Schedule daily bounce->source reconcile (06:20 UTC, before scrub+builder)

infra/ansible/roles/mail-pipeline/tasks/main.yml

@@ -118,3 +118,4 @@
     - pw-hc-rampcap
     - pw-ip-rehab
     - pw-warmup-tg-alert
+    - pw-bounce-reconcile

infra/cron/pw-bounce-reconcile

@@ -0,0 +1,19 @@
+# Daily bounce → source reconciliation. Propagates every HARD bounce Listmonk
+# recorded back into the fmcsa_carriers source of truth (demotes the matching
+# smtp_valid/send_confirmed row to 'hard_bounced', which the builders + the
+# fuel-tax newsletter both treat as never-sendable), then rebuilds the
+# pw_smtp_valid_stage staging table the newsletter joins to so it only ever
+# contains still-live, non-blocklisted addresses.
+#
+# WHY: the audience queries already filter to the clean pool, but nothing
+# demoted an address AFTER it hard-bounced on a real send. Trucking mailboxes
+# churn fast, so the "clean" pool silently decayed to ~12% dead and the
+# "filtered" newsletters were still running 10-15% bounce, re-damaging sender
+# reputation. This closes that feedback loop so the clean pool self-heals.
+# (See scripts/reconcile_bounces_to_source.py + docs/email-deliverability-runbook.md.)
+#
+# Runs as deploy (needs docker exec for psql). 06:20 UTC — after the every-5-min
+# bounce-sync has ingested the overnight bounces, before the 06:30 UTC consumer
+# scrub and well before the 08:00 UTC trucking builder. Includes the healthcare
+# listmonk_hc bounces too so the HC pool stays clean as that stream scales.
+20 6 * * * deploy cd /opt/performancewest && python3 scripts/reconcile_bounces_to_source.py --days 90 --include-hc >> /opt/performancewest/logs/pw-bounce-reconcile.log 2>&1