🫀 System health
A nightly check of the core invariants — stock conservation, no negative balances, one group per property, and mis-mapped turnovers.
The System health page (under Settings, admin-only) runs an automatic nightly check of the rules the whole system depends on, and shows the result with a green OK or a red FAIL for each one. If anything fails, a banner also appears on the Dashboard so you notice it without opening this page.
What it checks
- Linen conservation — for every linen item and every bucket (On Shelf / In Vault / At Laundries / Coming back), the per-group tabs must add up to the Total. A mismatch means stock is attributed to a group that no longer exists, or a per-group split has drifted.
- Perishables conservation — the same, for consumables stock.
- No negative stock — no item’s tenant-total balance in any bucket is below zero (a single group may sit negative and is clamped for display; the aggregate never should).
- No impossible per-group negatives — flags any item whose *per-group* slice in a bucket is below zero. These are physically impossible (a group can’t hold minus-N pieces) and are hidden on the pages because the display clamps them to zero, so the total can look fine while a slice sits negative. Surfaced here so it can be cleared with the superuser Reconcile tools (which raise the slice to zero while keeping the tenant total unchanged). A warning, not a hard failure.
- One group per property — no apartment is in two property groups at once.
- No mis-mapped turnovers — when the booking system is reachable, no single listing resolves to two different apartments (the tell-tale of a bad merge). This one is *skipped*, not failed, when the PMS is unreachable.
- Backup & replication freshness — surfaces a silent off-site-backup failure, a missed daily backup email, or (on an active cluster node) a stale replication push.
Running it on demand
Click Run check now to run it immediately instead of waiting for the nightly run — useful right after a data fix. The page also shows a short history of recent runs. It is read-only: it only inspects your data, it never changes stock. On a passive cluster replica the button is unavailable (the standby is read-only) — run the check on the active node.