diff --git a/supabase/migrations/206_locations_rededupe_by_city.sql b/supabase/migrations/206_locations_rededupe_by_city.sql
new file mode 100644
index 0000000..4bc4e6c
--- /dev/null
+++ b/supabase/migrations/206_locations_rededupe_by_city.sql
@@ -0,0 +1,142 @@
+-- Migration 206: Re-dedupe locations by (name, city, address)
+--
+-- Problem
+-- -------
+-- The original backfill (after migration 203) grouped venues by (name, address)
+-- only — so "Tractor Supply" in Brighton, CO and "Tractor Supply" in
+-- Canon City, CO both got rolled up to one location if they happened to share
+-- the same address (or both had a NULL address). The same affected:
+--   * Tractor Supply         — 58 stops at 10 distinct (city, address) combos
+--   * Ace Hardware           — 27 stops at 3 distinct combos
+--   * Murdoch's Ranch & Home — 25 stops at 3 distinct combos
+--   * Big Tool Box           — 15 stops at 2 distinct combos
+--   * JAX Farm & Ranch       — 13 stops at 2 distinct combos
+-- Result: 5 of the 26 venues were each represented as a single over-broad
+-- location record, hiding the real per-city venue from the Locations tab.
+--
+-- Fix
+-- ---
+-- Re-dedupe by (name, city, address). For each unique combo, create a fresh
+-- location row, re-link stops to it, and soft-delete the over-grouped originals
+-- (which by then have zero stop references). Idempotent: safe to re-run; on
+-- re-run the CTE returns no rows and the script is a no-op.
+--
+-- Result: 41 locations for the Tuxedo brand (up from 26). 269 stops still
+-- linked, just to the correct per-city venue record.
+
+DO $$
+DECLARE
+  v_brand_id UUID := '64294306-5f42-463d-a5e8-2ad6c81a96de';
+  v_inserted INT := 0;
+  v_relinked INT := 0;
+  v_soft_del INT := 0;
+BEGIN
+  -- 1. Insert one new location per distinct (name, city, address) tuple that
+  --    isn't already a current non-deleted location. Skip if a row with the
+  --    same (brand_id, name, lower(city), lower(address)) already exists.
+  WITH venues AS (
+    SELECT DISTINCT
+      s.brand_id,
+      s.location  AS name,
+      NULLIF(btrim(COALESCE(s.city, '')), '')    AS city,
+      NULLIF(btrim(COALESCE(s.address, '')), '') AS address
+    FROM public.stops s
+    WHERE s.brand_id = v_brand_id
+      AND s.deleted_at IS NULL
+      AND s.location IS NOT NULL
+      AND length(trim(s.location)) > 0
+  ),
+  ranked AS (
+    -- Stable, per-(name, city, address) ordering so slugs are reproducible
+    SELECT
+      v.*,
+      row_number() OVER (
+        PARTITION BY v.brand_id, lower(trim(v.name))
+        ORDER BY lower(COALESCE(v.city, '')), lower(COALESCE(v.address, ''))
+      ) AS rn
+    FROM venues v
+  ),
+  inserted AS (
+    INSERT INTO public.locations (
+      brand_id, name, city, address, active, slug
+    )
+    SELECT
+      r.brand_id,
+      trim(r.name),
+      r.city,
+      r.address,
+      true,
+      -- Slug: name, optionally suffixed with -city, then a numeric counter
+      -- only when multiple venues share the same name (e.g. Tractor Supply
+      -- appears in 10 cities, so we get -1, -2, … -10).
+      -- Use [[:alnum:]] so uppercase letters aren't stripped (e.g. "Tractor
+      -- Supply" → "tractor-supply", not "ractor-upply").
+      trim(BOTH '-' FROM lower(regexp_replace(trim(r.name), '[^[:alnum:]]+', '-', 'g')))
+        || CASE
+             WHEN r.city IS NOT NULL AND length(trim(r.city)) > 0
+               THEN '-' || trim(BOTH '-' FROM lower(regexp_replace(trim(r.city), '[^[:alnum:]]+', '-', 'g')))
+             ELSE ''
+           END
+        || CASE WHEN r.rn > 1 THEN '-' || r.rn::text ELSE '' END
+    FROM ranked r
+    WHERE NOT EXISTS (
+      SELECT 1 FROM public.locations l
+      WHERE l.brand_id  = r.brand_id
+        AND l.name      = trim(r.name)
+        AND l.deleted_at IS NULL
+        AND lower(COALESCE(l.city,    '')) = lower(COALESCE(r.city,    ''))
+        AND lower(COALESCE(l.address, '')) = lower(COALESCE(r.address, ''))
+    )
+    RETURNING id, name, city, address
+  )
+  SELECT COUNT(*) INTO v_inserted FROM inserted;
+  RAISE NOTICE 'Inserted % new per-city locations', v_inserted;
+
+  -- 2. Re-link each stop to the location whose (name, city, address) matches.
+  --    Use lower() on both sides for case-insensitive matching.
+  WITH new_links AS (
+    SELECT
+      s.id AS stop_id,
+      l.id AS location_id
+    FROM public.stops s
+    JOIN public.locations l
+      ON l.brand_id  = s.brand_id
+     AND lower(trim(l.name)) = lower(trim(s.location))
+     AND lower(COALESCE(l.city,    '')) = lower(COALESCE(NULLIF(btrim(s.city),    ''), ''))
+     AND lower(COALESCE(l.address, '')) = lower(COALESCE(NULLIF(btrim(s.address), ''), ''))
+     AND l.deleted_at IS NULL
+    WHERE s.brand_id = v_brand_id
+      AND s.deleted_at IS NULL
+  )
+  UPDATE public.stops s
+  SET location_id = nl.location_id
+  FROM new_links nl
+  WHERE s.id = nl.stop_id;
+
+  GET DIAGNOSTICS v_relinked = ROW_COUNT;
+  RAISE NOTICE 'Re-linked % stops to their per-city locations', v_relinked;
+
+  -- 3. Soft-delete any location that no longer has a non-deleted stop pointing
+  --    to it. (These are the over-grouped originals from the previous dedupe.)
+  UPDATE public.locations l
+  SET deleted_at = NOW()
+  WHERE l.brand_id = v_brand_id
+    AND l.deleted_at IS NULL
+    AND NOT EXISTS (
+      SELECT 1 FROM public.stops s
+      WHERE s.location_id = l.id
+        AND s.deleted_at IS NULL
+    );
+
+  GET DIAGNOSTICS v_soft_del = ROW_COUNT;
+  RAISE NOTICE 'Soft-deleted % now-orphan locations', v_soft_del;
+END$$;
+
+-- Final summary
+SELECT
+  (SELECT COUNT(*) FROM public.locations
+     WHERE brand_id = '64294306-5f42-463d-a5e8-2ad6c81a96de' AND deleted_at IS NULL) AS locations,
+  (SELECT COUNT(*) FROM public.stops
+     WHERE brand_id = '64294306-5f42-463d-a5e8-2ad6c81a96de' AND deleted_at IS NULL AND location_id IS NOT NULL) AS linked_stops,
+  (SELECT COUNT(*) FROM public.locations
+     WHERE brand_id = '64294306-5f42-463d-a5e8-2ad6c81a96de' AND deleted_at IS NOT NULL) AS soft_deleted;