fix(dispatch): bucket failures by geo cell while the backend omits zone_id

Every booking.assignment_failed observed in production carries no zone_id, so
the per-zone failure counter and the once-a-day alert limit collapsed into a
single global "unknown" bucket — one alert per day for the whole country.

zone_key() prefers the backend's zone_id and falls back to a coarse lat/lon
grid cell (1 dp, ~11 km, DISPATCH_GEO_BUCKET_DP). The failure reason is now
logged too.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012AJLYcbTHCe45fyFnMfEin

agents/dispatch_agent.py

@@ -47,6 +47,27 @@ MILER_UNAVAILABLE_STATUSES = {"break", "offline", "busy", "onbreak", "off_duty",
 
 AVAILABLE, UNAVAILABLE, UNKNOWN = "available", "unavailable", "unknown"
 
+# The Go backend does not yet send zone_id on booking.assignment_failed (every
+# event observed carries none). Without a fallback, every failure in the country
+# shares the single "unknown" bucket, so the per-zone counter and the once-a-day
+# alert limit would collapse into one global bucket. Derive a coarse grid cell
+# from the pickup coordinates instead: 1 decimal place is ~11 km, about the size
+# of a delivery zone. Real zone_ids, once sent, take precedence.
+GEO_BUCKET_DP = int(os.getenv("DISPATCH_GEO_BUCKET_DP", "1"))
+
+
+def zone_key(zone_id, lat, lon) -> str:
+    """Stable bucket for counting failures: the backend's zone_id when present,
+    else a coarse lat/lon grid cell, else 'unknown'."""
+    if zone_id and str(zone_id).lower() not in ("", "unknown", "none", "null"):
+        return str(zone_id)
+    if lat is not None and lon is not None:
+        try:
+            return f"geo:{round(float(lat), GEO_BUCKET_DP)},{round(float(lon), GEO_BUCKET_DP)}"
+        except (TypeError, ValueError):
+            pass
+    return "unknown"
+
 
 class DispatchAgent(SpecializedAgent):
     """
@@ -274,11 +295,15 @@ class DispatchAgent(SpecializedAgent):
         try:
             data = json.loads(msg.data.decode())
             booking_id = data.get("booking_id")
-            zone_id    = data.get("zone_id", "unknown")
             lat        = data.get("lat")
             lon        = data.get("lon")
+            # Falls back to a coarse geo cell while the backend omits zone_id.
+            zone_id    = zone_key(data.get("zone_id"), lat, lon)
 
-            logger.warning(f"[DISPATCH] booking.assignment_failed — booking={booking_id} zone={zone_id}")
+            logger.warning(
+                f"[DISPATCH] booking.assignment_failed — booking={booking_id} zone={zone_id} "
+                f"reason={data.get('reason')!r}"
+            )
 
             facts, count = await self._gather_assignment_facts(zone_id, lat, lon)
             logger.info(f"[DISPATCH] booking={booking_id} context facts: {facts}")