diff --git a/services/hasura/migrations/1741520400001_fix_teams_name_unique_constraint/up.sql b/services/hasura/migrations/1741520400001_fix_teams_name_unique_constraint/up.sql
index 36eee6f3..41622c65 100644
--- a/services/hasura/migrations/1741520400001_fix_teams_name_unique_constraint/up.sql
+++ b/services/hasura/migrations/1741520400001_fix_teams_name_unique_constraint/up.sql
@@ -6,10 +6,59 @@
 -- Application code (deriveTeamName, findTeamByName) already normalizes
 -- names to lower(trim(name)) before insert/lookup, so a plain constraint
 -- is sufficient and Hasura-compatible.
+--
+-- Idempotent across every possible starting state:
+--   (a) expression index teams_name_unique exists          → drop it, add constraint
+--   (b) plain UNIQUE constraint teams_name_unique exists   → no-op
+--   (c) neither exists                                     → add constraint
+-- Required because an earlier version of this migration blindly ran
+-- `DROP INDEX IF EXISTS` followed by `ADD CONSTRAINT`, which fails with
+-- "cannot drop index because constraint requires it" whenever the constraint
+-- had already been created (locally or via partial reruns). The pattern below
+-- lets `hasura-cli migrate apply` succeed on every environment regardless of
+-- prior state, including environments where a human manually added the
+-- constraint after the first attempt failed.
+
+-- Normalize any names that slipped through without lowering/trimming.
+-- Skip rows whose normalized form would collide with an existing row — the
+-- constraint below is case-sensitive on the exact stored string, so leaving
+-- those rows un-normalized is safe: they coexist with the normalized
+-- counterpart and the application only ever writes the normalized form.
+-- Blindly normalising would hit a duplicate-key error on the UPDATE itself.
+UPDATE public.teams t
+   SET name = lower(trim(t.name))
+ WHERE t.name <> lower(trim(t.name))
+   AND NOT EXISTS (
+     SELECT 1
+       FROM public.teams u
+      WHERE u.id <> t.id
+        AND u.name = lower(trim(t.name))
+   );
+
+DO $$
+BEGIN
+  -- Case (b): constraint already present, nothing to do.
+  IF EXISTS (
+    SELECT 1
+      FROM pg_constraint
+     WHERE conname = 'teams_name_unique'
+       AND conrelid = 'public.teams'::regclass
+  ) THEN
+    RETURN;
+  END IF;
 
--- Normalize any names that slipped through without lowering/trimming
-UPDATE public.teams SET name = lower(trim(name)) WHERE name != lower(trim(name));
+  -- Case (a): expression index exists without a constraint backing it.
+  -- DROP it so we can cleanly add the constraint below.
+  IF EXISTS (
+    SELECT 1
+      FROM pg_indexes
+     WHERE schemaname = 'public'
+       AND indexname = 'teams_name_unique'
+  ) THEN
+    EXECUTE 'DROP INDEX public.teams_name_unique';
+  END IF;
 
--- Drop the expression index and replace with a plain unique constraint
-DROP INDEX IF EXISTS public.teams_name_unique;
-ALTER TABLE public.teams ADD CONSTRAINT teams_name_unique UNIQUE (name);
+  -- Cases (a) and (c): add the plain constraint.
+  EXECUTE 'ALTER TABLE public.teams ADD CONSTRAINT teams_name_unique UNIQUE (name)';
+END
+$$;