Interruption timing should be chosen at subtask boundaries

What it is

The same interruption costs different amounts depending on when it happens. Placed at a subtask boundary — the gap right after one small piece of work has finished and before the next has begun — resumption cost is lowest; placed in the middle of a subtask's execution, resumption cost is clearly higher. This is about why a boundary is cheaper — a conclusion that follows directly from what resumption cost is made of, not a method for how to identify boundaries in practice.

Why it happens

The bulk of resumption cost comes from rebuilding the goal state that the interruption cut off: which sub-steps remain undone, where the last one left off, where intermediate results were stored. A subtask boundary is cheap precisely because almost none of this needs rebuilding at that point: the previous subtask has already delivered its result in full, with no half-finished intermediate state left to maintain, and the next subtask hasn't started yet, so there's no invested-but-incomplete progress either. An interruption landing in this vacuum finds nothing suspended in working memory that needs protecting; resumption only requires confirming "which subtask comes next," with no need to rewind and piece together a state that was cut off mid-way.

If the interruption lands inside a subtask, the situation is entirely different: that subtask's intermediate state — how far through a comparison it had gotten, what step of a calculation it had reached — is at that moment held entirely by working memory. Once the interruption occurs, that state either gets overwritten by the interrupting task or diluted by working memory's limited capacity, and resumption must rebuild that intermediate state before continuing — this is exactly the most expensive part of resumption cost. A subtask boundary is what eliminates that reconstruction.

Studying it

This is commonly studied by extending the interruption-resumption paradigm, specifically manipulating where the interruption falls relative to task structure: a multi-step task is divided into subtasks, and interruptions are placed either at the boundary between subtasks or midway through one, with all other conditions (interruption duration, content) held constant. Post-resumption reaction-time cost and error rate are then compared across the two placements.

Common independent variables: the interruption's position relative to subtask boundaries, and the length and complexity of the subtasks themselves. Common dependent variables: the post-resumption reaction-time increase, error rate, and whether errors involving a misremembered intra-subtask state occur.

This manipulation is commonly used in HCI to provide direct evidence for when a system should surface an interruptive prompt — comparing the cost of a boundary interruption against an interruption at an arbitrary moment quantifies how much resumption cost "choosing the right moment" actually saves.

A methodological caution: this finding only answers "why a boundary is cheaper," not "how to detect a boundary during real use." Real tasks rarely have completion markers as clean as a lab task's, and identifying a boundary in practice is a separate problem that needs its own treatment.

Where it stops holding

• This assumes the "subtask" itself has a clear completion marker. If there is no well-defined completion point between steps (continuous free-form writing, for instance), the cost gap between boundary and non-boundary placements narrows, because it becomes hard to define what counts as a genuine gap.
• The boundary's cost advantage applies only to the "rebuilding goal state" portion. The other part of resumption cost — remembering that an interrupted secondary task is still owed — does not go away just because the interruption happened at a boundary; that cost is independent.
• Over-relying on boundary interruptions erodes their protective advantage. If a user gets interrupted after completing every single subtask, they end up constantly cycling between "finish—restart," and even though each individual boundary interruption is cheap, the accumulated cost can still be substantial.
• This answers only why a boundary is cheaper; how to actually locate a boundary in a real task flow, and what granularities of boundary exist, are questions that need separate treatment.

Applying it

• When designing a feature that actively interrupts the user (a pop-up prompt, a forced notification display), prioritize detecting whether the user is currently in a gap between two subtasks, rather than firing at an arbitrary moment — even a simple check for "is there an unsubmitted action or an actively edited field right now" filters out most high-cost mid-task interruptions.
• For step-based task flows (forms, wizard-style interactions), treat the completion of each step as a potential safe interruption point, since a user's intermediate state at those points is naturally "empty" and needs no extra protective mechanism.
• When it's impossible to tell whether the user is at a boundary, prefer delaying the interruption over forcing it in at an uncertain moment — the cost of delay is usually smaller than the reconstruction cost caused by interrupting mid-subtask.
• How to check: record the difference in reaction time and error rate when the same type of interruption occurs at a boundary versus mid-subtask. The larger the gap, the more the current system needs to improve how it chooses interruption timing.

Related

• Same group: A5.08.2 The system should preserve the state and position from before the interruption · A5.08.3 A measurable resumption lag exists before performance returns to pre-interruption levels · A5.08.4 Resumption cost has two components — re-executing the primary task and remembering the interrupting task — that must be accounted for separately · A5.08.5 Self-initiated interruptions have lower resumption cost than externally imposed ones · A5.08.6 The more information an interruption shares with the primary task, the more reconstruction work resumption requires
• Nearby: A5.14 Interruption timing and break points (boundary granularity, notification batching, and do-not-disturb periods as concrete timing strategies)
• Search terms: subtask boundary · natural breakpoint · interruption timing · task resumption