It is 23:00 and you are the only nurse covering all 16 monitored beds tonight. Every bed reports to a central station that fires an alarm the moment a reading crosses a threshold.
Each alarm gives you two choices: Dismiss it on the spot, or Respond by walking to the bed to check the patient yourself. Walking takes about 8 seconds, and while you are gone new alarms keep arriving and stacking up behind you.
Somewhere in tonight's stream, one patient is genuinely deteriorating. Everything else is noise: a loose probe, a lead that slipped, a pump that cleared itself, a battery running low. Your job is to catch the one signal in the noise, without stopping for every alarm that fires.
Round 1 runs the ward as it is tonight. Round 2 runs the exact same night through a redesigned alarm system. Watch what changes.
Round 1 · The Ward As It Is
Walking to Bed …
New alarms keep firing while you are away.
8s
End of Round 1
Shift Change
The same night is about to run again, through a redesigned alarm system: duplicates suppressed, non-actionable conditions filtered, alarms tiered by severity, and readings correlated across parameters instead of fired one at a time.
The Debrief
Two Wards, One Patient
Bed 12 deteriorated on both nights, in exactly the same way. What changed between rounds was not your attention, it was the system deciding what deserved it.
The science
Clinical studies estimate that 72 to 99% of alarms in monitored hospital units are false or non-actionable (AHRQ, 2020).
The Joint Commission's Sentinel Event Alert #50 (2013) linked 98 alarm-related events, including 80 deaths, to alarm fatigue between 2009 and 2012.
Boston Medical Center cut its weekly cardiac monitor alarms from 87,823 to 9,967 (an 89% reduction) by redesigning thresholds and adding severity tiers, with no increase in adverse patient outcomes.
The redesign, not extra vigilance, is what let you hear Bed 12 in Round 2: cutting the volume of noise mattered more than trying harder to listen through it.
This demo is a scripted, self-contained browser simulation. Nothing you type is sent to an AI model and no live AI system runs behind it, even where it plays one. Its code and copy were built with generative AI (Anthropic’s Claude) and reviewed by Robert Barcik, who is responsible for what is published (LearningDoe s.r.o.). Disclosed in the spirit of Article 50 of the EU AI Act.