An alarm system with more signals than anyone can remember stops meaning anything

What it is

The first two layers solve "can it be heard" and "can it be told apart"; this layer addresses a third problem: hearing an alarm and being able to tell it apart from others still doesn't guarantee remembering what it means. Once memory fails, the alarm collapses to "something happened," and the operator still has to check another display to find out what. This isn't an acoustic-design failure — it's a memory-capacity failure. The number of arbitrary sound-to-meaning mappings a person can reliably hold in stable long-term memory is limited, and once a system's alarm inventory exceeds that number, the extra alarms are nominally "discriminable" but nobody can reliably say what they mean.

Why it happens

Binding a sound to a meaning is an absolute identification task: rather than comparing two sounds side by side for a relative difference, a listener must hear one sound alone and name which category it belongs to. When people perform absolute identification along a single sensory dimension, the number of categories they can reliably maintain has a roughly fixed ceiling; beyond that ceiling, identification accuracy drops off systematically as the category count grows, rather than degrading gently. Alarm-meaning mapping is fundamentally this kind of task: the more alarm types there are, the fewer "memory slots" each meaning effectively gets, and because most alarms trigger infrequently in real use, mappings that go unused for long stretches decay over time, further shrinking the usable identification capacity — an alarm list memorized during training, most of it unencountered for months, is easily forgotten by the time it's actually needed.

Studying it

The absolute-identification paradigm applies directly: train listeners to associate a set of sounds with meanings, test identification accuracy once they reach a proficiency criterion, and increase the number of alarm types step by step to find the point where accuracy starts dropping sharply. A more ecologically valid version adds a retention interval — testing weeks or months after training, rather than immediately, to capture the effect of forgetting, since a capacity ceiling measured right after training runs noticeably higher than what actually holds up after a long stretch of disuse in a real work setting. The independent variables are the number of alarm types and the post-training interval; the dependent variables are identification accuracy and reaction time.

This method is used to set an evidence-based ceiling on how many alarms a system should carry, rather than letting the count grow unchecked as features accumulate.

Where it stops holding

• This capacity ceiling is not a fixed constant — it shifts with training intensity, with whether an alarm carries a semantic cue that makes its meaning easier to remember (semantic alarms outperform abstract ones here, a point the next leaf develops), and with individual differences; no single number transfers universally.
• A frequently triggered alarm can maintain high identification accuracy even in a larger set, simply through repeated exposure; the capacity bottleneck mainly bites on rarely triggered alarms that rely on long-term memory rather than recent practice.
• Providing a supporting cue (matching text or an icon shown on screen when the alarm sounds) can route around the limits of pure auditory memory, but at the cost of requiring a shift of visual attention to confirm — not always feasible in a scenario that demands keeping eyes elsewhere.

Applying it

• Set a ceiling on the total number of alarm types early in system design; when a new state is added beyond that ceiling, prefer folding it into an existing alarm category (same sound, different urgency encoding) over adding an entirely new alarm sound without limit.
• For infrequently triggered alarms, don't assume one round of training sustains identification ability indefinitely — schedule periodic refresher training or keep an easily accessible sound-to-meaning reference available.
• How to verify it: weeks after training, without advance notice, play the alarm list in random order and record the proportion of correctly identified meanings. If accuracy is noticeably lower than it was right after training, the current alarm inventory exceeds what this user population can sustain at the actual rate of real-world use.

Related

• Same group: A3.07.1 an alarm must be audible above the noise of the environment it will sound in · A3.07.2 distinct alarms must be discriminable from each other · A3.07.4 urgency can be encoded by faster tempo and rising pitch without changing timbre · A3.07.5 abstract versus semantic alarm sounds trade off learning cost against cross-language applicability · A3.07.6 simultaneous alarms mask each other and require a priority suppression policy · A3.07.7 standardized alarms improve cross-system recognition while custom alarms improve scene discrimination, and the two conflict
• Nearby: A6.02 working memory capacity · A6.06 the forgetting curve
• Search terms: absolute identification · alarm fatigue · alarm overload · channel capacity