backend/src/common/digits.spec.ts

@@ -0,0 +1,31 @@
+import { toLatinDigits } from './digits';
+
+/**
+ * Exercised by the voice pipeline on two untrusted inputs: spoken dates, and the tooth
+ * code the extraction model echoes back — a Persian-digit "۲۶" that fails to normalise
+ * costs the clinician a tooth, silently.
+ */
+describe('toLatinDigits', () => {
+  it('normalises Persian digits and leaves everything else alone', () => {
+    expect(
+      toLatinDigits('\u06F1\u06F4\u06F0\u06F4/\u06F0\u06F7/\u06F2\u06F5'),
+    ).toBe('1404/07/25');
+    expect(toLatinDigits('1404/07/25')).toBe('1404/07/25');
+    expect(toLatinDigits('\u062F\u0646\u062F\u0627\u0646 \u06F1\u06F4')).toBe(
+      '\u062F\u0646\u062F\u0627\u0646 14',
+    );
+  });
+
+  it('also normalises the Arabic-Indic block, which ASR output can carry', () => {
+    // U+0660..U+0669, distinct code points from the Persian U+06F0..U+06F9 block.
+    expect(
+      toLatinDigits('\u0661\u0664\u0660\u0664/\u0660\u0667/\u0662\u0665'),
+    ).toBe('1404/07/25');
+  });
+
+  it('normalises a transcript that mixes both blocks with ASCII', () => {
+    expect(toLatinDigits('\u06F1\u06F4 and \u0661\u0665 and 16')).toBe(
+      '14 and 15 and 16',
+    );
+  });
+});

backend/src/common/digits.ts

@@ -0,0 +1,24 @@
+/**
+ * Persian (Extended Arabic-Indic, U+06F0–U+06F9) zero, and Arabic-Indic (U+0660–U+0669)
+ * zero. ASR output can carry either block, sometimes mixed with ASCII in one transcript.
+ */
+const PERSIAN_ZERO = 0x06f0;
+const ARABIC_INDIC_ZERO = 0x0660;
+
+/**
+ * Normalise Persian and Arabic-Indic digits to ASCII. Non-digits pass through.
+ *
+ * Deliberately wider than the frontend original, which only handles the Persian block:
+ * this parses model/ASR output rather than keystrokes, so both blocks must be accepted
+ * or a spoken date or tooth number silently degrades to "unresolved".
+ *
+ * Lives on its own rather than inside jalali.ts because tooth codes need it too, and a
+ * tooth module reaching into the calendar module would read as an accident.
+ */
+export function toLatinDigits(value: string): string {
+  return value.replace(/[۰-۹٠-٩]/g, (ch) => {
+    const code = ch.charCodeAt(0);
+    const base = code >= PERSIAN_ZERO ? PERSIAN_ZERO : ARABIC_INDIC_ZERO;
+    return String(code - base);
+  });
+}

backend/src/common/fdi.ts

@@ -6,6 +6,8 @@
  * midline pairs (11–21, 41–31) are neighbours, exactly as the chart treats them.
  */
 
+import { toLatinDigits } from './digits';
+
 export type Arch = 'upper' | 'lower';
 
 /** Which side of the *patient*, not of the screen. Quadrant 1 is the patient's upper right. */
@@ -65,6 +67,19 @@ export function isFdiTooth(value: unknown): value is string {
   return typeof value === 'string' && FDI_TOOTH_IDS.has(value);
 }
 
+/**
+ * Clean up a tooth code the extraction model echoed back, before it is matched.
+ *
+ * The model is transcribing Persian speech, so it can hand back "۲۶" in Persian digits or
+ * "2 6" from a digit-by-digit dictation. Neither matches an FDI code literally, and a
+ * near-miss here does not fail loudly — the tooth quietly turns into "not understood".
+ * Returns '' for anything that is not a string.
+ */
+export function normalizeFdiCode(value: unknown): string {
+  if (typeof value !== 'string') return '';
+  return toLatinDigits(value).replace(/\s+/g, '');
+}
+
 function archOrder(tooth: string): readonly string[] | null {
   if ((FDI_UPPER_ARCH_ORDER as readonly string[]).includes(tooth))
     return FDI_UPPER_ARCH_ORDER;

backend/src/common/jalali.spec.ts

@@ -5,7 +5,6 @@ import {
   jalaliDaysInMonth,
   jalaliToGregorian,
   jalaliToIsoDate,
-  toLatinDigits,
 } from './jalali';
 
 describe('jalali calendar', () => {
@@ -87,29 +86,4 @@ describe('jalali calendar', () => {
       expect(jalaliDaysInMonth(1404, 13)).toBe(0);
     });
   });
-
-  describe('toLatinDigits', () => {
-    it('normalises Persian digits and leaves everything else alone', () => {
-      expect(
-        toLatinDigits('\u06F1\u06F4\u06F0\u06F4/\u06F0\u06F7/\u06F2\u06F5'),
-      ).toBe('1404/07/25');
-      expect(toLatinDigits('1404/07/25')).toBe('1404/07/25');
-      expect(toLatinDigits('\u062F\u0646\u062F\u0627\u0646 \u06F1\u06F4')).toBe(
-        '\u062F\u0646\u062F\u0627\u0646 14',
-      );
-    });
-
-    it('also normalises the Arabic-Indic block, which ASR output can carry', () => {
-      // U+0660..U+0669, distinct code points from the Persian U+06F0..U+06F9 block.
-      expect(
-        toLatinDigits('\u0661\u0664\u0660\u0664/\u0660\u0667/\u0662\u0665'),
-      ).toBe('1404/07/25');
-    });
-
-    it('normalises a transcript that mixes both blocks with ASCII', () => {
-      expect(toLatinDigits('\u06F1\u06F4 and \u0661\u0665 and 16')).toBe(
-        '14 and 15 and 16',
-      );
-    });
-  });
 });

backend/src/common/jalali.ts

@@ -158,28 +158,6 @@ export function jalaliDaysInMonth(jy: number, jm: number): number {
   return isJalaliLeapYear(jy) ? 30 : 29;
 }
 
-/**
- * Persian (Extended Arabic-Indic, U+06F0–U+06F9) zero, and Arabic-Indic (U+0660–U+0669)
- * zero. ASR output can carry either block, sometimes mixed with ASCII in one transcript.
- */
-const PERSIAN_ZERO = 0x06f0;
-const ARABIC_INDIC_ZERO = 0x0660;
-
-/**
- * Normalise Persian and Arabic-Indic digits to ASCII. Non-digits pass through.
- *
- * Deliberately wider than the frontend original, which only handles the Persian block:
- * this parses model/ASR output rather than keystrokes, so both blocks must be accepted
- * or a spoken date silently degrades to "unresolved".
- */
-export function toLatinDigits(value: string): string {
-  return value.replace(/[\u06F0-\u06F9\u0660-\u0669]/g, (ch) => {
-    const code = ch.charCodeAt(0);
-    const base = code >= PERSIAN_ZERO ? PERSIAN_ZERO : ARABIC_INDIC_ZERO;
-    return String(code - base);
-  });
-}
-
 /** True when the triple is a real Jalali date inside the supported year range. */
 export function isValidJalaliDate(jy: number, jm: number, jd: number): boolean {
   if (!Number.isInteger(jy) || !Number.isInteger(jm) || !Number.isInteger(jd)) {

backend/src/modules/voice/extraction.wire.spec.ts

@@ -56,6 +56,33 @@ describe('VOICE_INTENT_JSON_SCHEMA', () => {
 });
 
 describe('toVoiceIntent', () => {
+  it('takes the explicit branch for a code the model wrote in Persian digits', () => {
+    // The model is reading Persian text back, so "۲۶" and a digit-by-digit "2 6" both
+    // reach us. Matching only ASCII drops the tooth into the positional branch with no
+    // quadrant, where it is reported as unresolved — the clinician loses a tooth and is
+    // told the words were the problem.
+    for (const raw of ['\u06F2\u06F6', '2 6', ' 26 ', '\u0662\u0666']) {
+      const [tooth] = toVoiceIntent(
+        wire({
+          teeth: [
+            {
+              spoken: '\u0628\u06CC\u0633\u062A \u0648 \u0634\u0634',
+              fdi: raw,
+              arch: null,
+              side: null,
+              position: null,
+            },
+          ],
+        }),
+      ).teeth;
+      expect(tooth).toEqual({
+        kind: 'explicit',
+        fdi: '26',
+        spoken: '\u0628\u06CC\u0633\u062A \u0648 \u0634\u0634',
+      });
+    }
+  });
+
   it('narrows a positional tooth', () => {
     const result = toVoiceIntent(wire({ teeth: [positionalTooth] }));
     expect(result.teeth[0]).toEqual({

backend/src/modules/voice/extraction.wire.ts

@@ -1,3 +1,4 @@
+import { normalizeFdiCode } from '../../common/fdi';
 import type {
   ConnectedSpanIntent,
   DueIntent,
@@ -180,7 +181,10 @@ const FDI_SHAPE = /^[1-8][1-8]$/;
 
 function toToothIntent(wire: WireToothIntent | undefined | null): ToothIntent {
   const spoken = typeof wire?.spoken === 'string' ? wire.spoken : '';
-  const fdi = typeof wire?.fdi === 'string' ? wire.fdi.trim() : '';
+  // Persian digits and digit-by-digit dictation ("۲۶", "2 6") are FDI codes that do not
+  // match literally; without normalising first they fall through to the positional branch
+  // with no quadrant and are reported as unresolved.
+  const fdi = normalizeFdiCode(wire?.fdi);
   // Only take the explicit branch for something actually FDI-shaped. A model that emits
   // fdi:"6" alongside correct arch/side/position would otherwise lose the tooth entirely.
   if (FDI_SHAPE.test(fdi)) {

backend/src/modules/voice/tooth-intent.resolver.spec.ts

@@ -134,6 +134,14 @@ describe('resolveToothIntents', () => {
     }
   });
 
+  it('reads a spoken number as its FDI code, digits in any script', () => {
+    // The product rule: the number the clinician says IS the tooth. 26 = quadrant 2
+    // (patient's upper left) + position 6 = first molar.
+    for (const raw of ['26', ' 26 ', '2 6', '\u06F2\u06F6', '\u0662\u0666']) {
+      expect(resolveToothIntents([explicit(raw, 'x')]).teeth).toEqual(['26']);
+    }
+  });
+
   it('trims an explicit code, matching normalizeTeeth', () => {
     expect(resolveToothIntents([explicit(' 14 ', 'x')]).teeth).toEqual(['14']);
   });

backend/src/modules/voice/tooth-intent.resolver.ts

@@ -1,4 +1,4 @@
-import { isFdiTooth, toFdi } from '../../common/fdi';
+import { isFdiTooth, normalizeFdiCode, toFdi } from '../../common/fdi';
 import type { ToothIntent, UnresolvedItem } from './voice.types';
 
 export type ToothResolution = {
@@ -9,10 +9,10 @@ export type ToothResolution = {
 
 /** Everything here parses untrusted model output, so nothing may throw. */
 function normalizedFdi(intent: ToothIntent): string {
-  const raw = (intent as { fdi?: unknown }).fdi;
+  // Same normalisation the wire layer used to pick this branch, so the two cannot
-  // Trimmed for parity with normalizeTeeth — '14 ' is tooth 14 through the treatment API
+  // disagree: '14 ' is tooth 14 through the treatment API and '۲۶' is tooth 26, and
-  // and must not be "malformed" here.
+  // neither may be reported as malformed here.
-  return typeof raw === 'string' ? raw.trim() : '';
+  return normalizeFdiCode((intent as { fdi?: unknown }).fdi);
 }
 
 /**