Focusing closely on one loop can quietly erase awareness of everything happening elsewhere
Aliases: context loss during drill-down · industrial human factors
What it is
Context loss during drill-down occurs when close work on one component or loop removes awareness of parallel anomalies, propagation paths, or the operator's place in the larger system. This is not inattentiveness — it is a side effect of attending closely to one thing: attention and the visible display both narrow at the same time, and it is the combination that causes the problem.
Why it happens
Detailed work consumes visual attention and working memory. At the same time, hierarchical navigation physically removes unselected objects from the display. Neither alone is fatal — narrowed attention does not mean nothing is visible, and a narrowed display does not mean nothing is remembered — but together they produce tunnel vision: the relevant information is neither on screen nor being actively recalled.
There is a distinction easy to miss here. Whether putting the missing information back on screen — breadcrumbs, a global alarm count — actually fixes the problem depends on the operator's workload at the time. Under a self-paced, low-pressure drill-down, restoring visibility is usually enough, because there is still spare attentional capacity to glance at the periphery. Under high workload and time pressure, the attentional strategy itself starts filtering out information that is visible but judged irrelevant to the current sub-task, so a passive cue sitting in a corner of the screen gets ignored outright; only something that can break through the current task — sound, flashing, a forced overlay for cross-area alerts — has any effect. This is the condition that separates "a breadcrumb is enough" from "you need an active interrupt": the higher the workload, the smaller the marginal value of passive visibility.
Studying it
Run a multi-fault simulation that injects an independent second event while a participant is engaged in local diagnosis, and measure detection of that second event, time to detect, the return path back to the original diagnosis, and situation probes collected on a display freeze (a SAGAT-style freeze-probe approach). Hold local-task difficulty constant across conditions, or a drop in secondary detection could simply reflect a harder local task pulling attention away, rather than drill-down itself.
One variable worth isolating is how tightly the second event is coupled to the primary fault. A downstream alarm caused by the same fault is detected far more often than an unrelated, independent anomaly, because operators are already alert to propagation. Mixing coupled and uncoupled events in the same analysis flattens the result and hides which kind of information drill-down actually costs you.
Where it stops holding
Focus is not always the wrong call. A short, well-isolated maintenance or calibration task that demands high precision benefits from fewer distractions, and forcing a global panel to stay on screen for that kind of task is counterproductive.
The hazard from lost context is conditional on parallel risk actually existing, or on the local action being able to affect other areas. If the component being drilled into is genuinely isolated and cannot cascade, the consequence of context loss is limited, and the display does not need to be designed against the worst case.
Applying it
Keep breadcrumbs, a scope map, the global alarm count, and interlock status visible as a persistent strip or badge during drill-down, not tucked behind a page the operator has to actively switch to.
Let high-consequence, cross-area events break through the local view — a popup, a colour change, or sound should take priority over "there's a number ticking in the corner." Routine, low-workload anomalies can stay passively visible; forcing every event to interrupt turns the interrupt mechanism itself into a new source of distraction.
How to check: inject an independent event unrelated to the primary fault during a drill and reconcile the operator's acknowledgement time against the alarm log's event timestamp. If that gap is substantially longer than the same operator's undistracted baseline response time, the current cueing is not actually breaking through attention under load and needs a stronger interrupt, not more information on screen.