wiregen — declared by cat-harness — source held

Applies when. Designing a USER INTERFACE, for example a page layout, a navigation scheme or a visualiser, where the choice between candidate designs has to be reviewed and, when reviewers disagree, adjudicated. It is a design-generation and design-evaluation method. It does not choose between non-UI options (use the decision-analysis methodology) or grade evidence. It does not by itself settle a disagreement: that is adjudication.

Origin. Sidong Feng, Mingyue Yuan, Jieshan Chen, Zhenchang Xing and Chunyang Chen, “Designing with Language: Wireframing UI Design Intent with Generative Large Language Models”, arXiv:2312.07755v1 [cs.HC], 12 Dec 2023. Ingested in full at cat-harness/library/arxiv-2312.07755v1/. The source states no licence, and its licence could not be established (see check:source-licence).

Ingested sources:

library/arxiv-2312.07755v1 · view · item page · source

Applied by skills:

Node file: wiregen.md

WireGen: from a written intent to a reviewed, mid-fidelity wireframe

Adopted in part, 2026-09-23, by the owner’s direction. In the owner’s words: “make methodology/utilize/integrate into existing processes. need both web and mobile layouts in usability reviews. wireframe is part of design process for adjudication”.

Integrated into CRDM, 2026-10-14. crdm-requirements-definition.bpmn calls Process_WireframeDesignReview (Call_Wireframe) whenever a requirement changes a rendered surface, before anything is built; the candidates become assets of the request’s proposal, and the requester or BA chooses one at BA_ReviewReqs. See wireframe-design-review.

Placed in cat-harness (owner, 2026-09-23: “put wireframe in cat-harness, not folio-assist(-core)”). The method is domain-neutral, and methodology-adoption §4 puts a domain-neutral method in the harness. It was first drafted in litlfred/ihris, and its first use was that folio’s data-model site. Issue #1023.

What the method says (rendered faithfully)

Section numbers below are the paper’s.

Design intent comes first, as language (§3, §5.2). A short natural-language description of what the screen is for and what it holds.
Mid-fidelity is the target (§3.1.1). The wireframe is monochrome and simplified, but it carries real content, semantic icons and interactive elements, not boxes and placeholder text.
HTML is the carrier (§3.1.2, §3.3). The screen is expressed as a UI-specific HTML-like language. Raw output is post-processed against UI guidance: icons chosen by meaning, typography, and layout guidelines (§3.3).
Evaluation is blind and comparative (§4.3). Independent raters with design background see the description and a shuffled grid of candidates, without knowing which generator produced which. They mark candidates significantly better or significantly worse.
Usefulness is asked separately (§5.2): whether a wireframe is effective for inspiration, whether it relates to the description, and whether the candidates are diverse. These are rated on a 5-point scale.
Refinement is iterative (§6). Intent, then wireframe, then a revised intent, then a revised wireframe.

What this platform adopts, and what it refuses

	adopted	refused, with the reason
intent statement	yes: every candidate set starts from one, and it is the criterion candidates are judged against	none
mid-fidelity with real content	yes: real content from the folio (for iHRIS: real form classes, fields, lists)	none
HTML carrier + guideline post-processing	yes: wireframes are HTML, and the mechanical checks are the post-processing gate	none
fine-tuning an LLM on Rico / Screen2Words (§3.1, §3.2)	no	That is how the authors generate; it is not a review method. Candidates here may come from a person, an agent or any tool. The method applies to what they produce.
blind, shuffled comparison	yes: reviewers are not told a candidate’s author	none
better/worse marks and Likert ratings	the structure, not the arithmetic	`methodology-adoption` refuses scoring and summing. A mean of Likert ratings reads as a measurement of a judgement. Reviewers record per-criterion `pass` / `warn` / `fail` with a reason.
mobile only (the paper’s data is Android, §3.1.1)	extended: both web and mobile layouts are required in every usability review (owner, 2026-09-23)	The source’s scope is narrower than our need. We state the extension rather than claim the paper supports it.
the reported numbers (77.5%, the user study)	no	They describe the authors’ model and datasets. Nothing here depends on them.

How it plugs into existing processes

Review record: each criterion gets entries from a script (the mechanical checks), an agent and a human, in block-qa/v1 sidecars. This is folio-assistant’s existing multi-reviewer primitive.
Disagreement: when entries for one criterion disagree, the candidate set enters Process_Adjudication (processes/sdlc/adjudication.bpmn). The adjudication leads, the checker’s entry is kept, and the dispensation carries its reason.
Choice between surviving candidates: a one-off choice, so it uses the decision-analysis methodology. Rejected candidates are kept with the reason they lost.
After build: the rendered result does not go through theme-ui-review.bpmn any more. Theme and UI review moved to ingestion only, by owner ruling (bean 9fdi, 2026-09-24): it runs when the graphical assets arrive, from ingest-theme.bpmn, and still requires both web and mobile layouts (#1023). A post-build visual look for a wiregen output is therefore not provided by that process — recorded so it is not assumed.

The executable form is processes/ui/wireframe-design-review.bpmn. The operating skill is wireframe-design-review. The mechanical checks are the Tool wireframe-check.

Refusals

Never review a single candidate as if it were a choice. The method is comparative; at least two candidates, or say why only one exists.
Never accept a wireframe reviewed at one viewport. A web-only or mobile-only review is incomplete, not passed.
Never show a reviewer who made a candidate. Blindness is the method’s evidence of fairness.
Never let placeholder text through. Lorem ipsum makes it a low-fidelity wireframe, and the method is about mid-fidelity.
Never average the ratings into a score.