Keeping Connections Healthy

How breakage presents

Almost never as an error message in front of a user.

It presents as absence. The calendar looks quiet. The inbox has no new mail. The sheet sync shows no changes. A scheduled run produces a thin, generic output. The agent says it could not find anything.

Every one of those is indistinguishable from 'nothing happened', which is why a broken connection can persist for days. The distinguishing question is never about the data — it is about the connection's own status and its last successful sync timestamp.

So the habit is: when something looks quiet, check whether it is connected before you conclude it is quiet.

Alerts, and the one-shot marker

Kavanah detects broken connections and alerts on them, in-app and by email.

One mechanical detail is worth knowing because it affects how you respond. The broken-connection alert is deliberately a one-shot: it fires once per breakage rather than repeatedly, so a persistently broken integration does not become a daily nag that everyone filters.

The consequence is that missing the first alert means not being reminded. So the alert is worth acting on when it arrives rather than filing, and it is worth making sure those emails reach a person rather than a shared address nobody opens.

The distinction that matters

This is the subtle part, and it is worth understanding because getting it wrong produces exactly the wrong action.

Some failures are yours: your grant expired, your password changed, your admin revoked the app. The fix is for you to reconnect.

Some failures are on the operator side: the platform's own credential for a service, a provider outage, an API change. Nothing you do to your connection helps, because your connection was never the problem.

These can look identical — both produce 'this integration is not returning data'. And treating the second as the first has a real cost: it marks healthy connections as broken and sends people 'action needed: reconnect' emails about a credential they do not hold. That has actually happened, and it burns the one-shot alert that a genuine breakage would have needed.

So Kavanah is careful about which failures are allowed to mark a connection broken. As a user, the practical version is: if you reconnect and the symptom persists identically, stop reconnecting. The problem is not your credential, and repeating the fix will not change that.

A maintenance habit

Monthly, and it takes five minutes.

Open Integrations and look at the status on each card. Anything not showing healthy gets reconnected now rather than when someone notices.

For the connections that back a sync — sheet sync especially — check the last successful sync timestamp rather than the status alone. A connection can be nominally fine while the sync it powers has stopped.

And for large data sources, spot-check completeness occasionally. A sync can report success while having fetched only part of a large calendar or mailbox, and the symptom is data that looks thin rather than data that looks broken.

Whoever owns the workspace should own this. It is small, it is boring, and it prevents the specific failure where a feature everyone relies on has quietly not worked for a fortnight.