Human Track — Taste & Judgment Catalog

The Human Track is the catalog of design calls a machine can draft but cannot score. The job is to look, not to memorize a list of named effects. Entries were sorted by Design Two-Track Extraction; the mechanical core of anything here lives on the Agent Track. A seam is the cut that puts that core on the Agent Track and the leftover judgment here. This is the catalog that is actually read and looked at.

Developing the eye

Interfaces worth admiring are rebuilt from scratch, with no peeking at devtools. The friction of matching the original surfaces the invisible tricks — tighter heading line-height, letter-spaced caps, doubled shadows. The drill is the calibration. Calling it the highest-yield drill is the book’s talk, not a measurement. [RUI]

The hunt is for decisions that would never have been made. On any design that works, the question is what the designer did that would not have occurred: an inverted datepicker, a button sitting inside the input, a two-color headline. Those get collected. [RUI]

Elimination trains comparison. Given a value system, the trainable perception is seeing that two of three options are obviously wrong. The fixed set itself is agent-work. [RUI]

Type attention is cultivated on purpose. Sustained looking at typography on good sites converts “I can’t say why” into a fast good-or-bad call. [RUI]

At the finish, the eyes beat the numbers. Systematic palettes, proportions, and scales get most of the way there. The last saturation or lightness tweak is made by eye. [RUI]

The aesthetic-usability effect is real, and it is a trap. Attractive things get rated as more usable, so beauty buys forgiveness — and it can hide genuine faults from the designer. Both get judged, separately. [UPOD]

When to break the rule

Dense interfaces are legitimate. “Start with too much whitespace” is overridden on purpose for a dashboard. The test is whether density is a choice, not a default. [RUI]

A grid is a tool, not a master. The break to fixed widths happens the moment fluid columns make something wider on a medium screen than on a large one, or shrink it past a usable minimum. [RUI]

A line longer than seventy-five characters can sometimes work. The call is made only while knowing the territory is risky. The safe range lives on the Agent Track. [RUI]

Justification is for mimicking print — a magazine or newspaper feel. Even then, left-align is usually fine. [RUI]

MAYA is Most Advanced Yet Acceptable. Novelty is pushed exactly as far as the audience will tolerate, and no further. Where that line sits is a read of the room. [UPOD]

Satisficing is knowing when a solution is good enough to ship, and when it is still worth more polish. The rules cannot make that call. [UPOD]

What looking actually is

The last tweak is by eye. Any leftover claim that the systems get “ninety percent” of the way is the book’s talk, not a finding. Beauty buys forgiveness, and it can hide faults from the person who made the surface. Judge the look and the use separately. The named effects below are calls with a tell, not a syllabus.

Aesthetic coherence

A personality is chosen, and every choice serves it. Font, color feel, border-radius, and copy tone cohere toward one character — serious, playful, or elegant. Calibration is against sites the audience already uses. Direct competitors are not aped. [RUI]

Border-radius: consistency is a rule, amount is a feel. Mixing square and round looks worse. How much round communicates personality. [RUI]

Color “psychology” explains a choice after the fact. It does not generate one. What looks right is the practice. [RUI]

Realism has a stop. Once light can be simulated, the temptation is to tune toward a photograph. Knowing when to stop is taste. [RUI]

Hierarchy is what makes something feel designed. Noticing when it is off — noisy, flat, competing — is a perceptual skill, not a checklist. [RUI]

Form follows function, and faith-follows-function is the warning the inverse fails. Functional-looking is not functional. The call is whether honest form actually fits, versus mere styling cues of competence. [UPOD]

Wabi-sabi sits next to gloss bias and contour bias as a stance, not a law. People are drawn to glossy surfaces and to curved-over-angular forms, and they also find beauty in imperfection and age. Whether to lean into polish, softness, or deliberate roughness is the call. [UPOD]

Symmetry reads as stable and formal. Asymmetry reads as energetic. Which one fits the feel is judgment. [UPOD]

The back of the dresser is finished anyway. Code, empty states, and edge layouts — a craft standard held, not a checkable rule. [UPOD]

Visual perception

Gestalt grouping is perceived, not measured. Proximity, similarity, closure, common fate, good continuation, figure-ground. An agent can space things. Whether the eye reads the intended groups, and where figure and ground flip ambiguously, still needs a look. [UPOD]

Surface area drives perceived weight. Bold type and solid icons feel heavy. The eye is trained to feel imbalance before contrast is used to fix it. [RUI]

Hue has a perceived brightness independent of HSL lightness. Yellow reads lighter than blue at equal L. The formula is agent-work. Judging rich versus dull, warm versus muddy, after a hue rotation is the eye. [RUI]

Where attention lands first is verified by looking. Focal point, entry point, reading gravity. Only looking tells whether the eye actually went where it was sent. [UPOD]

Problem-framing

A feature comes before a layout. The real problem is framed before any pixels. The shell cannot be designed until a few features exist. [RUI]

The smallest shippable version is designed pessimistically. Functionality that cannot be built is not implied. Short build-and-test cycles replace imagining every edge case. [RUI]

Detail comes later. Grayscale first forces spacing, contrast, and size to carry the hierarchy, before color can rescue a weak one. [RUI]

User-centered is not user-driven. People report problems well and prescribe solutions badly. How much to weight what they ask for against what they need is judgment. [UPOD]

The mental model is the whole game. The user’s model is built only from what the surface shows. Judging whether that surface actually induces the right model — and surfacing it when “synced” or “cloud” breaks — is interpretive. [UPOD]

The remembered experience is designed at the peak and the end, not the average. Which moment is the peak, and how the thing ends, is a taste and empathy call. [UPOD]

Desire lines tell the truth. Where people actually go is paved, rather than the path that was drawn. [UPOD]

Examples

A raised door panel vs an inset cabinet panel.
Light from above: the source of every UI shadow decision.

Typography on well-designed sites is inspected. Steal from people who care. [RUI]

Asset sources worth knowing are a class, not a storefront: a free photo library, an SVG pattern library. [RUI]

tsumugu is the language-learning reader and dictionary these catalogs are meant to judge. Two pairs from those surfaces, looked at rather than named:

Guess-first vs the answer shown.
Good: the reader has to produce the word before the gloss arrives.
Bad: a peek that hands the answer and calls it practice.

A control vs an emoji standing in for a name.
Good: the action is labeled in words the hand and the reader can both use.
Bad: a lone pictogram carrying the only name the control has.

A judgment catalog invites collecting named effects the way a rules catalog invites collecting numbers. Looking at two surfaces beats naming twelve effects. Rebuilding from scratch with no devtools is the expensive drill; the hunt, the elimination, and the type attention are cheaper and still train the same eye. If a pass produces no “decision I would not have made” and no off-hierarchy — noisy, flat, competing — the catalog was read, not used. After one rebuild, one invisible trick can be named that would not have been thought of. After grayscale-first, color is not carrying the hierarchy.

Following the rules is not enough to tell whether the result is good. The eye is the catalog.

Related

• Design Two-Track Extraction — the classification test that produced this catalog.
• the Agent Track — the mechanical core of anything here.
• Front-End Web Design — the tsumugu surfaces these good/bad pairs are drawn from.

Open Questions

Whether more tsumugu pairs land here once the reader and dictionary chrome stabilize.

Whether seam notes on the Agent Track become jump anchors once these headings stay put.

Sources

Wathan & Schoger, Refactoring UI (2018). Lidwell, Holden & Butler, Universal Principles of Design, 3rd ed. (2023). Kurosu & Kashimura (1995); Tractinsky, Katz & Ikar (2000), aesthetic-usability. Loewy, MAYA. Simon (1956), satisficing. Bar & Neta (2006), contour bias. Wertheimer, Gestalt grouping. Helmholtz–Kohlrausch effect. Mack & Rock (1998); Simons & Chabris (1999), inattentional blindness. Kahneman, Fredrickson, Schreiber & Redelmeier (1993), peak-end.