SCHEDULER_DISABLED=1 environment
variable (useful for multi-process deployments where only one instance should
run jobs). Per-shop failures are isolated: one store’s error never stops
another store’s run.
Schedule (all times UTC)
Timing expectations that follow from this schedule: insights can be up to 30 minutes behind, and a fresh checkout appears
in dashboards within about a minute thanks to the throttled order sync.
Insight generation
Every 30 minutes the app evaluates finding rules per store — high views with low cart, high cart with low purchase, low stock with high interest, best promotional window, restock suggestions, and related types. Each finding stores its evidence, the period it covers, and a suggested next step. The same insight type for the same product is suppressed for 24 hours so the feed stays fresh, and merchants can dismiss findings or trigger detection manually from the Insights page.Order catch-up sync
Webhooks are the primary order path; the sync is the reconciliation layer:- First run for a store: backfills roughly 90 days of orders through the Admin GraphQL API.
- Subsequent runs: fetches only orders updated since the last watermark, with a 1-hour overlap to heal edits and missed webhooks.
- Triggers: the 30-minute cron, plus a throttled check (at most once per minute per store) when the admin is open.
- Attribution limit: Admin API order queries expose no cart token, so synced orders count as orders but can never be attributed — attribution stays webhook-driven. The UI says so rather than implying coverage.
Rollups and retention
- The nightly rollup job aggregates recent days’ events and orders into per-product daily stats (views, sessions, cart events, orders, units, attributed orders, net revenue) with idempotent upserts, so re-runs heal instead of duplicating. Rollups are the long-term record: raw events are pruned after the retention window, rollups are not.
- The retention job deletes events, sessions, and orders older than the configured retention (default 90 days) and sweeps presence rows past a hard 7-day backstop. See Data model for the full lifecycle and deletion matrix.
- Billing housekeeping runs nightly: usage storage estimates are reconciled from real row counts, and plan status is re-verified against the Partner API so quota enforcement stays honest.