Integrations

Per-provider state, from what was last recorded — never from a live call.

Always 200 unless our own side is broken. A customer's expired token is their integration being red, not our service being down, and a monitor that went red for it would be red permanently and therefore ignored.

That paragraph was true and the code did not honour it. The top-level status carried the worst CUSTOMER-side state, so one connection needing a reconnect reported the whole product as down — measured on 2026-09-23, when it did exactly that and the uptime monitor called HeyMetra DOWN over a single WooCommerce connection that was waiting to be re-entered.

The same promise had a second half it was not keeping: "always 200 UNLESS our own side is broken". A side that could not read its own state answered 200 with the word unknown in the body, so an uptime check had to be configured to read the body to notice — and a check that needs the right keyword configured is a check that is one edit away from watching nothing. It refuses with 503 now, which is what /health/ready above already does and what its own test says out loud: an orchestrator reads the status code, not the word in the body.

So the two are separate now. status is about US and takes one of two values: ok, or unknown when this endpoint cannot read its own state. connections is about the customers, in the same vocabulary as the per-provider map below, so nothing is lost by the split.

Deliberately carries no counts, no account ids, no names and no figures. The endpoint is unauthenticated, like the probes above, and "how many Shopify stores are connected" is nobody's business but ours.