Add deterministic geography: precomputed nearest-office distances injected per turn

The 2026-07-05 Weston call (CA3e88ce) showed the model guessing geography: it
recommended Boca Raton for a Weston caller (real answer: Pembroke Pines, ~9 mi),
invented distances ('Weston to Hialeah is twenty miles'), and confirmed a
nonexistent Miami Lakes office when the caller insisted.

- practice.py: office lat/lon + ~70-place South Florida gazetteer + haversine;
  geography_block() renders a full distance table (closest first) from the
  caller's area to every office. 'Westin' aliased to Weston (STT spelling).
- callstate.py: extractor gains caller_area; CallStateGroomer appends the
  GEOGRAPHY block to the system message — zero-lag via a synchronous gazetteer
  scan of user turns until the extractor catches up.
- bot.py: DISTANCES prompt rule (answer only from GEOGRAPHY, never estimate),
  only-these-8-offices guard, caller-area cities added to Whisper hotwords
  ('Weston' was heard as 'Western Florida').

Verified: staged replay of the failed exchanges on activeblue-avc:latest —
correct office, correct distances, Miami Lakes denied (5/5 + 3/3 runs).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

bot.py

@@ -103,7 +103,11 @@ WHISPER_COMPUTE = os.environ.get("WHISPER_COMPUTE", "float16")
 WHISPER_HOTWORDS = os.environ.get(
     "WHISPER_HOTWORDS",
     "Advanced Vision Care, eye exam, annual exam, appointment, optometry, contact lens, "
-    "Hialeah, Kendall, Tamarac, Pembroke Pines, Lauderdale Lakes, Miami Gardens, Boca Raton",
+    "Hialeah, Kendall, Tamarac, Pembroke Pines, Lauderdale Lakes, Miami Gardens, Boca Raton, "
+    # Common caller-area cities (not offices) — the 2026-07-05 call heard "Weston" as
+    # "Western Florida" repeatedly, which broke the nearest-office flow.
+    "Weston, Davie, Miramar, Plantation, Sunrise, Hollywood, Coral Springs, Doral, "
+    "Miami Lakes, Aventura, Coral Gables, Homestead, Cutler Bay",
 )
 
 # Twilio sends 8 kHz mu-law on the wire, but faster-whisper assumes any numpy array is
@@ -245,7 +249,15 @@ SYSTEM_PROMPT = (
     "that matches one of our offices (for example they say 'Kendall' and we have a Kendall "
     "office), simply confirm THAT office and move on — do NOT offer, compare, or mention any "
     "other office or city, and do NOT ask them to choose between offices. Only if their area "
-    "matches no office should you name the single nearest one. List offices only if asked.\n"
+    "matches no office should you name the single nearest one. List offices only if asked. "
+    "The offices in PRACTICE FACTS are our ONLY offices — never agree we have an office "
+    "somewhere else, even if the caller insists you mentioned one.\n"
+    "- DISTANCES / CLOSEST OFFICE: when a GEOGRAPHY section appears below, answer distance and "
+    "closest-office questions from it word for word — it already lists approximate driving "
+    "miles from the caller's area to every office, closest first. If there is no GEOGRAPHY "
+    "section yet, ask what city or area they're in. NEVER estimate, compute, or guess a "
+    "distance or drive time yourself, and never claim an office is 'close to' or 'near' a "
+    "place unless the GEOGRAPHY section says so.\n"
     "- INSURANCE — log only, never promise, never guess: ask open-endedly what insurance they "
     "have (for example, 'What insurance do you have?'). Do NOT read out or suggest plan names "
     "from the list — let the caller tell you. Capture ONLY what the caller actually says; NEVER "

callstate.py

@@ -28,6 +28,8 @@ from loguru import logger
 from pipecat.frames.frames import BotStoppedSpeakingFrame, Frame, LLMContextFrame
 from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
 
+import practice
+
 # Short, in-call variant of the post-call extractor (extract.py): only what's needed to
 # build the checklist, temperature 0, capped output. Runs on the local Ollama model.
 _STATE_INSTRUCTIONS = (
@@ -54,6 +56,9 @@ _STATE_INSTRUCTIONS = (
     '"billing question"). Only \'an appointment\' with no visit reason is NOT a reason — '
     "use null then.\n"
     '  "location": string or null — the office/city the caller wants\n'
+    '  "caller_area": string or null — the city/area where the CALLER themselves is located '
+    "or wants an office near (\"I'm in Weston\", \"what's closest to Davie\") — NOT an office "
+    "they merely asked about. Corrections win here too.\n"
     '  "patient_name": string or null — ONLY a name the caller explicitly gave as their own '
     "(\"my name is…\", \"this is…\", or answering a name question); NEVER a name guessed "
     "from other words, a garbled transcription, or a name only the assistant used\n"
@@ -252,12 +257,34 @@ class CallStateGroomer(FrameProcessor):
         )
         self._task.add_done_callback(self._extract_done)
 
+    def _geo_block(self, messages) -> str:
+        """Precomputed office-distance table for the caller's area (see
+        practice.geography_block). Anchor = the LLM-extracted caller_area when we have
+        one (it understands context and corrections); before that (extraction lags one
+        turn) a synchronous gazetteer scan of the user's own words — first place that
+        isn't one of our office cities, since office names in caller speech are usually
+        about our offices, not where they live."""
+        state = self._state or {}
+        for key in ("caller_area", "location"):
+            v = state.get(key)
+            if isinstance(v, str) and v.strip():
+                anchor = practice.match_place(v)
+                if anchor:
+                    return practice.geography_block(anchor)
+        hits = []
+        for m in messages:
+            if m.get("role") == "user" and isinstance(m.get("content"), str):
+                hits += practice.places_in_text(m["content"])
+        non_office = [h for h in hits if h not in practice.OFFICE_PLACE_KEYS]
+        anchor = (non_office or hits or [None])[0]
+        return practice.geography_block(anchor) if anchor else ""
+
     def _groom_context(self):
         messages = merge_consecutive_user_messages(list(self._context.messages))
-        block = build_state_block(self._state)
+        blocks = [b for b in (build_state_block(self._state), self._geo_block(messages)) if b]
         for i, m in enumerate(messages):
             if m.get("role") == "system":
-                content = self._base_system + ("\n\n" + block if block else "")
+                content = "\n\n".join([self._base_system, *blocks])
                 if m.get("content") != content:
                     messages[i] = {**m, "content": content}
                 break

practice.py

@@ -16,18 +16,20 @@ from loguru import logger
 # ─────────────────────────────────────────────────────────────────────────────
 # Real facts from advancedvisioncareflorida.com
 # ─────────────────────────────────────────────────────────────────────────────
+# "latlon" is block-level approximate (good enough for nearest-office ranking and
+# "about X miles" answers; never used for navigation).
 LOCATIONS = [
     # Broward County
-    {"city": "Hollywood / Fort Lauderdale", "address": "2873 Stirling Rd, Fort Lauderdale, FL 33312", "phone": "(954) 983-4969"},
+    {"city": "Hollywood / Fort Lauderdale", "address": "2873 Stirling Rd, Fort Lauderdale, FL 33312", "phone": "(954) 983-4969", "latlon": (26.047, -80.183)},
-    {"city": "Tamarac", "address": "5865 N University Dr, Tamarac, FL 33321", "phone": "(954) 720-2720"},
+    {"city": "Tamarac", "address": "5865 N University Dr, Tamarac, FL 33321", "phone": "(954) 720-2720", "latlon": (26.198, -80.253)},
-    {"city": "Pembroke Pines", "address": "246 S Flamingo Rd, Pembroke Pines, FL 33027", "phone": "(954) 443-1230"},
+    {"city": "Pembroke Pines", "address": "246 S Flamingo Rd, Pembroke Pines, FL 33027", "phone": "(954) 443-1230", "latlon": (26.005, -80.312)},
-    {"city": "Lauderdale Lakes", "address": "3682 W Oakland Park Blvd, Lauderdale Lakes, FL 33311", "phone": "(954) 730-8087"},
+    {"city": "Lauderdale Lakes", "address": "3682 W Oakland Park Blvd, Lauderdale Lakes, FL 33311", "phone": "(954) 730-8087", "latlon": (26.166, -80.198)},
     # Miami-Dade County
-    {"city": "Hialeah", "address": "1770 W 32nd Pl, Hialeah, FL 33012", "phone": "(305) 885-4477"},
+    {"city": "Hialeah", "address": "1770 W 32nd Pl, Hialeah, FL 33012", "phone": "(305) 885-4477", "latlon": (25.876, -80.300)},
-    {"city": "Kendall", "address": "11605 N Kendall Dr, Miami, FL 33176", "phone": "(305) 982-8927"},
+    {"city": "Kendall", "address": "11605 N Kendall Dr, Miami, FL 33176", "phone": "(305) 982-8927", "latlon": (25.686, -80.381)},
-    {"city": "Miami Gardens", "address": "4771 NW 183rd St, Miami Gardens, FL 33055", "phone": "(305) 390-2467"},
+    {"city": "Miami Gardens", "address": "4771 NW 183rd St, Miami Gardens, FL 33055", "phone": "(305) 390-2467", "latlon": (25.942, -80.272)},
     # Palm Beach County
-    {"city": "Boca Raton", "address": "21673 State Road 7, Boca Raton, FL 33428", "phone": "(561) 470-2310"},
+    {"city": "Boca Raton", "address": "21673 State Road 7, Boca Raton, FL 33428", "phone": "(561) 470-2310", "latlon": (26.358, -80.202)},
 ]
 
 PRACTICE_FACTS = {
@@ -94,6 +96,195 @@ def _find_location(name: str):
     return None
 
 
+# ─────────────────────────────────────────────────────────────────────────────
+# Geography — deterministic nearest-office answers (no LLM, no external API).
+#
+# Added after the 2026-07-05 Weston call: the model guessed which office was close
+# ("Boca Raton is relatively close", "Weston to Hialeah is twenty miles" — invented)
+# and hallucinated a Miami Lakes office. Instead we precompute distances from the
+# caller's stated area to every office and inject them via CallStateGroomer, so the
+# model only ever repeats numbers it was handed.
+#
+# Coordinates are approximate residential centers — fine for ranking offices and
+# "about X miles", which is all a phone answer needs.
+# ─────────────────────────────────────────────────────────────────────────────
+GAZETTEER = {
+    # Broward
+    "weston": (26.080, -80.390),
+    "davie": (26.076, -80.277),
+    "southwest ranches": (26.059, -80.337),
+    "cooper city": (26.057, -80.272),
+    "plantation": (26.127, -80.250),
+    "sunrise": (26.157, -80.300),
+    "lauderhill": (26.164, -80.208),
+    "tamarac": (26.203, -80.250),
+    "north lauderdale": (26.217, -80.226),
+    "margate": (26.245, -80.206),
+    "coconut creek": (26.284, -80.184),
+    "coral springs": (26.271, -80.271),
+    "parkland": (26.310, -80.237),
+    "pompano beach": (26.238, -80.125),
+    "deerfield beach": (26.318, -80.100),
+    "lighthouse point": (26.276, -80.087),
+    "oakland park": (26.172, -80.132),
+    "wilton manors": (26.160, -80.139),
+    "fort lauderdale": (26.122, -80.137),
+    "lauderdale lakes": (26.166, -80.208),
+    "hollywood": (26.011, -80.150),
+    "dania beach": (26.052, -80.144),
+    "hallandale beach": (25.981, -80.148),
+    "pembroke pines": (26.008, -80.297),
+    "pembroke park": (25.988, -80.178),
+    "west park": (25.984, -80.199),
+    "miramar": (25.977, -80.303),
+    # Miami-Dade
+    "miami": (25.775, -80.194),
+    "miami beach": (25.790, -80.130),
+    "brickell": (25.758, -80.193),
+    "little havana": (25.773, -80.215),
+    "north miami": (25.890, -80.187),
+    "north miami beach": (25.933, -80.163),
+    "aventura": (25.956, -80.139),
+    "sunny isles beach": (25.943, -80.122),
+    "miami gardens": (25.942, -80.246),
+    "opa-locka": (25.902, -80.250),
+    "opa locka": (25.902, -80.250),
+    "miami lakes": (25.909, -80.309),
+    "hialeah": (25.858, -80.278),
+    "hialeah gardens": (25.865, -80.324),
+    "medley": (25.838, -80.332),
+    "doral": (25.819, -80.355),
+    "sweetwater": (25.763, -80.373),
+    "fontainebleau": (25.772, -80.346),
+    "westchester": (25.745, -80.327),
+    "west miami": (25.757, -80.296),
+    "coral gables": (25.721, -80.268),
+    "south miami": (25.707, -80.293),
+    "pinecrest": (25.667, -80.308),
+    "kendall": (25.679, -80.355),
+    "west kendall": (25.706, -80.439),
+    "kendale lakes": (25.708, -80.407),
+    "the hammocks": (25.670, -80.445),
+    "tamiami": (25.758, -80.402),
+    "country walk": (25.633, -80.432),
+    "richmond west": (25.611, -80.430),
+    "palmetto bay": (25.622, -80.320),
+    "cutler bay": (25.575, -80.336),
+    "homestead": (25.469, -80.478),
+    "florida city": (25.448, -80.479),
+    "key biscayne": (25.691, -80.163),
+    # Palm Beach (southern)
+    "boca raton": (26.359, -80.130),
+    "west boca": (26.351, -80.210),
+    "delray beach": (26.461, -80.073),
+    "boynton beach": (26.525, -80.087),
+    "lake worth": (26.617, -80.072),
+    "wellington": (26.659, -80.241),
+    "west palm beach": (26.715, -80.054),
+}
+
+_PLACE_ALIASES = {
+    "westin": "weston",  # frequent STT spelling of Weston (hotel-chain bias)
+    "ft lauderdale": "fort lauderdale",
+    "ft. lauderdale": "fort lauderdale",
+    "kendall west": "west kendall",
+    "hallandale": "hallandale beach",
+    "sunny isles": "sunny isles beach",
+    "the pines": "pembroke pines",
+}
+
+# Gazetteer keys that ARE one of our office cities (used to prefer the caller's own
+# area over office names they merely asked about).
+OFFICE_PLACE_KEYS = frozenset({
+    "hollywood", "fort lauderdale", "tamarac", "pembroke pines", "lauderdale lakes",
+    "hialeah", "kendall", "miami gardens", "boca raton",
+})
+
+# Which office a gazetteer key belongs to, for the "right in <city>" special case.
+_OFFICE_FOR_KEY = {
+    "hollywood": "Hollywood / Fort Lauderdale",
+    "fort lauderdale": "Hollywood / Fort Lauderdale",
+    "tamarac": "Tamarac",
+    "pembroke pines": "Pembroke Pines",
+    "lauderdale lakes": "Lauderdale Lakes",
+    "hialeah": "Hialeah",
+    "kendall": "Kendall",
+    "miami gardens": "Miami Gardens",
+    "boca raton": "Boca Raton",
+}
+
+# Longest names first so "north miami beach" wins over "north miami" over "miami".
+_PLACE_RE = re.compile(
+    r"\b(" + "|".join(
+        re.escape(k) for k in sorted(list(GAZETTEER) + list(_PLACE_ALIASES), key=len, reverse=True)
+    ) + r")\b"
+)
+
+# Straight-line → driving miles fudge factor for the South Florida grid.
+_ROAD_FACTOR = 1.3
+
+
+def _haversine_miles(a, b) -> float:
+    import math
+    lat1, lon1, lat2, lon2 = map(math.radians, (*a, *b))
+    h = (math.sin((lat2 - lat1) / 2) ** 2
+         + math.cos(lat1) * math.cos(lat2) * math.sin((lon2 - lon1) / 2) ** 2)
+    return 3958.8 * 2 * math.asin(math.sqrt(h))
+
+
+def _road_miles(a, b) -> int:
+    return max(1, round(_haversine_miles(a, b) * _ROAD_FACTOR))
+
+
+def match_place(text: str):
+    """Canonical gazetteer key for a free-text place mention, or None."""
+    if not text:
+        return None
+    m = _PLACE_RE.search(text.lower())
+    if not m:
+        return None
+    k = m.group(1)
+    return _PLACE_ALIASES.get(k, k)
+
+
+def places_in_text(text: str):
+    """All gazetteer places mentioned in the text, canonical, in order of appearance."""
+    if not text:
+        return []
+    return [_PLACE_ALIASES.get(k, k) for k in _PLACE_RE.findall(text.lower())]
+
+
+def geography_block(place_key: str) -> str:
+    """Precomputed distance table from the caller's area to every office, closest
+    first — injected into the system prompt so distance / nearest-office answers are
+    read, never guessed. Returns "" for an unknown place."""
+    coords = GAZETTEER.get(place_key)
+    if not coords:
+        return ""
+    pretty = place_key.title()
+    ranked = sorted(
+        ((l["city"], _road_miles(coords, l["latlon"])) for l in LOCATIONS),
+        key=lambda x: x[1],
+    )
+    table = "; ".join(
+        f"{city} about {mi} miles" + (" (CLOSEST)" if i == 0 else "")
+        for i, (city, mi) in enumerate(ranked)
+    )
+    home = _OFFICE_FOR_KEY.get(place_key)
+    if home:
+        situated = f"We have an office right in {pretty} (the {home} office)."
+    else:
+        situated = (f"We do NOT have an office in {pretty}; the closest is "
+                    f"{ranked[0][0]}, about {ranked[0][1]} miles away.")
+    return (
+        f"GEOGRAPHY (precomputed, approximate driving distances from {pretty} — the ONLY "
+        f"source you may use for any distance or closest-office answer; NEVER estimate a "
+        f"distance yourself and NEVER confirm an office at any place not in PRACTICE FACTS):\n"
+        f"- {situated}\n"
+        f"- From {pretty}: {table}."
+    )
+
+
 # ─── Tools (used when ENABLE_TOOLS=true and the model supports tool-calling) ──
 
 def persist_appointment(record: dict) -> str: