Signal clinics for overnight cohort jobs

Overnight jobs feel magical until they are not. A signal clinic is deliberately boring: we read job schedules aloud, note expected latency, and compare that to what operators actually need before morning sends.

We document null paths with examples, not theory. A null playbook entry should say, "If billing event is null for more than 36 hours, downgrade to stage X," with an owner and an escalation path. Without that specificity, nulls become shrug emojis in Slack threads.

Drift watchlists are seasonal. Retail teams care about holiday skew; SaaS teams care about release windows. We ask each table to name one seasonality risk they are ignoring because it is inconvenient. Those risks go on the watchlist even if no action exists yet — acknowledgment matters.

The clinic ends with a pilot monitoring checklist taped to the runbook metaphorically. If a checklist never gets opened, that is signal too: either the job is stable or the team stopped caring. Both are worth knowing.