The Maker's Eye in Designed Systems
A prose revision, a comparison screen, and the limits of borrowing from literary craft
By Evelyna Bellamy
“An essay and an interface both select what a reader encounters first, but their purposes can conflict. I revise a sentence from this archive and examine the Perspective Lab’s existing comparison hierarchy, then work through a declared error-message example. Literary attention supplies questions for design; it does not replace evidence about usability or system behaviour.”
Editorial essay by Evelyna Bellamy; not externally peer reviewed. 1102 words. Estimated at 200 words per minute for the abstract and main prose; references and interactive studies take additional time. Publication details revised 10 October 2026.
Revised 2026-10-10. Developed an actual before/after manuscript revision, a source-based reading of the existing Perspective Lab and a labelled constructed interface example. Claims of direct transfer to AI explainability and orchestration were removed.
I. A sentence that hides its own choice
The earlier manuscript of this essay made a large promise in a small sentence: “This mental operation applies identically whether the substrate is an essay by Joan Didion, a Cloudflare Pages deployment pipeline, or a multi-tenant database schema.” The sentence names three impressive objects and gives the reader no operation to inspect. Its confidence comes from the word identically, precisely where its explanation is missing.
I would replace it with this: “A paragraph and a screen can both withhold the fact a reader needs next. Revising them begins by identifying that fact and deciding where it belongs.” The revised sentence names a choice that can be demonstrated. It gives up a universal claim about equivalence in exchange for a narrower question about sequence. This is an actual revision of the archive’s previous text, not a quotation from Didion or a report of a design experiment.
One alternative would preserve the old sentence and add definitions of substrate, mental operation and architecture. That might make its vocabulary clearer without making its argument stronger. I prefer to supply the missing action before supplying more terminology. The rejected alternative shows why revision is more than simplifying a word: it can change what the paragraph is prepared to claim.
This is what I mean here by a maker’s eye. Read for a choice that produced an effect, then try a different choice on the same material. The effect might be emphasis, omission, pace or a clearer next action. The question remains useful even when the original maker’s intentions are unavailable, provided we identify our reading as interpretation.
II. An interface you can read for structure
Narratologium’s existing Perspective Lab provides a small interface to inspect. It introduces one event, identifies its accounts as an illustrative teaching scenario, offers two perspective selectors and then aligns what each account foregrounds and leaves outside its frame. The people, quotations and figures are constructed examples. They should be read as teaching material rather than historical testimony.
The hierarchy makes a decision before the reader makes one: establish the shared event, then compare accounts. That ordering gives each account a reference point. Placing every quotation above the event description would offer voices sooner, but the reader would have to infer what they were describing. Both arrangements are possible; the current one spends its opening space on orientation.
The comparison also exposes a limit of its own instrument. Its “Show only observations unique to each account” control compares exact stored strings. Two differently worded statements could therefore remain visible even if they describe similar observations. It is a display filter, not a semantic proof that the accounts disagree. Naming that limit prevents the shape of a comparison from claiming more than its operation establishes.
Reading the screen this way does not demonstrate that its present hierarchy is best for every visitor. A usability claim would need observed tasks and responses. What we can identify from the artifact is the order, the labels, the distinction between source and teaching material, and the operation of its controls. That gives us a reasoned hypothesis to test, rather than a verdict dressed as a measurement.
III. What to reveal first
Consider a constructed file-upload message: “Processing failed. Request exception 42. Your submission is being handled.” This example belongs to no claimed deployed product. Suppose the file was never saved. The message makes the reader decode an internal condition while leaving the most consequential fact uncertain. It may also encourage the person to wait for a submission that does not exist.
A proposed revision would begin: “Your file was not saved.” It would then identify the known cause in ordinary language and give an available next action. If the system only knows that the connection failed, it should say that rather than inventing a file-format diagnosis. “Try again” belongs in the message only when retrying is actually available and its consequences are understood.
This revision rearranges the same kind of material a prose editor handles: event, cause, qualification and consequence. The opening carries the fact needed to decide what to do. The explanation carries only what the system knows. The control carries the next permitted action. The U.S. Web Design System’s alert guidance supports concise, readable messages and explicit next steps; it does not validate this particular hypothetical implementation.
There is a further distinction between the written sentence and the functioning interface. W3C’s guidance on status messages concerns how such updates become available to assistive technology without requiring focus. A clear sentence that is never announced or cannot be found still leaves a reader outside the sequence. Editorial clarity and programmatic behaviour need separate checks. Neither can stand in for the other.
IV. The analogy has to know where to stop
Literature can make withholding productive. A later revelation may change how we read a character’s earlier account. A sentence can sustain uncertainty because the uncertainty is part of the experience. An interface handling an unsaved file has another obligation: the person needs enough information to decide whether to wait, retry or stop. Transferring the literary pleasure of suspense into that moment could make the design worse.
The same limit applies to perspective. Comparing a resident’s and an engineer’s constructed accounts can reveal what an editorial frame foregrounds. It cannot establish that two database permissions should be implemented as two unreliable narrators, or that an AI explanation is faithful because its prose is persuasive. Permissions need enforceable rules. An explanation needs a defensible relationship to the actual system and its available evidence. Narrative vocabulary can identify a question without solving those engineering problems.
My preference is to carry one choice across the analogy at a time. Start with a passage or screen, identify what is present and omitted, change its sequence, and inspect what the change makes easier or harder. Preserve the facts during the revision. Name the unresolved tradeoff. Where the goal is task success or accessibility, test that goal rather than treating an elegant arrangement as its proof.
The maker’s eye becomes useful when it returns us to the artifact. In the opening sentence, that meant replacing an unsupported equivalence with an inspectable choice. In the lab, it meant identifying a hierarchy and a filter’s real limit. In the upload example, it meant putting the unsaved state before the technical explanation. These are related editorial moves with different obligations. Keeping those obligations visible is what lets the comparison teach us something.
APA 7th Edition Citations & Primary Sources
Primary site artifact used for the hierarchy reading. Its accounts are constructed examples; its observation filter compares stored strings. This source supports implementation descriptions, not measured usability effects.
Primary source textOfficial component guidance, Usability guidance: next steps, plain language and polite error messages. The file-upload example in this essay is constructed, not an official USWDS example.
Primary source textOfficial explanatory guidance on communicating status updates through roles or properties without moving focus. It does not substitute for testing a particular implementation.
Primary source text