Phase 5: voice consistency pass

Tier-1 voice-consistency pass: whole-document LLM review surfacing passages
that read tonally out of place (formal/over-polished/paraphrased-too-closely),
as honey-decorated `voice` flags with no correction (awareness-only).

- internal/llm/voice.go: RunVoice sends the whole document (no TruncateDoc),
  MaxTokens 2048, 20s per-doc floor (VoiceInterval). Standalone voice prompt
  in prompts.go (not bundled with the grammar checkpoint, per spec).
- internal/suggestions: POST /api/docs/:id/voice. replacePending is now
  family-scoped (pendingScope) so grammar and voice never clobber each other's
  pending flags; both passes return the unified pending set. check/voice share
  one runPass helper. TestVoicePassCoexists covers both directions.
- Frontend: api.voiceDoc, useCheckpoint voicing/runVoice, honey "Check my
  voice" toolbar pill, breathing honey dot in StatusBar.

Claude-Session: https://claude.ai/code/session_016Yr6jELuRc7hyzYLccQKZd

internal/llm/prompts.go

@@ -33,6 +33,44 @@ func CheckpointMessages(contentText string) []Message {
 	}
 }
 
+// voiceSystemPrompt drives the Tier-1 voice-consistency pass. It is a distinct
+// pass from the grammar checkpoint (spec: "do not bundle them") — the model
+// reads the whole document to learn the writer's natural voice, then flags
+// passages that read as tonally out of place. `replacement` is null: these are
+// awareness-only, with no correction to apply.
+const voiceSystemPrompt = `You are a warm, encouraging writing assistant helping someone who speaks English as a second language. ` +
+	`You are reviewing a COMPLETE document for VOICE CONSISTENCY only — not grammar.
+
+Read the whole document to learn the writer's natural voice, then identify any passages (2 or more sentences) ` +
+	`that feel tonally inconsistent with the surrounding writing — unusually formal, unusually polished, or phrased ` +
+	`in a way that differs from the writer's established voice elsewhere in the document. These often signal text ` +
+	`that was paraphrased too closely from another source. Do not flag the first paragraph (there is no baseline yet). ` +
+	`Do not flag grammar or spelling mistakes — only voice.
+
+Respond ONLY with valid JSON. No preamble, no markdown fences. Format:
+{
+  "suggestions": [
+    {
+      "original": "exact passage from the document that feels inconsistent",
+      "replacement": null,
+      "explanation": "friendly one-sentence note, e.g. 'This passage sounds more formal than the rest of your writing — worth reviewing.'",
+      "type": "voice"
+    }
+  ]
+}
+
+If the voice is consistent throughout, return: {"suggestions": []}`
+
+// VoiceMessages builds the message array for a voice-consistency pass. Unlike
+// the checkpoint, the caller passes the WHOLE document (no truncation) — voice
+// consistency is judged against the established voice everywhere else.
+func VoiceMessages(contentText string) []Message {
+	return []Message{
+		{Role: "system", Content: voiceSystemPrompt},
+		{Role: "user", Content: contentText},
+	}
+}
+
 // askPetalSystemTemplate is the Ask Petal tutor prompt. The suggestion context
 // is interpolated in; the user's own messages are appended after this system
 // turn by the caller.

internal/llm/voice.go

@@ -0,0 +1,32 @@
+package llm
+
+import (
+	"context"
+	"time"
+)
+
+// VoiceInterval is the minimum gap between voice-consistency passes for one
+// document. The pass is whole-document and slow, and it runs on an explicit
+// user action rather than a typing cadence, so this floor only guards the
+// inference endpoint against the button being mashed.
+const VoiceInterval = 20 * time.Second
+
+// RunVoice sends the WHOLE document — deliberately NOT TruncateDoc'd, since
+// voice consistency is judged against the established voice everywhere else —
+// for a Tier-1 voice pass and parses the JSON result. It reuses the checkpoint's
+// tolerant parser and a larger token budget, since one pass may flag several
+// passages. Each flag carries a null replacement (awareness-only).
+func RunVoice(ctx context.Context, client LLMClient, contentText string) ([]RawSuggestion, error) {
+	raw, err := client.Complete(ctx, CompletionRequest{
+		Messages:          VoiceMessages(contentText),
+		MaxTokens:         2048,
+		Temperature:       0.3,
+		RepetitionPenalty: 1.15,
+		TopP:              0.9,
+		Stop:              []string{"```", "\n\n\n\n"},
+	})
+	if err != nil {
+		return nil, err
+	}
+	return ParseCheckpoint(raw)
+}