Abstract
When two people revise the same sentence, a collaborative system has more work to do than preserve keystrokes. It must decide, display, or defer a relationship among edits made from different moments, different roles, and sometimes different theories of what the sentence should mean. A system may successfully converge on one document state while leaving the decisive editorial question unanswered: whose proposition is now being published, on what authority, and with what record of the alternative?
This paper calls the hidden rule set behind that outcome the editorial concurrency constitution. It is not a software specification. It is a way to make visible the assumptions a collaborative environment makes about textual scope, sequence, authority, visibility, reversibility, and escalation. Drawing on concurrent-editing architecture, primary documentation of suggested changes, workspace-awareness research, cognitive-load research, and professional editorial standards, the paper introduces the Editorial Conflict Assumption Ledger. The ledger allows a team to identify what a system has done to document state and what a responsible editor must still decide about the text’s public meaning.[1] [2] [3]
Working principle: A converged sentence is a state of the document. It is not, by itself, a settled editorial decision.
A clean merge can still make an unmade decision
Begin with a constructed scenario. It is deliberately small because the problem becomes easier to miss in a large document.
Base sentence: “The council will publish the report after review.”
Editor A, legal: “The council may publish the report after review.”
Editor B, communications: “The department will publish the report after review.”
Converged output: “The department may publish the report after review.”
The resulting sentence is grammatical. It might even be the best final wording. But no arrangement of characters can answer the questions created by the combination. Did the council delegate publication to the department? Is may the authorized legal formulation, or merely a cautious suggestion? Did Editor A intend the modal change to apply if the department, rather than the council, becomes the actor? Were the two changes meant to coexist at all?
It would be tempting to describe this as a merge failure. That diagnosis is too simple. A synchronization method can accomplish the technical objective it was given: make independently edited replicas converge on a shared document. In a CRDT account of collaborative text editing, each replica can edit independently and later incorporate information from the other; concurrent changes must nevertheless be ordered or otherwise represented in a way that reaches a common state.[1] The technical success is real. The editorial incompleteness is also real.
The paper’s problem is not to decide which editor should win. It is to teach readers to recognize the moment when a system’s successful operation has quietly crossed into an editorial decision. The key distinction is between document state and publication judgment. A document state says what characters, formatting, and structure the system presently represents. A publication judgment says whether that state accurately, responsibly, and authoritatively expresses what an organization intends to issue.
The editorial concurrency constitution
An editorial concurrency constitution is the set of rules—explicit, inherited, or concealed—by which a collaborative environment treats simultaneous changes to a text. It answers six questions that ordinary interface language often leaves unasked.
| Constitutional question | What a system or workflow must assume | What can go wrong if the assumption stays hidden |
|---|---|---|
| What is the textual object? | A character range, sentence, paragraph, citation, table cell, document version, or proposition is the relevant unit. | A system can correctly combine character changes while missing a collision at the level of claim, attribution, or commitment. |
| What counts as simultaneous? | Changes are concurrent, sequential-but-unseen, derived from different base states, or simply timestamped in an order. | Display order can be mistaken for causal order or editorial priority. |
| What does it mean to preserve intent? | The system preserves placement, operations, formatting spans, a user’s local action, or a more specific formal property. | Technical “intent preservation” can be confused with preservation of a writer’s or editor’s substantive purpose. |
| Who holds authority? | An owner, role, branch, approval path, or last writer can accept, reject, or supersede a change. | A privilege rule becomes an invisible theory of authorship and institutional responsibility. |
| Who sees the alternatives? | Differences remain visible to all collaborators, a subset, a reviewer, or no one once composed. | A consequential combination may appear as a neutral final text rather than a choice with alternatives. |
| When must human judgment re-enter? | The system composes, orders, preserves, blocks, or escalates a collision. | A workflow either creates needless review burden or silently settles a decision that requires accountable judgment. |
The word constitution is useful because it directs attention to governing rules rather than individual personalities. A well-designed constitution can support routine speed while reserving attention for high-consequence ambiguity. A poor one can make an arbitrary order look like authorized meaning, or can force every comma-level overlap into an exhausting ceremony.
The technical distinction matters. The Zed team’s description of a CRDT-based editor explains a common distributed problem: remote operations cannot be applied naively when concurrent local changes have shifted positions. Its implementation uses durable identifiers and anchors to resolve locations, and it orders same-location insertions consistently using logical timestamps and replica identifiers.[1] In that account, a stable order makes replicas agree. It does not say that the earlier-looking insertion has more editorial authority, that the displayed order expresses institutional priority, or that the combined sentence has received human review as a proposition.
A peer-reviewed rich-text CRDT paper makes the boundary sharper. Peritext observes that rich-text systems need an explicit model of intent preservation and that existing open-source implementations do not always preserve user intent. The paper then proposes a particular model with specified convergence, causality, and intent-preservation properties.[3] Those are substantial technical achievements. They remain properties of a defined model of concurrent text and formatting operations. They cannot decide whether a legal editor’s modal verb or a communications editor’s subject change should govern a public commitment.
Professional editing standards explain why that remaining task is not ornamental. Editors Canada defines stylistic editing as work to clarify meaning, ensure coherence and flow, and refine language. Its standards include resolving ambiguity, revising logical connections, reordering elements for coherence, and considering tone, audience, intended effect, and authorial goals.[4] In other words, a local textual operation may carry a judgment whose relevant unit is neither the character nor the cursor position. It may be the proposition, the reader’s inference, or the institutional promise.
Five ways a system can treat simultaneous change
There is no single virtuous response to concurrency. Systems and teams can choose among at least five broad treatments. Each delivers a different benefit and carries a different editorial risk.
| Treatment | What happens to document state | Primary gain | Editorial risk | Best question to ask |
|---|---|---|---|---|
| Automatic composition | Changes judged compatible are combined. | Routine work continues with little friction. | The combination can create an unreviewed proposition. | What new assertion exists only after the combination? |
| Deterministic ordering | Concurrent operations receive a stable order. | Replicas converge predictably. | System order can be mistaken for chronology, priority, or authority. | What does the displayed order mean to a reader? |
| Alternative preservation | Proposals, variants, or tracked changes remain available. | Reviewers can compare choices and recover context. | Low-value alternatives can overwhelm a reader. | Which differences need durable visibility? |
| Privileged selection | A role, owner, branch, or rule selects one outcome. | Responsibility and speed become clearer. | A power decision can be hidden in an apparently technical rule. | Is the privilege explicit, legitimate, and reviewable? |
| Human escalation | The system refuses silent settlement and asks for a decision. | Consequential ambiguity receives accountable attention. | Interruption can be slow, expensive, or performative. | What consequence justifies stopping routine flow? |
A documented suggestion interface shows that proposed change and final replacement can be kept distinct. Google’s public documentation says that a user may suggest edits without changing original text; the change becomes visibly marked, deletions appear struck through, comments may be attached, and the file owner can decide whether to retain the suggestion after receiving notice.[2] This is not an argument that every document should use one product or one approval hierarchy. It is evidence of a legible pattern: proposal, visible difference, context, notice, and an identified approval decision can be separated instead of silently fused.
The pattern is important because the alternatives in the opening scenario are not simply redundant text. Will and may create different degrees of commitment. Council and department distribute agency differently. Automatic composition preserves a possible final sentence, but it does not make either premise visible unless the environment preserves the changes or the reviewers inspect their combination. Deterministic order helps a system work; alternative preservation helps a person see what the system has worked with.
Scholarly editorial guidance provides a demanding, intentionally non-routine comparison. The Modern Language Association’s guidelines for scholarly editions frame reliable editing around accuracy, adequacy, appropriateness, consistency, and explicitness. They call for documentation of alterations and variants, clear rationale for emendation, and clear responsibility for accuracy.[5] A shared newsroom draft or internal policy memo does not need a scholarly textual apparatus. It may, however, need a lightweight version of the same insight: when a consequential text changes, the team should be able to say what alternative existed, who could decide, and why the final wording was chosen.
One collision under three editorial regimes
The council/department example becomes more instructive when the exact same character changes are placed under different editorial regimes. The physical operations remain identical. The required review path does not.
| Editorial regime | What is at stake | What automatic composition leaves unresolved | Appropriate ledger response |
|---|---|---|---|
| Legally controlled public statement | Institutional authorization, obligation, possible reliance by a public reader. | Whether the department has authority and whether may is the approved scope of commitment. | Preserve both changes, surface the combined proposition, and route a decision to a named policy or legal owner. |
| Routine communications draft | Clarity, consistency, voice, and accurate description of a settled policy. | Whether the wording reflects a known delegation decision or merely a communications preference. | Compose only if the underlying authority is settled; otherwise notify the accountable content owner with both premises visible. |
| Scholarly edition or critical text | Representation of variants, rationale for emendation, and future reconstructability. | Whether the output conceals a significant variant or unrecorded editorial intervention. | Retain variants and rationale at a level proportionate to the edition’s stated methodology.[5] |
The distinction does not mean that legal texts are always more important than other texts. A small wording change can matter greatly in a classroom policy, accessibility statement, news correction, grant report, or medical publication. Nor does it mean that every routine sentence must be routed through an owner. The point is proportionality: the response should be keyed to the consequence of the combined proposition, not merely to the fact that two cursors met at the same character range.
Workspace-awareness research explains why the presence of one shared document is not enough. Gutwin and Greenberg define workspace awareness as an up-to-the-moment understanding of another person’s interaction with a shared workspace: where they are working, what they are doing, and what they may do next. They describe its role in coordination, anticipation, discussion, and assistance, while noting that distributed groupware can provide far less perceptual information than a face-to-face workspace and may not present even the information it has.[6]
For editorial work, the lesson is practical. Editor A does not merely need to know that a character was changed. A may need to know that Editor B is changing the text’s institutional actor, that B has a communications mandate rather than legal authority, and that B expects a fast public release. Conversely, Editor B may need to know why A softened the modal verb. The absence of that awareness does not automatically require a live presence indicator or a flood of alerts. It does require the team to decide which contextual information must travel with a consequential change.
The Editorial Conflict Assumption Ledger
The Editorial Conflict Assumption Ledger is a decision sheet for one simultaneous editorial collision. It is not a conflict-resolution algorithm. Its job is to force an editorial team to put the system’s outcome and the team’s judgment into adjacent fields. A change is first described as an operation; only then is its possible premise, authority, consequence, and resolution recorded.
| Field | Reader records | Why the field matters | Example from the constructed collision |
|---|---|---|---|
| 1. Textual object / scope | Clause, sentence, paragraph, table, citation, or version state affected. | Scope may be conceptual rather than typographic. | “Subject and modal verb of a publication commitment.” |
| 2. Original proposition | The claim or commitment before either edit. | A team cannot judge a collision without a stated baseline. | “The council will publish after review.” |
| 3. Change A — observed operation | Literal insertion, deletion, move, or restyle. | Description should precede motive attribution. | “Will → may.” |
| 4. Change B — observed operation | The second change in the same disciplined form. | Both alternatives deserve independent description. | “Council → department.” |
| 5. Concurrency relation | Simultaneous, sequential-but-unseen, incompatible bases, or unknown. | Display order is not necessarily decision order. | “Independent changes from the same base.” |
| 6. System outcome | Composed, ordered, preserved, selected, or blocked. | The actual system treatment must be stated, not inferred. | “Both substitutions composed.” |
| 7. Editorial premise A / B | The concern each change appears to carry. | This exposes the substantive issue beneath the operation. | “Legal commitment” / “institutional agency.” |
| 8. Authority and decision owner | Role, mandate, approval right, or uncertainty. | Authority should not be deduced from cursor position or last modification. | “Legal owner for commitment; communications owner for public wording.” |
| 9. Visibility and notice | Who sees proposals, alternatives, and final combination; who is notified. | Awareness is a design choice with a cost. | “Both editors see their own changes; no review of the combined proposition is requested.” |
| 10. Reversibility and record | Undo ability, version history, retained alternative, reason, or loss. | Reversibility is valuable but is not the same as an accountable rationale. | “Source edits recoverable; decision reason absent.” |
| 11. Publication consequence | Low, medium, or high; name the relevant trust, factual, legal, attribution, or audience concern. | A uniform escalation rule wastes attention or hides risk. | “High: public authority and commitment change together.” |
| 12. Resolution action and rationale | Compose, preserve, select, or escalate; name who decides and why. | The final action becomes reviewable without requiring a full scholarly apparatus. | “Escalate to named policy owner; record delegation decision.” |
The ledger has two built-in restraints. First, it does not call every difference a conflict. A spelling correction and a replacement of a publication date may intersect in an interface but have radically different consequences. Second, it does not use the word authority as an excuse for opacity. A named decision owner may properly select an outcome, but the ledger still asks whether the alternatives, rationale, and consequence remain reconstructable enough for the task.
The instrument can also prevent a familiar category error: equating undo with accountability. A version history can allow a person to restore yesterday’s text. It may not reveal why the current wording was selected, who had delegated authority, or whether a reviewer ever saw the decisive combination. Reversibility is a safety feature. The record of an editorial decision is a different responsibility.
The cost of making every difference visible
The case for escalation needs a counterweight. A document in active collaboration can contain dozens of harmless overlaps, short-lived alternatives, and cosmetic changes. A system that interrupts every instance will displace the work it claims to protect.
Research on collaborative cognitive load offers a useful limit. Kirschner, Sweller, Kirschner, and Zambrano describe collaboration as involving transactive activities with inherent costs and argue that outcomes vary with factors such as group process, prior experience, and information distribution. They place those costs within the limits of working memory, especially where people must process novel, interacting information.[7] Their account concerns collaborative learning, not the publication workflow in this paper. It does not yield a numeric rule for alerts. It does support the modest inference that coordination is not free.
The answer is not to return to silent composition. It is to make the exposure threshold responsive to consequence. For low-consequence compatibility—two independent spelling corrections, for example—automatic composition or deterministic ordering may be better than a ritualized conflict screen. For a collision that changes a promise, attribution, evidence claim, identity, scope of authorization, or safety statement, the cost of a short, named review may be lower than the cost of publishing an unowned proposition.
Gutwin and Greenberg’s framework supports the same proportionality. It is not a demand to show every action all the time. It asks designers to consider what information matters, how it is gathered, how it is presented, and when it is useful in a particular task and group situation.[6] The editorial concurrency constitution should do the same: set visibility rules that are adequate to the textual risk rather than maximal by habit.
Exposure threshold: when must a human re-enter?
A team should surface a simultaneous change to a responsible human rather than silently settle it when one or more of these questions has a clear “yes.”
| Threshold question | What a “yes” means | Provisional response |
|---|---|---|
| Does the combined wording alter a commitment, authorization, attribution, evidence claim, safety statement, or identified public obligation? | The collision has consequence beyond mechanical text placement. | Escalate or preserve alternatives for a named decision owner. |
| Do the two changes carry different editorial premises rather than merely different mechanics? | The system is combining judgments about meaning, not just operations. | Show the combined proposition and state both premises. |
| Is responsibility ambiguous or hidden by system order? | Convergence may be mistaken for authorized choice. | Make the authority path and delegation explicit. |
| Would silent composition prevent a later reviewer from reconstructing a consequential decision? | The document state lacks sufficient rationale or retained alternatives. | Keep a compact decision record and a reversible path. |
If none of these questions is true, automatic composition may be entirely appropriate. The ledger is not a campaign against automation. It is a method for deciding where automation ends. The result is not a universally correct workflow or a preferred collaboration product. It is a legible constitution: a statement of what the system may settle, what the team must see, who may decide, and why the final sentence has earned the authority to stand.
References
[1]: https://zed.dev/blog/crdts Zed Industries, “How CRDTs Make Multiplayer Text Editing Part of Zed’s DNA,” Zed Blog, 1 December 2022.
[2]: https://support.google.com/docs/answer/6033474?hl=en Google, “Suggest Changes to a File,” Google Docs Editors Help, accessed 19 August 2026.
[3]: https://dl.acm.org/doi/10.1145/3555644 Martin Kleppmann et al., “Peritext: A CRDT for Collaborative Rich Text Editing,” Proceedings of the ACM on Programming Languages, 2022.
[4]: https://editors.ca/publications/professional-editorial-standards/stylistic-editing/ Editors Canada, “C. Standards for Stylistic Editing,” Professional Editorial Standards 2024, accessed 19 August 2026.
[5]: https://www.mla.org/Resources/Guidelines-and-Data/Reports-and-Professional-Guidelines/Guidelines-for-Editors-of-Scholarly-Editions Modern Language Association, “Guidelines for Editors of Scholarly Editions,” last revised 4 May 2022.
[6]: https://grouplab.cpsc.ucalgary.ca/grouplab/uploads/Publications/Publications/2002-DescriptiveFramework.JCSCW.pdf Carl Gutwin and Saul Greenberg, “A Descriptive Framework of Workspace Awareness for Real-Time Groupware,” Computer Supported Cooperative Work 11 (2002): 411–446.
[7]: https://pmc.ncbi.nlm.nih.gov/articles/PMC6435105/ Paul A. Kirschner, John Sweller, Femke Kirschner, and Jimmy Zambrano R., “From Cognitive Load Theory to Collaborative Cognitive Load Theory,” International Journal of Computer-Supported Collaborative Learning 13 (2018): 213–233.