Abstract
A small-screen edit is often reported as a single failure: I could not move the sentence, the selection kept slipping, or dragging text is broken. Such reports may be true as experience, but they are too compressed to teach from. They hide a sequence of separate demands: finding the target, establishing the intended span, seeing enough surrounding context, invoking an operation, locating the changed text, and recovering a relation when its outcome is uncertain. This paper offers precision debt as a diagnostic construct for that sequence. Precision debt is the additional motor, visual, contextual, and recovery work that accumulates when an editor must make a relation exact before the textual consequence of an edit can be inspected.
The paper’s instrument, the Precision Friction Field Card, records one editorial objective, target relation, selection action, occlusion/context condition, movement operation, error/recovery path, alternative input route, observed burden, research-needed question, and mandatory non-conclusion. It is not a product score, accessibility audit, clinical assessment, or universal theory of touch. It gives a writing instructor, UX researcher, quality-assurance lead, or editor a way to move from an undifferentiated complaint to a bounded question: where did this edit begin to demand more precision, memory, or recovery work than the textual relation itself required?
Scope
This is a teaching instrument for analyzing an observed or reconstructed editorial micro-interaction. It does not compare products, rank platforms, certify accessibility conformance, diagnose motor or cognitive conditions, or assume that a difficulty belongs either to the user or to the software. Its constructed incidents are illustrative, not empirical studies. The research record is used within its stated settings: controlled target-acquisition tasks, mobile text-selection and correction prototypes, a study of expert shortcuts, a practitioner account of testing, and official web-content guidance. The Field Card is an original Wexley construct that keeps a diagnosis provisional.
1. A sentence moved; a relation did not
The editor is not trying to write a paragraph. The paragraph already exists, and that is the problem.
At 4:47 p.m., between two short trains, she opens a proof on her phone to make what appears to be a contained correction. A sentence beginning “The reduction may be temporary” sits near the end of a paragraph. The word may matters. It keeps the sentence attached to a footnote that reports a provisional agency estimate rather than a settled fact. A later paragraph gives the political response. The editor decides the qualified sentence belongs there instead: evidence first, response second.
The desired revision is ordinary: select the sentence, move it, keep its footnote, and inspect the transition.
What happens is easy to dismiss. The first long press selects one word. Two handles appear at a line break. While extending the span, the editor’s finger covers the phrase she must preserve and a menu replaces the intended action. She starts again, cuts the full sentence, loses its marker when the page jumps, then pastes it into a plausible sequence.
The sentence has moved. The relation has not yet been verified.
The footnote may have travelled. It may remain above. It may now appear to support a neighboring claim. The move might be semantically right and editorially wrong; it might be mechanically complete but evidentially unresolved. The editor cannot responsibly report “the edit is done” until she has reconstructed what changed. By then, the task is no longer a simple sentence move. It has become a chain of target finding, selection, menu state, displaced context, citation checking, and recovery.
This paper calls the additional work in that chain precision debt. Precision debt is not a property of a person’s fingers, a device size, or a platform’s reputation. It is the extra motor, visual, contextual, and recovery work that accumulates when an edit requires a relation to become exact before its textual consequence can be inspected. It begins when a writer or editor must do more than make the intended change. They must also discover whether the change landed where it was meant to land, whether the surrounding material still says what it said, and how to get back if it does not.
The term is deliberately not a score, and a small screen does not automatically create it. A practiced shortcut or alternate input route may change a cumbersome selection path. Zhao and colleagues, for example, evaluated a smartphone prototype that combined approximate touch, speech, and language context so that an editing or correction command did not depend on a precisely established text span in the same way as a touch-only technique.[1] That counterexample matters because it blocks a lazy conclusion: an observed friction is not necessarily inevitable.
Still, an alternative does not erase the need to see where a burden enters. Text selection is often described as one gesture. In practice, it may include activation, first and second endpoint adjustment, view preservation, command invocation, and post-operation location. Rivu and colleagues describe a baseline touch-selection path with activation followed by adjustments to start and end anchors in their study of a gaze-supported alternative.[2] Their study does not prove that gaze is a general answer. It makes a more durable lesson available: a short intention can contain several interaction commitments.
2. The revision path is not one action
An editorial instruction normally names a textual outcome: delete the duplicate phrase, move the sentence, preserve the qualification, correct the citation, restore the omitted word. The interface does not receive that outcome. It receives a sequence of contacts, pauses, drags, releases, menu choices, visual shifts, and possible reversals. If the result is uncertain, the editor has to return through part of the same sequence while remembering what the original relation was supposed to be.
The distinction can be made visible by separating five questions.
| Path question | What the editor is trying to establish | A possible precision burden | What a reported difficulty does not establish |
|---|---|---|---|
| Can I acquire the target? | A caret location, handle, word, sentence, marker, or command control. | Repeated aiming, adjacent target conflict, or a target obscured by the contacting finger. | That the whole document is unusable or that the editor lacks skill. |
| Can I form the intended selection? | The exact start and end of a meaningful text span. | Activation ambiguity, endpoint drift, selection expansion, or a span that is not visible together. | That a touch system cannot select text at all. |
| Can I keep the relation in view? | The selected words and their surrounding paragraph, note, heading, version, or citation relation. | Occlusion, reflow, scroll displacement, menu coverage, or loss of nearby evidence. | That smaller displays necessarily destroy context. |
| Can I carry out the operation? | A move, cut, copy, paste, replacement, formatting action, or alternate command. | Unexpected mode, command location, focus change, or an operation whose visible effect is ambiguous. | That direct manipulation is always the best or worst route. |
| Can I recover and verify? | A return to a known relation after uncertainty, interruption, or apparent error. | Reselecting, searching, comparing, undoing, retyping, or deferring the check. | That the existence of undo removes the cost of detecting what needs undoing. |
These questions are not a universal procedure. A word can be easy to touch while difficult to select as a span; a span can be accurate while its footnote disappears; a move can execute while its qualification remains uncertain. Even undo requires the editor first to notice what needs undoing.
Research on finger input supports keeping at least the first question distinct. Bi, Li, and Zhai propose an extension of Fitts’ law for finger touch and report experiments in one-dimensional and two-dimensional target acquisition and touchscreen keyboard typing.[3] Their contribution here is not a formula for a manuscript. It is a methodological warning: finger acquisition has its own variables and should not be casually treated as identical to conventional pointing, text selection, or editorial judgment.
Parhi, Karlson, and Bederson make a related distinction at the task level. Their two-phase study of one-handed thumb input on a small touchscreen separated discrete single-target work from serial keypad work, varied target size and location, and recorded time, errors, hit positions, and subjective comfort.[4] The hardware and setting are historically specific: the study used an HP iPAQ, right-thumb interaction, and carefully delimited target tasks. It cannot tell us how a current editor will move a citation-bearing paragraph. It can teach something more precise: even before a revision reaches selection or rearrangement, the physical acquisition condition depends on the kind of action and its location.
That matters for language. “Touch is imprecise” is too large to be useful. A target can be reachable while a selection remains ambiguous. A selection can be accurate while its context disappears. A move can occur while the evidence relation remains in doubt. The Field Card therefore records an observed burden at the point where it occurs, rather than assigning an overall verdict to the phone, the application, or the editor.
3. Precision debt is a diagnostic, not a trait
Precision debt has four related forms. They should be read as possible locations of burden, not as a typology of people.
Acquisition debt appears when the editor must spend additional actions locating or reaching the point where work begins. It can involve a small or crowded control, a text location that is difficult to distinguish under a finger, a handle that competes with nearby text, or a view in which the intended marker is no longer adjacent to the visible paragraph. Acquisition debt concerns the start of the action. It does not by itself say whether a selected phrase will be correct.
Selection debt appears when the intended linguistic unit is not the unit the interaction readily offers. A writer may need a clause while the system initially selects a word; an editor may need a sentence while the visible endpoints sit across a line break; a revision may concern two nonadjacent passages while the default selection model presumes one contiguous span. The word selection hides those differences. Huan, Tu, and Li’s Text Pin record is useful because it compares endpoint tapping with adjustment-handle dragging and reports faster performance with comparable selection accuracy in its studied Android comparison.[5] The record does not settle the general design question. It demonstrates that the relation between intended endpoints and finger movement is open to design.
Context debt appears when the exact operation displaces the material needed to judge it. The problem may arise through finger occlusion, magnification, a pop-up menu, scroll movement, reflow, or the simple fact that an editorial relation crosses more than one visible area. Context debt is not the same as reading difficulty. A paragraph can be legible while the editor lacks a simultaneous view of the paragraph, the relevant footnote, and the destination where a sentence is being placed.
Recovery debt appears when uncertainty has entered and the editor must restore a relation before proceeding. Undo is one recovery path, but not the only one. The editor may reselect the original text, search for a phrase, compare versions, place a temporary bracketed note, retype a passage, move to another surface, or leave a visible unresolved question. Scott Jenson’s practitioner account of mobile editing is not a general performance study, but it is valuable as a QA vocabulary: it distinguishes cursor placement, handles, selection extension, menus, and workarounds such as retyping.[6] A Field Card can use such language to identify what was seen without claiming how often it happens.
The debts can compound: endpoint adjustment may hide a source marker, and a completed move may land out of view. The aim is not to total them but to preserve the moment when a relation stopped being easy to inspect.
4. The Precision Friction Field Card
The Precision Friction Field Card records one gesture path associated with one editorial objective. It is most valuable immediately after a report such as “I lost the sentence,” “the marker stayed behind,” or “I kept selecting the wrong part.” The record can be completed by an editor reflecting on a task, a writing instructor planning a mobile revision exercise, a UX researcher preparing a study, or a quality-assurance lead turning an informal report into a reproducible question.
| Field | What to record | Prompt | Boundary control |
|---|---|---|---|
| Editorial objective | The textual consequence sought: correct, qualify, relocate, preserve a citation, restore, or verify. | “What needs to be different in the document?” | Do not record “edit text.” |
| Target relation | The relation that must survive: clause-to-claim, sentence-to-footnote, paragraph-to-evidence, or version-to-approval. | “What could be lost even if the visible operation completes?” | Do not assume a visual move preserved the relation. |
| Selection action | The observed path: tap, press, endpoint adjustment, drag, keyboard route, voice-plus-touch, or another named action. | “How was the span or insertion point made?” | Do not call a gesture difficult without observation or a stated source. |
| Occlusion/context condition | What became hidden, compressed, reflowed, off-screen, or nonadjacent during the action. | “What must remain visible or recallable to judge this edit?” | Do not infer context loss from display size alone. |
| Movement/rearrangement operation | Cut/paste, drag, command, mode shift, replacement, or alternate route. | “What state-changing action was actually invoked?” | Do not treat direct manipulation as the only legitimate route. |
| Error/recovery path | Undo, reselect, search, comparison, temporary marker, retype, transfer, defer, or explicit unknown. | “If the relation is uncertain, how will it be reconstructed?” | Do not assume the possibility of undo cancels recovery cost. |
| Alternative input route | A shortcut, hardware input, assistive route, voice-plus-touch, larger context, or no known route. | “What comparison would make this burden testable?” | Do not claim the alternative is available or accessible without evidence. |
| Observed burden | Repeat targeting, endpoint drift, lost context, unexpected mode, recovery loop, or no burden observed. | “What happened before interpretation became uncertain?” | Do not label the editor or assume root cause. |
| Research-needed question | A bounded question for observation, QA, or study. | “Under what condition does this burden recur or disappear?” | Avoid questions that presuppose the conclusion. |
| Non-conclusion | What the record cannot show. | “This does not establish whether…” | Mandatory. The Card is incomplete without it. |
The final field changes the instrument’s ethical character. “The editor missed the handle” can become a judgment about dexterity; “the product failed accessibility” can become a conclusion without a conformance method. The recorder must instead state what remains unobserved: This does not show whether the selection path is difficult for all editors, whether an alternate keyboard route is available, or whether the interface fails an accessibility standard.
The Card is not a defect ticket. It begins before expected behavior, reproduction conditions, and severity can be responsibly specified, preserving the editorial aim and observed uncertainty.
5. Three comparative field readings
The following records are constructed examples, not assessments of named systems or accounts of real people. They demonstrate why one generic category—“mobile editing”—cannot teach the differences among a word correction, a sentence move, and a citation-adjacent restoration.
| Field | Case A: correct one repeated word | Case B: relocate a qualified sentence | Case C: restore a note relation after paste |
|---|---|---|---|
| Editorial objective | Delete a duplicated word in a short paragraph. | Move a qualified sentence below a related paragraph. | Determine whether a footnote still supports the intended claim. |
| Target relation | Word-to-neighboring words. | Sentence-to-paragraph sequence and sentence-to-footnote. | Footnote marker-to-provisional claim. |
| Selection action | Place caret near duplicate; delete locally. | Activate selection; establish two sentence endpoints; cut or move. | Locate marker, claim, and destination after the operation. |
| Likely context condition | Nearby wording remains visible. | Selection handles, menu, line break, reflow, and off-screen destination may split the relation. | Claim and note may not be visible together. |
| Recovery path | Undo or retype the local word. | Search for the sentence, compare the original location, restore a marker, or defer to a second review. | Compare the source claim and marker, restore relation, or flag uncertainty. |
| Provisional observed burden | None, or local acquisition/placement burden. | Selection and context debt may compound before the sentence is moved. | Context and recovery debt may remain after a visible move succeeds. |
| Research-needed question | Does local caret placement recur as a problem under stated text size and input conditions? | At which step does sentence selection or verification become uncertain: endpoint setting, command, reflow, or destination check? | What view or alternate route permits claim and note relation to be checked together? |
| Non-conclusion | This does not prove short corrections are always easy. | This does not prove paragraph moves are impossible on a small screen. | This does not prove the note was lost or that a different surface would solve the problem. |
Case A may end with no meaningful precision debt. That is an important result. The instrument is not designed to find friction everywhere. It is designed to prevent an isolated awkward gesture from standing in for the entire practice of editing.
Case B is the paper’s anchor because the textual relation is more demanding than the movement itself. To relocate the sentence responsibly, the editor must preserve a qualification, recognize a citation relation, and inspect a new transition. The physical path may be perfectly manageable while the verification path remains strained. WEXLEY-WP-013 asks whether a task relation fits a surface within a wider writing ecology. WEXLEY-WP-017 narrows the unit: where, inside one exact revision, does precision debt begin to accumulate?
Case C makes the difference between a completed command and a completed editorial task explicit. If the claim and its marker cannot be judged together, the editor does not yet know whether the revision is complete. The appropriate next step may be recovery; it may also be deferral. Deferral is not failure when it is recorded as an intentional handoff rather than silently mistaken for completion.
6. Alternate paths are evidence, not escape hatches
A precision-debt diagnosis becomes unfair if it treats a default touch path as the only imaginable path. The point of recording an alternative route is not to tell every editor to acquire another device, master a shortcut, or disclose a disability. It is to create a comparison question: does the burden belong to the textual objective, the current interaction path, the available context, or an untested combination of them?
Fuccella, Isokoski, and Martin compared a gestural technique drawn on a soft keyboard with a widget-based smartphone technique for caret movement, selection, and clipboard control. They report a performance benefit for their gestural implementation across their evaluated font-size conditions.[7] The paper does not make a universal case for gestures. It makes a narrower design point: caret movement, selection, and command access can be arranged through a different interaction area, and that arrangement can be evaluated.
Le and colleagues offer a second counterpressure. Their work begins with shortcut use among 15 expert hardware-keyboard users over five workdays and interviews about text-heavy activity, then moves through gesture elicitation and a prototype for fully touch-sensitive smartphones.[8] Their results should not be turned into a claim that every phone should have rear or side gestures. The prototype has specific hardware assumptions, and expert shortcut behavior is not an ordinary-phone baseline. Yet the study helps the reader see that “precise editing” has more than one possible grammar. Caret movement, selection, clipboard access, and navigation can sometimes be approached as alternate commands rather than as repeated direct targeting.
Zhao and colleagues extend that comparison across modalities. Their studied prototype uses touch to indicate an approximate location while speech and language context contribute to inferring an intended edit or correction.[1] This is not a recommendation to use speech in public, a claim about privacy, or a way to avoid later editorial verification. It is evidence that a system can choose to tolerate approximate location rather than requiring exact span selection at the outset.
An alternate path can introduce new burdens: a shortcut may be undiscoverable, a gesture may conflict, a hardware route may be unavailable, and speech may be unwanted or inaccessible. The Card does not rank those trade-offs; it records an alternative only as a research comparison and declares what was not tested.
7. When a report becomes a research question
The Field Card becomes useful when it changes the next action. A vague report may need no escalation if the operation was completed, verified, and not burdensome. It may need a local recovery if the relation is known but temporarily displaced. It may need targeted QA if the same burden can be reproduced under stated conditions. It may need accessibility investigation when a specific input route, control, or user-agent behavior is implicated. The correct response depends on what the Card records—not on whether the report sounds dramatic.
| If the record shows… | The next useful action is… | The record must still not claim… |
|---|---|---|
| A one-time uncertain move with a visible recovery route | Preserve the objective and recovery step; defer or repeat only if the relation cannot be verified. | A general usability defect. |
| Repeated inability to establish the intended span under specified conditions | Reproduce the path with the same text relation, context condition, and action sequence. | That all users experience the same failure. |
| An apparent conflict between a selection path and an alternative route | Compare the routes while recording availability, learnability, and verification consequences. | That the alternative is superior or accessible for everyone. |
| A question about target size or control spacing | Determine whether the item is actually an applicable target-size control and preserve relevant exceptions. | Native-app or product conformance. |
| A move that completes without relation verification | Create a second-check or handoff step for the claim, citation, version, or surrounding sequence. | That visible completion equals editorial completion. |
W3C’s explanation of WCAG 2.2 Success Criterion 2.5.8 is useful precisely because it restrains misuse. It addresses pointer-target size and spacing, but it includes exceptions for inline targets and user-agent-controlled targets; it also treats an editable area used to position a cursor as one spatial target.[9] That guidance can help a reader ask whether a target-size question is structurally applicable. It cannot certify a native text-selection interaction, establish the accessibility of a named editor, or explain a particular person’s experience.
The escalation discipline is modest: record the objective and relation, reproduce only the relevant path, identify what left view, record recovery and alternatives, then state what remains unknown. A sound question may be: When a qualified sentence crosses a line break, does endpoint adjustment obscure the footnote marker before or after selection is established? Unlike phone editing is impossible, it can be observed and compared.
8. Limits of the instrument
The Precision Friction Field Card does not measure editing quality, productivity, disability, expertise, or accessibility. It cannot decide whether a writer should use a phone, tablet, keyboard, voice interface, assistive technology, or another surface. It does not turn an editor’s report into an empirical result, and it does not replace usability research, accessibility testing, or an application’s documented behavior.
Its focus is intentionally narrow. It does not analyze the broader allocation of writing tasks across surfaces; that is WEXLEY-WP-013’s task–surface question. It does not argue that direct manipulation is inherently good or bad; that conceptual inquiry belongs to WEXLEY-WP-037. It does not map all interface complexity, all network constraints, all physical device limits, or all privacy conditions. Its unit is one editorial relation under one attempted path of exact selection, movement, and recovery.
The narrowness is a protection. A small-screen failure can too quickly become evidence of poor design, a user’s supposed limitation, platform inadequacy, or a general theory of mobile work. The Card interrupts that speed by recording what was preserved, attempted, uncertain, and still unproven.
9. Conclusion: name the debt before assigning blame
A sentence can move while its qualification, note, evidence relation, or sequence remains unresolved. That difference is easy to miss because the visible operation has finished. Precision debt names the work that begins after the interface has apparently done what was asked: the work of regaining a target, reconstructing a span, restoring context, finding a recovery route, and verifying that the textual relation still holds.
The right conclusion is not that small screens cannot support serious editing. Nor is it that any editor who completes a visible action has overcome the relevant problem. The reader’s task is more exact. Before assigning failure to a person, a platform, or a gesture, ask: Where did this edit begin to demand more precision, memory, or recovery work than the textual relation itself required?
References
[1] Zhao, Maozheng, Wenzhe Cui, IV Ramakrishnan, Shumin Zhai, and Xiaojun Bi. 2021. “Voice and Touch Based Error-tolerant Multimodal Text Editing and Correction for Smartphones.” Proceedings of the 34th Annual ACM Symposium on User Interface Software and Technology (UIST ’21). https://doi.org/10.1145/3472749.3474742
[2] Rivu, Radiah, Yasmeen Abdrabou, Ken Pfeuffer, Mariam Hassib, and Florian Alt. 2020. “Gaze’N’Touch: Enhancing Text Selection on Mobile Devices Using Gaze.” CHI Conference on Human Factors in Computing Systems Extended Abstracts. https://doi.org/10.1145/3334480.3382802
[3] Bi, Xiaojun, Yang Li, and Shumin Zhai. 2013. “FFitts Law: Modeling Finger Touch With Fitts’ Law.” Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/2470654.2466180
[4] Parhi, Pekka, Amy K. Karlson, and Benjamin B. Bederson. 2006. “Target Size Study for One-Handed Thumb Use on Small Touchscreen Devices.” MobileHCI ’06. https://www.microsoft.com/en-us/research/wp-content/uploads/2006/01/parhi-mobileHCI06.pdf
[5] Huan, Wei, Huiying Tu, and Zhicong Li. 2019. “Enabling Finger Pointing Based Text Selection on Touchscreen Mobile Devices.” Proceedings of Chinese CHI ’19. https://doi.org/10.1145/3332169.3332172
[6] Jenson, Scott. 2023. “The Invisible Problem.” https://jenson.org/text/
[7] Fuccella, Vincenzo, Poika Isokoski, and Benoît Martin. 2013. “Gestures and Widgets: Performance in Text Editing on Multi-touch Capable Mobile Devices.” Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/2470654.2481385
[8] Le, Huy Viet, Sven Mayer, Maximilian Weiß, Jonas Vogelsang, Henrike Weingärtner, and Niels Henze. 2020. “Shortcut Gestures for Mobile Text Editing on Fully Touch-Sensitive Smartphones.” ACM Transactions on Computer-Human Interaction 27 (5): Article 33. https://doi.org/10.1145/3396233
[9] World Wide Web Consortium. 2023. “Understanding SC 2.5.8: Target Size (Minimum) (Level AA).” Web Content Accessibility Guidelines (WCAG) 2.2. https://www.w3.org/WAI/WCAG22/Understanding/target-size-minimum.html