {
 	writeText("audit-reports/framing.md", iace.RenderFramingQueue(in.MachineType, framing))
 	writeJSON("audit-reports/framing.json", framing)
 
+	// Type 3: vocab->tag proposals (unknown narrative tokens that pattern text
+	// names as a whole word, with a dominant shared required tag).
+	vocab := audit.RunVocabulary(map[string]any{"narrative": in.Narrative})
+	var vgaps []audit.DictionarySuggestion
+	for _, s := range vocab.SuggestedDictionaryEntries {
+		if len(s.SuggestedTags) > 0 {
+			vgaps = append(vgaps, s)
+		}
+	}
+	writeText("audit-reports/vocab.md", renderVocabQueue(in.MachineType, vgaps))
+	writeJSON("audit-reports/vocab.json", vgaps)
+
 	printSummary("Method P — Dedup Proposer ("+judge.Name()+")", map[string]int{
 		"fired_patterns": len(fired),
 		"candidates":     len(candidates),
 		"in_queue":       len(proposals),
 		"gt_blocked":     blocked,
 		"framing_flags":  len(framing),
+		"vocab_gaps":     len(vgaps),
 	})
 	if gt == nil {
 		fmt.Fprintln(os.Stderr, "note: no ground truth provided — GT wall NOT applied (candidates not recall-screened)")
@@ -145,3 +160,19 @@ func envFloat(key string, def float64) float64 {
 	}
 	return def
 }
+
+func renderVocabQueue(machine string, entries []audit.DictionarySuggestion) string {
+	var b strings.Builder
+	fmt.Fprintf(&b, "# Vocab→tag review queue — %s\n\n", machine)
+	fmt.Fprintf(&b, "%d unknown token(s) appear in pattern text but map to no dictionary tag. Propose-only — a human (or the LLM) confirms the tag, then adds a keyword_dictionary entry and pins a GT case.\n\n", len(entries))
+	for i, s := range entries {
+		tag := "<tag>"
+		if len(s.SuggestedTags) > 0 {
+			tag = s.SuggestedTags[0]
+		}
+		fmt.Fprintf(&b, "## %d. \"%s\"  → suggested tag(s): %s\n", i+1, s.Token, strings.Join(s.SuggestedTags, ", "))
+		fmt.Fprintf(&b, "- named by %d pattern(s): %s\n", len(s.PatternIDs), strings.Join(s.PatternIDs, ", "))
+		fmt.Fprintf(&b, "- suggested action: add keyword_dictionary entry {%q → %s} so narratives mentioning it trigger those patterns; human confirms\n\n", s.Token, tag)
+	}
+	return b.String()
+}