Operational runbooks

The authoritative runbooks live in the backend repo at trile/docs/runbooks/ (next to the code and migrations they reference). This page indexes them so on-call engineers know what exists.

Runbook	Use when
Balance reconciliation (balance-reconciliation.md)	A reconciliation run reports drift between a wallet balance and the sum of its ledger. Covers running the script, reading the output, and when to escalate.
DLQ inspection (dlq-inspection.md)	Jobs are landing in the BullMQ dead-letter queue. Covers inspecting, diagnosing, and recovering failed jobs.
Manual webhook replay (manual-replay.md)	The automatic relay poller missed outbox rows, or delivery is degraded. Covers inspecting and replaying failed webhook deliveries by hand.
Backup & restore (backup-restore.md)	Validating DB backups, point-in-time recovery, and verifying health after a restore.

Keep them in the backend repo

Runbooks reference scripts, queue names, and table structures that change with the code. Keeping them in trile/docs/runbooks/ lets them version alongside what they describe. This page is just the directory — update the source files, not this list, when a procedure changes.

Related concepts

• The wallet/ledger model the reconciliation runbook acts on: The wallet model.
• The outbox/delivery design behind webhook replay: Backend architecture.
• Queue/process roles (DLQ, scheduler vs. worker): Backend architecture.