feat(voice): keep the transcript on the server

The review sheet rendered the full dictation at the top of the modal,
unconditionally. A raw transcript can carry the patient's spoken name — the
spec said so itself in §9, while §10 said telemetry must never contain it.

The transcript no longer reaches the browser by any route:

- removed from the success response (VoiceExtractionResponse is now plain
  ResolvedExtraction, which never had the field)
- removed from the error body. VOICE_EXTRACT_FAILED carried
  details.transcript for a salvage dialog that was never built, and
  ApiError['details'] is an array, so the shape never even matched — it was
  serialized onto the wire and dropped
- removed from the sheet, and from VoiceExtractionResult. tsc proves that
  <p> was the only reader in the whole frontend

It is logged instead: one info line per recording, written immediately after
the emptiness check so a failed extraction still records it, and deliberately
outside logTelemetry so that method's patient-free guarantee stays literally
true.

Accepted consequence, recorded in §10: patient words now persist in
production server logs at default level, so whatever retention and access
control applies to those logs applies to dictation. The repo's other
sensitive-text path (openrouter.provider.ts) uses debug level with
truncation; moving this line to debug is a one-word change.

Transcript salvage is dropped rather than deferred, which settles §11 open
item 16 by taking its second option. When extraction fails the clinician
re-dictates; an operator can read the words in the log, the person who spoke
them cannot.

Spec: §7, §9 and §10 rewritten, item 16 resolved, decisions 51-53 added, and
decisions 10 and 25 marked superseded so the log stops contradicting itself.

No automated coverage for the response shape or the log line: there is no
voice.service.spec.ts — the service is I/O orchestration and has never been
unit tested. Removing the type field is what proves no reader survives.

Gates: backend 216 tests, nest build, ESLint clean on the voice module;
frontend tsc --noEmit clean, 52 Vitest tests, next build clean.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

backend/src/modules/voice/voice.service.ts

@@ -35,9 +35,12 @@ export type VoiceAvailability = {
   maxRecordingMs: number | null;
 };
 
-export type VoiceExtractionResponse = ResolvedExtraction & {
-  transcript: string;
-};
+/**
+ * The transcript is deliberately absent. A raw dictation can carry the patient's spoken name, so
+ * it never leaves the server — it is logged there instead (§10). Nothing the client renders needs
+ * it, and what is not sent cannot leak through the network tab or an error reporter.
+ */
+export type VoiceExtractionResponse = ResolvedExtraction;
 
 @Injectable()
 export class VoiceService {
@@ -119,13 +122,20 @@ export class VoiceService {
       );
     }
 
-    // Stage 2 — structure it. On failure the transcript still goes back to the client so
-    // the words the clinician already paid for are not lost (transcript salvage).
+    // The transcript's only destination. Logged before extraction so it survives an extraction
+    // failure too, and kept out of logTelemetry so that method's patient-free guarantee stays
+    // true. This line DOES carry what the clinician said, which may include a patient's name.
+    this.logger.log(
+      `voice transcript [${catalogLocale}]: ${transcript.trim()}`,
+    );
+
+    // Stage 2 — structure it. A failure here returns a code only; the transcript stays in the
+    // server log above, never in the response.
     let resolved: ResolvedExtraction;
     let llmCost: number | null = null;
     try {
-      // Inside the try: the transcript is already paid for, so a catalog/DB failure here
-      // must still salvage it rather than becoming a generic 500 that throws it away.
+      // Inside the try: a catalog/DB failure here must surface as VOICE_EXTRACT_FAILED, which
+      // the clinician can act on, rather than a generic 500.
       const catalog = await this.buildCatalog(organizationId, catalogLocale);
       const result = await extraction.extract(
         transcript,
@@ -148,7 +158,7 @@ export class VoiceService {
         linkedLabIds: new Set(catalog.labs.map((l) => l.id)),
       });
     } catch (error) {
-      throw this.toAppException(error, 'extraction', transcript);
+      throw this.toAppException(error, 'extraction');
     }
 
     this.logTelemetry({
@@ -160,7 +170,7 @@ export class VoiceService {
       resolved,
     });
 
-    return { ...resolved, transcript };
+    return resolved;
   }
 
   private assertOrganization(user: { organizationId?: string }): string {
@@ -303,7 +313,6 @@ export class VoiceService {
   private toAppException(
     error: unknown,
     stage: 'asr' | 'extraction',
-    transcript?: string,
   ): AppException {
     if (error instanceof Error && error.name === 'AbortError') {
       // The clinician cancelled; not a failure worth a translated message.
@@ -318,11 +327,9 @@ export class VoiceService {
       stage === 'asr'
         ? ErrorCode.VOICE_ASR_FAILED
         : ErrorCode.VOICE_EXTRACT_FAILED;
-    return new AppException(
-      code,
-      HttpStatus.BAD_GATEWAY,
-      transcript ? { transcript } : undefined,
-    );
+    // No details: the transcript used to ride along here for a salvage dialog that was never
+    // built, so it was serialized onto the wire and dropped. It stays on the server now.
+    return new AppException(code, HttpStatus.BAD_GATEWAY);
   }
 
   /** Structured and patient-free: never the transcript, never audio, never a patient id. */

docs/specs/voice-treatment-entry/spec.md

@@ -664,9 +664,9 @@ assignment's existing targets. An item raised outside any assignment (a tooth sp
 > same constraint the realtime soft-refresh already lives under: never remount the
 > treatment form, never clear a draft.
 
-- Renders the transcript, then one row per extracted field in the app's own vocabulary:
-  translated catalog labels, and a mini FDI chart for the teeth rather than a list of
-  numbers.
+- Renders one row per extracted field in the app's own vocabulary: translated catalog labels,
+  and a mini FDI chart for the teeth rather than a list of numbers. **The transcript is not
+  shown** — it never reaches the browser at all (§10).
 - Each row has a checkbox. Ticked rows apply; nothing else is touched. Confirm is also
   what creates the new detail — see §2.
 - Rows default to ticked **except** the lab row when `labMatchExact` is false — shipping to a
@@ -822,26 +822,16 @@ function already had for Safari versions that shipped no `isTypeSupported`; `ons
 the real container from `recorder.mimeType`, as it already did. `VOICE_UNSUPPORTED_FORMAT` is
 left for a browser that genuinely cannot record.
 
-**Transcript salvage — specified, NOT built.** The backend half exists: `VOICE_EXTRACT_FAILED`
-carries `details.transcript` and `HttpExceptionFilter` forwards it. The client half was
-never written — `onError` only resolves a message through `getUserFacingError`, which never
-reads `details`, so the transcript is shipped in an error body and dropped. Either build the
-dialog below or stop returning the transcript; shipping dictation to the client and
-discarding it is the worst of both.
+**Transcript salvage — dropped, not deferred.** `VOICE_EXTRACT_FAILED` used to carry
+`details.transcript` so a failure dialog could offer the words back as a note. It was never
+built, and the transcript no longer reaches the client at all (§10), so the dialog as specified
+cannot be built either. `toAppException` now returns a code and no `details`.
 
-When ASR succeeded and only extraction failed, the response still
-carries the transcript and the failure dialog offers *"افزودن به یادداشت"*. That action
-**creates a new detail with only `comment` set to the transcript** — everything else left
-at `newDetail()` defaults. The words were captured and paid for; only the structure was
-lost.
+The trade, stated plainly: when ASR succeeded and only extraction failed, the words were
+captured and paid for, and the clinician cannot be offered them. They are in the server log,
+readable by an operator, not by the person who spoke them. The clinician re-dictates.
 
-This keeps the feature's one invariant intact: **voice never writes into an existing
-detail.** Dictating into an already-filled detail is a separate, later feature with its
-own voice-to-text control scoped to that field (§1, out of scope).
-
-It also does not bypass the confirmation rule — the dialog shows the transcript, and the
-dentist taps to accept it. That review matters, because a raw transcript carries ASR
-errors and may contain the patient's spoken name, and `comment` is persisted (§10).
+Reversing this needs a decision about the transcript leaving the server, not just client code.
 
 ---
 
@@ -850,7 +840,20 @@ errors and may contain the patient's spoken name, and `comment` is persisted (§
 - Audio is held **in memory for the request only**. Never written to disk, never a Prisma
   row. Note this is deliberately unlike treatment attachments, which do persist to
   `backend/uploads/treatments`.
-- The transcript goes to the browser for the review sheet and dies with it.
+- **The transcript never leaves the server.** It is not in the success response and not in any
+  error body. A raw dictation can carry the patient's spoken name, and the review sheet has no
+  need of it — the resolved rows are what the clinician confirms.
+- The transcript **is** written to the server log, once per recording, at **info** level
+  (`voice.service.ts`, immediately after the emptiness check so an extraction failure still
+  records it). This is a deliberate exception to the rule below, and the only durable trace
+  besides `comment`. It is logged before extraction, not inside `logTelemetry`, so that
+  method's patient-free guarantee stays literally true.
+
+  > ⚠ Consequence to accept: patient words persist in production server logs at default level.
+  > Whatever retention and access control applies to those logs now applies to dictation. The
+  > repo's other sensitive-text path — a vendor error body that can echo the request back —
+  > uses `debug` level and truncates to 500 characters
+  > (`openrouter.provider.ts`). Moving this line to `debug` is a one-word change.
 - The `comment` field persists a cleaned version of what was said — that is legitimate
   clinical record-keeping and is the only durable trace.
 - Telemetry is **structured and patient-free**: clip duration, which fields resolved,
@@ -969,11 +972,10 @@ enabling this for real clinics.
     pre-authentication allocation and nobody has decided whether that is acceptable. Not
     changed in this revision.
 
-16. **Transcript salvage is still specified and not built.** `VOICE_EXTRACT_FAILED` carries
-    `details.transcript`, `HttpExceptionFilter` forwards it, and nothing on the client reads
-    it — `getUserFacingError` resolves a message only. The field also does not match
-    `ApiError['details']`, which is an array. Either build the dialog in §9 or stop returning
-    the transcript. Carried forward unchanged; it is orthogonal to the prosthesis model.
+16. ~~**Transcript salvage is still specified and not built.**~~ — **resolved 2026-09-10:** of
+    its two options, "stop returning the transcript" was taken. The transcript is no longer in
+    the success response or in `details`, the review sheet no longer shows it, and it is logged
+    on the server instead (§9, §10). The `ApiError['details']` shape mismatch goes away with it.
 
 ---
 
@@ -1073,14 +1075,14 @@ Settled in a grilling session on 2026-08-20.
 | 7 | Due date | Intent + deterministic resolver |
 | 8 | Resolver location | Backend, Jalali math ported |
 | 9 | Prosthesis | Default type + overrides, all-or-nothing |
-| 10 | Retention | Discard audio and transcript, non-PHI telemetry only |
+| 10 | Retention | Discard audio and transcript, non-PHI telemetry only — **partly superseded by 52**: the transcript is now logged on the server at info level |
 | 11 | Capture | Tap to start/stop, hard cap (see 27) |
 | 12 | Failure UX | Stage-aware codes, transcript salvage |
 | 13 | Locales | Provider registry per locale; all three locales enabled |
 | 14 | Reachability | Registry now, slots filled per deployment |
 | 23 | ASR model | `openai/whisper-1` for **every** locale; registry kept so `fa` can diverge |
 | 24 | Extraction model | **`google/gemini-3.7-flash`**; escalation path documented in §4 |
-| 25 | Salvage target | Creates a new detail with only `comment` set — voice never writes into an existing detail |
+| 25 | Salvage target | Creates a new detail with only `comment` set — voice never writes into an existing detail. **Superseded by 53**: salvage is dropped |
 | 26 | Throttle | Configurable; v1 default 6 requests / 60s per user |
 | 27 | Duration cap | **2 minutes**, configurable via `maxMs` |
 | 28 | Review sheet | Modal on desktop, full-screen overlay (not a route) on mobile; candidate chips are its only interactive part |
@@ -1139,3 +1141,11 @@ Corrections to the v1 text found in the same pass: the endpoint is `POST /voice/
 `/treatments/voice-extract` (§3); `GET /voice/availability` does exist and only the plan check is
 deferred (§11 item 13); the catalog has 5 subcategories, not 4, and the disjointness test asserts
 against the live catalog rather than a written count (§5).
+
+Transcript handling revised on 2026-09-10.
+
+| # | Question | Decision |
+|---|---|---|
+| 51 | Who sees the transcript | Nobody outside the server. It is absent from the success response and from every error body, and the review sheet does not render it — a raw dictation can carry the patient's spoken name, and what is not sent cannot leak through the network tab or an error reporter (§7, §10) |
+| 52 | Where it goes instead | One **info**-level server log line per recording, written before extraction so a failed extraction still records it, and outside `logTelemetry` so that method stays patient-free. Accepted consequence: patient words persist in production logs at default level; `debug` is a one-word change (§10) |
+| 53 | Transcript salvage | Dropped, not deferred. Reversing it needs a decision about the transcript leaving the server, not just client code. This supersedes decision 25 (§9) |

frontend/src/components/treatment/voiceReviewRows.spec.ts

@@ -59,7 +59,6 @@ const LAB_DEPENDENT = new Set(['prosthesis']);
 
 function baseResult(overrides: Partial<VoiceExtractionResult> = {}): VoiceExtractionResult {
   return {
-    transcript: '',
     treatmentType: 'restoration',
     teeth: [],
     toothSelectionGroups: [],

frontend/src/components/ui/treatment/VoiceReviewSheet.tsx

@@ -219,10 +219,6 @@ export function VoiceReviewSheet({
           {t('voiceReviewTitle')}
         </h2>
 
-        <p className="mt-2 rounded-[var(--radius-md)] bg-background-card/60 px-3 py-2 text-sm text-text-secondary">
-          {effective.transcript}
-        </p>
-
         {nothingToApply ? (
           <p className="mt-4 text-sm text-text-secondary">{t('voiceNothingExtracted')}</p>
         ) : (

frontend/src/types/voice.ts

@@ -46,7 +46,10 @@ export interface VoiceProsthesisAssignment {
 }
 
 export interface VoiceExtractionResult {
-  transcript: string;
+  /**
+   * No transcript. A raw dictation can carry the patient's spoken name, so the server never
+   * sends it — it is logged there instead (spec §10).
+   */
   treatmentType: string | null;
   teeth: FdiToothId[];
   toothSelectionGroups: ToothSelectionGroup[];