From Vague Taste to a Committed Direction
Once product truth and visual authority are separated, Impeccable still has a difficult decision to make: what should this particular surface become? This chapter focuses on the direction round, where the agent and the user move from a product mechanism and an audience situation to a chosen visual form. The useful result is not guaranteed novelty. It is a visible decision that preserves the product facts, gives the builder a concrete first-viewport promise, and makes later implementation review possible.
Start with what the surface must prove
A category template starts with appearance. Impeccable's new-work flow starts with a different question: what does this surface need to make clear, and to whom? The answer might be a visitor's reason to act, an operator's task, a reader's question, or an experience's central artifact. That question constrains the direction before the system begins discussing palettes, type, or materials.
The workflow first resolves visual authority. If the project already has a coherent system, a new surface inherits it. If the request is a redesign, the old look becomes evidence and an anti-reference, while product truth, content, function, native affordances, and technical constraints remain in force. If no visual authority exists, the system can create a world with the user. The important distinction is between extending a world and replacing it. A missing DESIGN.md does not automatically authorize a replacement identity.
For a genuinely open surface, Impeccable asks for the product's unique mechanism, the audience's real scene, the cultural world around that audience, and what the first surface must prove. It then derives concrete visual possibilities from that world. The guidance is intentionally broader than a list of design adjectives. It asks for artifacts, places, rituals, graphic traditions, publications, data displays, and interfaces that the audience already recognizes. A direction based on “clean and premium” gives an agent little to build. A direction based on a specific notation system, exhibition format, workshop ritual, or publication tradition can carry decisions about density, hierarchy, material, and movement.
The source asks for seven possibilities, but the purpose is not to make the user choose from a menu of seven moods. The candidates should be materially different and grounded in the audience's world. Near-duplicates count once. If all seven come from the same obvious material family, the exploration has stopped at the first artifact that came to mind. The exercise is a way to expose the design space before the model collapses it back into a familiar category pattern.
Imagine an illustrative surface for a tool that helps a team inspect frontend design quality. A generic direction might produce a dark dashboard with glowing metrics, tiled cards, and a large “improve your UI” headline. It is legible, but it says little about the act of inspection. A grounded direction could borrow from an annotated proof sheet, a browser inspector, a design review table, or a publication that marks corrections in the margin. Each source suggests a different relationship between evidence and action: compare two states, isolate a defect, explain a judgment, or move from finding to repair. The product facts remain the same. What changes is the form through which those facts become visible.
The example is not a recommendation for a particular visual style. It shows why the cultural-world step matters. The source of a direction should carry a system grammar that can shape hierarchy, controls, and transitions, not merely supply a color or a decorative texture. If the borrowed reference cannot change how the visitor understands the product's mechanism, it is probably a costume rather than a direction.
Impeccable then uses concept-seed to assign a direction and deal challengers. A challenger is not a second theme to paste on top of the first. It brings a different form and system grammar that must be fused with the product's actual facts. The comparison uses two questions: does the direction help the audience recognize themselves, and does it make the product clear? A visually striking idea that hides the mechanism loses on product clarity. A perfectly clear idea that looks like every competitor loses on audience identification. A direction has to carry both.
That evaluation is a safeguard against two common errors. The first is visual novelty without product meaning. The second is product clarity delivered through a generic arrangement that could be copied into any neighboring category. Impeccable's workflow does not solve those tensions by pretending they do not exist. It makes the trade-off explicit before code makes it expensive.
Let the user choose the risk
The direction round gives the user a real decision rather than asking for a vague approval after the agent has already built the page. The assigned direction leads, challengers remain visible as alternatives, and the user can steer or re-roll. A declined challenger is not silently erased. The workflow records what the selected direction kept from it, such as a more committed palette, a denser grid, or a clearer structural discipline, without copying the challenger's surface costume.
This is where Impeccable's conventional standing exit matters. The category standard is offered as a legitimate path the user can choose. It is not supposed to be the agent's quiet recommendation, and it is not treated as a joke. If the user wants the familiar path, the system asks what neighboring products the surface should sit beside and uses their craft level as the bar. Familiarity can be the right choice when trust, audience expectation, or delivery constraints outweigh differentiation. The design failure is not convention itself. It is convention chosen by inertia and then described as a deliberate point of view.
Re-rolling also has a cost. New directions consume attention, and each discarded hand can move the work farther from implementation. Impeccable therefore keeps the user in control of re-rolls and asks what quality is missing after repeated attempts. The agent may re-roll on factual grounds, such as a direction that cannot carry the product's task or truth. Taste alone is not a reason to secretly discard an assigned direction. A user-pinned constraint wins over the roll, and a material mismatch can be translated rather than treated as permission to abandon the direction.
The result is a useful division of authority. The system can force the decision into the open, vary the options, and test them against the audience and product. The user decides how much familiarity or risk is acceptable. The source supports this process as a design contract, not as evidence that every generated direction will be original or superior. Its anti-default rules are calibration devices. An explicit brief, an existing visual system, or a user decision can earn back a pattern that the generic workflow would otherwise question.
The diagram's loop is important. A challenger can reveal that the assigned direction lacks a discipline worth keeping, but that does not mean the page should become a collage of alternatives. The system raises the assigned direction, names the contribution, and preserves one coherent world. The product supplies the facts; the chosen form supplies the structure.
Commit the world, then apply the floor
After the user chooses, Impeccable records a direction contract in the surface brief. The contract has six jobs. Its thesis states what the surface owns and which category-default arrangement it refuses. Its own-world block describes the palette and component language well enough to recognize the system without its content. Its story states what the visitor should understand, believe, and do. Its first-viewport block fixes the composition, scale, and primary action. Its form records the chosen direction and its origin. Its finish line defines what must be reviewed before the work is considered complete.
Each block protects against a different kind of retreat. Without a thesis, the page can become a polished arrangement with no central idea. Without an own-world description, components can drift back toward the model's stock vocabulary. Without a story, the composition can be visually committed but behaviorally empty. Without a first-viewport promise, the strongest part of the direction may be postponed below the fold. Without a finish condition, the team can call the page done while the central interaction or its responsive behavior is still unreviewed. The contract is compact because it records decisions, not every implementation detail.
This contract is not implementation code and should not be smuggled into the delivered page as a hidden comment or serialized prop. It is a development authority that later work can reload. That distinction protects the contract from becoming another browser artifact and keeps it useful to a reviewer who needs to compare the final render against the decision that produced it.
The direction contract also makes a subtle promise: the first viewport must express the product mechanism, not merely announce it. A page can have a beautiful navigation bar, a large headline, and a polished button while saying nothing distinctive about the product. Impeccable's own guidance treats the first viewport as a thesis. If a visitor leaves after one viewport and remembers only a mood, the direction has not yet done enough. The composition must show the subject doing its job or make a product-specific consequence immediately legible.
Only after the direction is settled does the craft floor enter. The floor checks contrast, spacing, typography, motion, states, browser surfaces, copy, and brief coverage. It refuses common generated habits such as identical icon-and-heading cards, gradient text, decorative glass, thick side stripes, fake technical monospace, glyphs used as icons, and organic silhouettes approximated by geometric clipping. It also checks practical failures: content overflow, cramped padding, missing focus treatment, hidden content, and weak reading measure.
Those refusals are not the visual world. A committed world can make a different choice when the brief earns it. The repository's own design system uses a dark lacquer ground, gold and patina roles, real texture, and small radii. Another project could choose a bright, restrained system and still satisfy the same floor. The floor asks whether the result is legible, coherent, usable, and free of predictable shortcuts. It does not decide whether the project should feel like lacquer, paper, a control room, or a gallery.
That order protects both sides of the work. If the floor came first, the agent could optimize for avoiding flagged patterns and still produce a bland page. If the direction came first with no floor, a confident visual world could excuse poor contrast, hidden content, cramped type, or fake states. A chosen identity supplies ambition; the floor supplies obligations that ambition cannot waive. The two are strongest when a reviewer can say both what the page is trying to be and whether it works on its own terms.
That separation solves a recurring design argument. One person says, “Do not use an eyebrow, a gradient, or a card grid.” Another points to a successful product that uses one. Impeccable's answer is conditional: a default pattern should not appear merely because the category suggests it, but a real brief or established world can justify it. The question is not whether the pattern is morally forbidden. The question is whether the product has a reason for it that survives inspection.
A committed direction therefore gives the agent something stronger than taste words and more flexible than a ban list. It fixes the product truth, the visual world, the visitor's path, and the first proof. The craft floor then tests the implementation without replacing that decision. The next challenge is to carry this commitment into code without letting an approved direction degrade into a mood board or a safer interpretation. That is where Impeccable's measurable and code-led build paths begin.