The paragraph that no longer knows where it was
Constructed opening claim. “A notification broke my flow.”
The sentence conveys a real frustration. It also compresses several events into one verdict. What was the writer doing at the moment of interruption: generating a sentence, locating an argument, comparing sources, repairing a citation, reading a comment, deciding a structure, or waiting for an editor? What kind of cue arrived? Was it ignorable, mandatory, relevant, expected, self-initiated, or socially difficult to defer? What remained visible when the writer returned? And what does “flow” mean here: speed, continuity, confidence, memory for the next step, pleasure, or something else?
Those questions do not make an interruption harmless. They make it researchable.
The temptation to solve the problem with one interface slogan is strong. “Turn off notifications.” “Protect deep work.” “Let writers choose.” Each phrase may point toward a useful experiment. None should substitute for an account of the task, interruption, return cue, and writer’s actual experience. A visible alert can be disruptive, but it can also carry a time-sensitive editorial request. A silent badge can avoid an immediate break while building an unseen backlog. A self-initiated pause can restore context for one writer and fragment a difficult revision for another.
Keus van de Poll and Sörqvist provide unusually direct evidence from word-processed writing. In their experiments, participants wrote stories and sometimes shifted to a secondary task; writing speed was lower immediately after interruption and recovered over a short interval in that study. [1] The study does not measure all writing, all notifications, or a writer’s creative state. It makes a narrower lesson available: return is a distinct part of the interruption event.
This paper introduces the Return Cue Ledger, an original Wexley teaching instrument that records one episode of sustained writing or revision, the interruption condition, the cue left behind, the return route, and the smallest testable change. It does not diagnose attention, set a notification policy, score a writer, claim medical benefit, or promise a productivity result. It asks a more practical question: what would help this person return to this work, under these conditions, without pretending that all interruption is the same?
“Flow” needs a task before it can be interrupted
“Creative flow” can sound precise while naming almost nothing. In a manuscript, a writer may mean rapid composition, a held image, a source relationship, a structural plan, a revision decision, or a feeling of unbroken engagement. These are not interchangeable states. An interruption that occurs during a one-minute proofreading check may have a different consequence from one that arrives while a writer is holding several paragraph-level relationships in mind.
Altmann, Trafton, and Hambrick studied short interruptions in a sequence task and found higher sequence-error rates after even brief interruptions in their controlled setting. [2] The task was not writing, and the result should not be recast as a statement about imagination. Its relevance is structural: a task position can be lost. A writer returning to a draft may need to know not only the next word, but the next unresolved claim, the current source, the alternative being weighed, or the editorial reason for a change.
This suggests a different opening question:
What representation of the unfinished task will still be available after the writer has attended to something else?
The representation might be a highlighted clause, a short next-step note, an open source, a visible comment, a mark in an outline, or a consciously completed stopping point. It might also be absent. The Ledger does not prescribe any one representation. It records whether one existed, whether it helped, and what should be tested next.
The interruption is not only the alert
An interruption can be a sound, banner, badge, message, meeting, person, device vibration, software prompt, system warning, physical need, or self-initiated detour. It can be expected or surprising. It can be relevant to the work or entirely unrelated. It can require an action or simply compete for attention. A ledger that calls all of these “notifications” loses the condition it hopes to improve.
Ohly and Bastin’s field experiment examined work interruptions caused by communication applications and found that reducing such interruptions was beneficial for performance and strain in that defined setting, with individual factors such as fear of missing out and telepressure moderating the result. [3] The study does not authorize a universal rule for authors. It matters because it refuses a simple mechanism: the effect of a notification is connected to what the interruption means to the person and workplace.
Research beyond writing complicates the picture further. Chen and colleagues found resumption and cumulative-interruption effects in a working-memory task. [4] Chevet and colleagues’ ecological study of digital reading found varied types of attentional disruption and did not yield a single uniformly negative result across interruption categories. [5] Neither source proves a pattern for writing. Together, they give the Return Cue Ledger a discipline: record type, timing, task state, value, control, and return condition before claiming that a particular cue “helps focus.”
Return cues should be concrete, not inspirational
A return cue is any intentionally available trace that helps a writer identify the unfinished work after an interruption. It is not necessarily a visual mark, and it is not necessarily sufficient. A cue may be:
- a brief note that names the next argument rather than merely saying “continue”;
- a selected phrase that shows where a revision stopped;
- an outline item that identifies the next structural choice;
- an open comment that states an editor’s question;
- a source citation kept beside the sentence it supports;
- a low-risk checkpoint made before leaving a difficult edit; or
- a conscious statement that no cue was left and that the next action is uncertain.
The cue’s value depends on the task. A highlighted word can be enough for correcting punctuation but insufficient for recovering the reason a paragraph was moved. An outline may preserve a structural goal while leaving a sentence-level wording problem unresolved. The instrument records the gap rather than pretending that a cue preserves the whole mental state.
Xue and colleagues report that interruption source and value type influenced different phases of an interruption process in a writing-task experiment. [6] The details cannot be generalized into an authoring rule. They support a useful design hesitation: the same visible cue may need a different treatment depending on whether it is an urgent editorial request, a planned break, or a social message unrelated to the manuscript.
The Return Cue Ledger
The Ledger is a non-validated Wexley teaching instrument. Its unit is one observed or constructed interruption episode, not a person’s capacity for focus or a product’s quality.
| Field | Record this | Do not infer |
|---|---|---|
| 1. Work boundary | The writing or revision task, document state, time window, and intended next action. | A general measure of creativity or productivity. |
| 2. Active task representation | What the writer was holding or using: clause, source, outline, comment, comparison, or decision. | That the representation remained in memory after interruption. |
| 3. Interruption condition | Source, modality, expectedness, relevance, urgency, and whether response was required. | That every alert of the same form has the same effect. |
| 4. Control and deferral relation | Who initiated it, what choice to defer existed, and what social or task constraint limited that choice. | That the writer freely chose the timing. |
| 5. Return cue | Visible or nonvisual trace left before or after the interruption. | That a cue restores all context. |
| 6. Return route | What the writer did to resume: reread, inspect outline, re-open source, ask, restart, or abandon. | That return time alone measures the cost. |
| 7. Observed consequence | A specific result: mistaken sequence, rereading, delay, changed wording, relief, or uncertainty. | A causal conclusion from one episode. |
| 8. Counterweight | The interruption’s possible relevance, required coordination, or value to the project. | That relevance eliminates disruption. |
| 9. Small test and non-conclusion | One reversible change to try and what the Ledger cannot decide. | A notification policy, diagnosis, or universal design verdict. |
Fields 2 and 5 are the center of the instrument. Without them, a writer can report an interruption but not identify what the interface or workflow needed to preserve. A helpful cue is not necessarily sophisticated. It only has to name the unfinished task better than a blank return.
A constructed revision episode
The following example is fictional. It does not report a study result or a real writer’s behavior.
Initial claim: “Message banners destroy revision.”
Bounded episode: A fictional editor is reorganizing two paragraphs in a long essay. The editor is comparing an author’s claim with a source note and has not yet decided whether the source belongs before or after a counterexample.
| Ledger field | Constructed record | What remains open |
|---|---|---|
| Work boundary | One revision decision between two paragraphs in a 90-minute editing session. | Any claim about the editor’s general focus. |
| Active representation | Source relation, proposed order, and a counterexample still to be placed. | Whether the editor had all relevant context in view. |
| Interruption condition | A banner announces a request from the managing editor; it appears relevant but not visibly urgent. | Actual urgency and whether response was required. |
| Control and deferral | The editor can defer the banner but worries about a deadline. | Whether the social pressure is justified. |
| Return cue | No note marks the unresolved paragraph order; the source remains open. | Whether a brief next-step note would have been enough. |
| Return route | The editor rereads both paragraphs and the source note before deciding. | Whether rereading caused an error or merely a short delay. |
| Observed consequence | The editor spends several minutes reconstructing the comparison and feels unsure of the prior option. | A causal result or a general banner effect. |
| Counterweight | The message may concern a deadline that matters to the same essay. | That relevance makes the interruption harmless. |
| Small test | Before responding to a nonurgent banner, write one sentence naming the unresolved structural choice. The Ledger cannot decide to disable all banners. | A toolwide configuration or policy. |
The constructed example does not prove that the cue will work. It makes a precise test possible. If the editor tries the note in three comparable episodes, the observation can become richer without being converted into a universal theory of flow.
A notification setting is only one part of the return environment
It is tempting to treat the interruption as a setting that can be switched on or off. Settings matter, but a return environment also includes document design, task scale, social expectations, device arrangement, editor practice, and the writer’s available cues.
For example, a writer can suppress banners and still lose the next task because an outline does not show what remains unresolved. A visible alert can be deferred while an unread message becomes an increasingly distracting obligation. A team can agree that routine questions wait for a scheduled review window, but then need a clear route for a genuine deadline or safety concern. A cleaner interface can help while a document’s own complexity still creates a difficult return.
The Ledger locates the intervention at the actual point of loss. If the active representation was a source comparison, the relevant repair may be a short source-linked note rather than an audio setting. If the interruption was required because the writer is an editor for others, the relevant repair may be a clear escalation label rather than suppression. If a writer was simply between tasks, an elaborate cue may be unnecessary.
The question is not how to eliminate all visible demands. It is how to make the state of unfinished work recoverable when a demand must be attended to.
Observe one condition before telling a story about a person
The word “distractible” is easy to attach to a person after a difficult session. It is rarely useful for improving a writing environment. A Ledger entry should describe an episode: one task, one interruption type, one return cue, and one observed consequence. It should not label a writer or infer a stable trait.
This matters because an identical interface can be experienced differently on different days and projects. A writer revising a familiar page may return easily with a simple selection. The same writer may be constructing a new argument from scattered sources and need a much richer cue. A notification that is welcome during copyediting can fracture an interview transcript review. The instrument’s scope prevents the design team from treating an observed episode as an explanation of a person.
When an observation repeats, the next research question can become more careful rather than more totalizing. Does a specific kind of interruption arrive during a particular kind of work? Does a visible next-step note improve the return route? Do writers describe a social pressure that makes a nominal deferral setting meaningless? These questions can inform a small design trial. They cannot be answered by a universal slogan about focus.
Use the Ledger without turning it into surveillance
The Ledger belongs to the person or team trying to improve a bounded workflow. It should not become a log of every pause, a measure of worker obedience, or evidence that an editor has failed to respond quickly enough. If a team needs to learn from patterns, it can use aggregated, voluntary, purpose-limited notes and state what will not be recorded.
The instrument’s final field is a protection against overreach. The next result might be, “Try marking the live decision before opening the editorial inbox for one week,” or “Ask whether this alert class needs a visible urgency label.” Neither asks a system to monitor private thought. The work of writing deserves conditions that support return, not a new layer of compulsory attention measurement.
Different writing operations leave different return problems
Drafting a sentence, revising an argument, fact-checking a name, responding to a comment, and formatting a citation do not leave the same kind of unfinished state. A sentence draft may leave a word-level choice. A structural edit may leave a hierarchy and a reason for moving it. A source check may leave a question about evidence and scope. A review comment may leave a social commitment as well as a textual task.
This is why a single “save my place” feature is not enough to explain return. A cursor position can preserve where the writer was, but not what the writer was deciding. An automatically reopened tab can restore a source without marking which claim required it. An outline can restore a sequence while hiding a sentence-level uncertainty. Each can be useful; none is a substitute for asking what representation the current task needs.
The Ledger makes this difference available without demanding a detailed annotation at every pause. A writer can use a lightweight cue when the task is simple and a richer one only when a decision is genuinely vulnerable to loss. Over time, the notes may show that certain operations—perhaps structural revision or source comparison—benefit from a named cue more than others. That is a local workflow finding, not a theory of all writers.
The practical design implication is modest: interfaces should help people leave and recover a task representation at several levels of scale. They should not assume that the latest cursor location is the same as the next meaningful action.
Do not turn a ledger into a productivity regime
A return cue can be helpful and still become coercive if it is transformed into a demand for constant self-tracking. The Ledger should be used lightly, for one bounded question. It is not a system for measuring every pause, assigning blame, or optimizing an author’s life.
The same caution applies to design. “Minimize interruption” can hide a crucial editorial dependency. A fact-checker’s urgent question may need a visible path. A writer may need a deliberate stop to consult a collaborator or return to the physical world. The counterweight field requires the paper’s own bias to remain visible: uninterrupted time is valuable for some tasks, but not the only value in an editorial practice.
The most appropriate intervention is often small and reversible: change a cue, group a class of low-value alerts, put the current decision in an outline, pause after a natural sentence boundary, or create a separate channel for genuinely time-sensitive coordination. None is justified merely by the language of “flow.” Each has to be tried in a stated task and checked against an observed return.
Make the return visible
Return to the opening claim: “A notification broke my flow.” After the Ledger, the sentence can become:
During a paragraph-order revision, an apparently relevant banner arrived before the editor had recorded the unresolved comparison. On return, the editor reread the source and both paragraphs; a one-sentence decision cue is a small next test. This record does not prove a general notification effect.
The revision is not less human because it is more specific. It gives the writer, editor, or interface designer a place to act: preserve an unfinished decision, distinguish a required message from a merely visible one, and leave a return path without pretending to remove every interruption from creative work.
References
[1] Marijke Keus van de Poll and Patrik Sörqvist, “Effects of Task Interruption and Background Speech on Word Processed Writing,” Applied Cognitive Psychology 30 (2016): 430–439, https://doi.org/10.1002/acp.3221.
[2] Erik M. Altmann, J. Gregory Trafton, and David Z. Hambrick, “Momentary Interruptions Can Derail the Train of Thought,” Journal of Experimental Psychology: General 143 (2014): 215–226, https://doi.org/10.1037/a0030986.
[3] Sandra Ohly and Luca Bastin, “Effects of Task Interruptions Caused by Notifications from Communication Applications on Strain and Performance,” Journal of Occupational Health 65 (2023), https://doi.org/10.1002/1348-9585.12408.
[4] Yue-Yuan Chen et al., “The Effect of Task Interruption on Working Memory Performance,” Human Factors 66 (2024): 1132–1151, https://doi.org/10.1177/00187208221139017.
[5] Guillaume Chevet et al., “What Breaks the Flow of Reading? A Study on Characteristics of Attentional Disruption During Digital Reading,” Frontiers in Psychology 13 (2022), https://doi.org/10.3389/fpsyg.2022.845532.
[6] Xue et al., “Interruption Value Type and Source Matter in Different Phases of an Interruption Process: Emotional/Cognitive Impact and Neural Evidence,” Frontiers in Neuroscience 16 (2022), https://doi.org/10.3389/fnins.2022.985778.