Ninety days per service, and every incident written up afterwards - including the ones where nobody noticed. An incident you cannot read about is an incident you have to take on trust.
Any period where a service did not do its job for a customer, including slow. A board that takes eleven seconds to load is not up, whatever the health check says.
Admins are notified by email for outages affecting their workspace, and the board carries a banner for the duration. We would rather tell you twice than have you discover it.
Agents queue their reports locally and send them on reconnect, so an intake or dashboard problem does not create gaps in your inventory history.
A queue consumer stopped acknowledging messages after a deployment, and intake backed up rather than failing. No mail was lost - every message in the backlog was processed, and tickets were created with their original received timestamp so response timers were not affected.
The deployment check now verifies consumer acknowledgement rather than process liveness, which is what made this look healthy for the first forty minutes.
An upstream messaging provider outage stopped WhatsApp messages reaching us. Email and web form intake were unaffected, and we notified every workspace with WhatsApp intake enabled within nine minutes.
Messages sent during the window were delivered by the provider on recovery and became tickets normally. We have since added a board banner so the channel shows as unavailable rather than quiet.
A database index rebuild ran during business hours in the ap-south region, making board queries take between four and eleven seconds. Agent sync and intake were unaffected, so no data was delayed - only reading it was unpleasant.
Index maintenance is now restricted to the regional maintenance window, and long-running operations are capped.
A DNS change on our intake domain propagated unevenly and one large mail provider deferred delivery. Mail was deferred, not bounced, so everything arrived once the record settled.
Intake DNS now has a longer TTL and changes go through a staged rollout with per-provider delivery checks.
Uptime is measured from outside our network against the same endpoints customers use, at one-minute intervals, from three regions. Scheduled maintenance inside a published window is excluded; everything else counts.
$infotechbang init --trial 14d --card none