Skip to content
Basin Thread

Alarms

Configuring alarm conditions, severity tiers, acknowledgment and state, escalation with on-call rotation, and using rationalization to remove noise.

Conditions

An alarm condition can be set on any measured or derived tag — thresholds, rates of change, or conditions derived from more than one point.

Conditions carry the same units discipline as the rest of the platform. A threshold is set in the tag’s engineering units, and a tag whose units change is flagged rather than silently re-evaluated against a number that no longer means the same thing.

Severity

Severity tiers separate an urgent condition from an informational one.

Severity is never communicated by colour alone. Every state carries a word and a shape in addition to its colour, so it reads correctly for a colour-blind operator, on a greyscale print, and on a phone screen in direct sunlight.

Red and amber are reserved for alarm and warning state throughout the platform. They are not used decoratively anywhere, so when something is red it is telling you something.

State

An alarm moves through four states:

State Meaning
Active The condition is currently true
Acknowledged Someone has seen it and taken responsibility
Cleared The condition is no longer true
Shelved Deliberately suppressed, with a reason and an expiry

Acknowledgment is attributed — who acknowledged it and when.

Shelving is deliberate and time-bounded. It requires a reason and it expires. A permanent mute with no record is how alarm systems rot, so the platform does not offer one.

Escalation and on-call

Call-out follows the on-call rotation rather than a static distribution list that went stale a year ago.

  • Routing follows the schedule, so the person actually on shift is the one called.
  • An alarm that goes unacknowledged escalates.
  • Acknowledgment works from a phone, because that is where the person on call is.

Rationalization

Alarm flooding is the single most common reason operators stop trusting a SCADA system. When everything alarms, nothing does.

The rationalization surface makes pruning the list an ordinary maintenance task with evidence behind it:

  • Which alarms fire most, ranked. A handful of conditions usually generate most of the volume.
  • Which are chronically unacknowledged. A condition nobody ever acts on is a strong candidate for removal.
  • Which fire in bursts. A flood usually has one cause, and finding it removes many alarms at once.

The decision to remove a condition needs history — how often did this fire, did anyone act on it, did it ever coincide with something that mattered. The platform keeps that history so the conversation is evidence rather than opinion.