diff --git a/app/database/migrations/20260721000000_repair_orphaned_day_offs/migration.sql b/app/database/migrations/20260721000000_repair_orphaned_day_offs/migration.sql
new file mode 100644
index 00000000..fa8ad2c0
--- /dev/null
+++ b/app/database/migrations/20260721000000_repair_orphaned_day_offs/migration.sql
@@ -0,0 +1,139 @@
+-- Repairs day-off events that were miscategorized as freeform during the
+-- EventKind → template migration series on 2026-07-20, and backfills the
+-- hasConflict flag on any part/service-part assignment that overlaps a
+-- day-off owned by the same member (the miscategorisation had also hidden
+-- those overlaps from refreshConflictFlags).
+--
+-- Root cause: both 2026-07-20 migrations end with a "route orphans to
+-- freeform" fallback (20260720100000_drop_event_kind step 3 when the
+-- freeform template happens to exist, 20260720200000_seed_system_templates
+-- step 3 after it seeds one). Congregations that had never been seeded via
+-- `pnpm prisma db seed` had NEITHER system template, so
+-- drop_event_kind step 2 (off → day-off) was a no-op, its own step 3
+-- (any → freeform) was a no-op too, and the events kept templateId=NULL.
+-- The kindId column was dropped between the two migrations, permanently
+-- erasing the "this used to be an `off`" signal. seed_system_templates
+-- then seeded both templates and its step 3 defaulted every remaining
+-- orphan to freeform — that's the write that actually stamped the wrong
+-- template in production (`we can't tell which orphaned rows were
+-- originally day-offs vs. freeform-like, so route them all to freeform —
+-- the safest default`).
+--
+-- Symptom: legacy day-offs surface on /programs and are absent from
+-- /programs/days-off, because they now carry template.key = 'freeform'.
+--
+-- Recovery heuristic: a freeform event that (a) has no eventParts,
+-- (b) has no eventServiceParts, and (c) spans at least 20 hours is a
+-- day-off with overwhelming likelihood as of 2026-07-21. No in-app path
+-- produces a freeform event with startDate/endDate on different calendar
+-- days: handleFreeformMode (features/events/routes/programs/new.tsx)
+-- passes a single `date` twice to combineLocalDateTime before calling
+-- createFreeformEvent, so a well-formed interval is < 24 h. Day-offs
+-- from /me/days-off run through createDayOff; the form's end-date `min`
+-- attribute enforces endDate ≥ startDate + 1 day, though the enforcement
+-- is browser-only (createDayOff only rejects startDate > endDate), so a
+-- scripted POST could slip a same-day day-off past the server — the 20 h
+-- floor would then miss that row in the repair, accepted as a limitation.
+--
+-- One-shot repair, not idempotent for future writes: a legitimate
+-- multi-day freeform event created between two runs (e.g. an overnight
+-- assembly entered as freeform) would match the heuristic and get
+-- converted to a day-off. Postgres records this migration as applied
+-- after the first successful run so `prisma migrate deploy` will not
+-- re-execute it; do not manually re-run.
+--
+-- Assumes no in-flight imports: importEvents (features/settings/server/
+-- import-events.server.ts) writes events before importEventParts adds
+-- their parts, so a snapshot taken between the two would show a legit
+-- templated event matching the heuristic. Migration deploys are not
+-- concurrent with imports today.
+
+-- 1. Re-link freeform events that match the day-off shape.
+DO $$
+DECLARE
+  v_count int;
+BEGIN
+  UPDATE "Event" e
+     SET "templateId" = dayoff."id"
+    FROM "EventTemplate" freeform
+    JOIN "EventTemplate" dayoff
+      ON dayoff."key" = 'day-off'
+     AND dayoff."congregationId" = freeform."congregationId"
+   WHERE freeform."key" = 'freeform'
+     AND e."templateId" = freeform."id"
+     AND e."endDate" - e."startDate" >= interval '20 hours'
+     AND NOT EXISTS (SELECT 1 FROM "EventPart" p WHERE p."eventId" = e."id")
+     AND NOT EXISTS (SELECT 1 FROM "EventServicePart" sp WHERE sp."eventId" = e."id");
+  GET DIAGNOSTICS v_count = ROW_COUNT;
+  IF v_count > 0 THEN
+    RAISE NOTICE 'Day-off recovery: re-linked % freeform event(s) with the day-off template.', v_count;
+  END IF;
+END $$;
+
+-- 2. Backfill hasConflict on parts/services whose event overlaps a day-off
+--    owned by the same member.
+--
+--    While the day-offs above lived on the freeform template,
+--    refreshConflictFlags and checkDayOffConflict (features/events/server/
+--    event-part-assignments.server.ts) filtered them out via
+--    `NOT: { template: { key: 'day-off' } }`, so every overlap for a
+--    misclassified day-off was skipped and hasConflict stayed false when
+--    it should have been true. Re-linking the template (step 1) does not
+--    retrigger those refreshes, so the flags stay stale until the member's
+--    next day-off create/delete.
+--
+--    Runs against every day-off (not only the ones re-linked in step 1) —
+--    the invariant "any assignment overlapping a day-off must have
+--    hasConflict = true" is unconditional, and constraining to the
+--    just-repaired set would leave any pre-existing drift in place. Only
+--    flips false → true; a legitimately-set true from a co-participant
+--    conflict is never cleared.
+DO $$
+DECLARE
+  v_parts int;
+  v_services int;
+BEGIN
+  -- Comma-join in FROM (rather than explicit JOIN) so the target-table
+  -- alias `ep` remains reachable from the join predicate — Postgres does
+  -- not allow the UPDATE target to be referenced inside an explicit JOIN
+  -- clause. Same story for the EventServicePart update below.
+  UPDATE "EventPart" ep
+     SET "hasConflict" = true, "updatedAt" = NOW()
+    FROM "Event" dayoff,
+         "EventTemplate" doTpl,
+         "UserAccount" creator,
+         "Event" host
+   WHERE doTpl."id" = dayoff."templateId"
+     AND doTpl."key" = 'day-off'
+     AND creator."id" = dayoff."createdById"
+     AND host."id" = ep."eventId"
+     AND creator."memberId" IS NOT NULL
+     AND (ep."assigneeId" = creator."memberId" OR ep."assistantId" = creator."memberId")
+     AND host."startDate" <= dayoff."endDate"
+     AND host."endDate" >= dayoff."startDate"
+     AND host."id" <> dayoff."id"
+     AND ep."hasConflict" = false;
+  GET DIAGNOSTICS v_parts = ROW_COUNT;
+
+  UPDATE "EventServicePart" esp
+     SET "hasConflict" = true, "updatedAt" = NOW()
+    FROM "Event" dayoff,
+         "EventTemplate" doTpl,
+         "UserAccount" creator,
+         "Event" host
+   WHERE doTpl."id" = dayoff."templateId"
+     AND doTpl."key" = 'day-off'
+     AND creator."id" = dayoff."createdById"
+     AND host."id" = esp."eventId"
+     AND creator."memberId" IS NOT NULL
+     AND esp."assigneeId" = creator."memberId"
+     AND host."startDate" <= dayoff."endDate"
+     AND host."endDate" >= dayoff."startDate"
+     AND host."id" <> dayoff."id"
+     AND esp."hasConflict" = false;
+  GET DIAGNOSTICS v_services = ROW_COUNT;
+
+  IF v_parts > 0 OR v_services > 0 THEN
+    RAISE NOTICE 'Conflict backfill: flipped hasConflict → true on % event part(s) and % service part(s) overlapping a member''s day-off.', v_parts, v_services;
+  END IF;
+END $$;