Design Two-Track Extraction
Design Two-Track Extraction
Process note. Mine the design books for technique, but split the catalog at the line AI now draws through the discipline. Books like Refactoring UI were written before agents could execute the code portion of UI; the mechanical rules they teach are now largely delegatable. What stays with the human is the judgment those rules can’t encode. So every technique gets sorted into one of two tracks, and the sorting itself is the product.
Why two tracks
A design agent can apply a deterministic rule perfectly and still produce something tasteless, because taste is the part that doesn’t reduce to a rule. The split protects both halves: agents get a clean, executable ruleset they can build off; the human keeps a curated catalog of what only the eye can judge — which is the only part worth spending scarce reading time on.
The classification test
For each technique, ask: can this be reduced to a deterministic rule an agent applies without perceptual judgment?
- Yes → Agent Track. Parameterizable, checkable, no taste required. (“Body line-height 1.5×.” “Spacing scale 4/8/12/16/24/32.” “Disabled state: lower opacity, not gray.”)
- No → Human Track. Requires perception, context, or taste to apply well. (“Does this hierarchy feel right?” “When to break the grid.” “Which of three correct layouts is better.”)
- Both (most techniques) → split at the seam. State the mechanical core in the Agent Track and the judgment edge in the Human Track, cross-linked. The seam is where the real thinking lives.
The two tracks
- Agent Track — Executable UI Technique Catalog — machine-consumable rules; could later harden into a skill or a
tsumugu-coreruleset / CLAUDE.md block. - Human Track — Taste & Judgment Catalog — what Wedge needs to see and judge; example-driven, comparative, read by the human eye only.
Processing queue
Sources staged at raw/sources/design/ (text extracted to raw/sources/design/_text/).
| Source | File | Status | Notes |
|---|---|---|---|
| Refactoring UI (Wathan & Schoger) | raw/sources/design/Refactoring UI.pdf | ✅ extracted 2026-06-30 | Confirmed as a rules engine — heavy agent-track yield across spacing, type, color, depth. |
| Universal Principles of Design (Lidwell et al.) | raw/sources/design/Universal Principles of Design.pdf | ✅ extracted 2026-06-30 (UI-scoped) | Scoped to the book’s UI/UX/Graphic/Product “most useful” lists + core Gestalt & interaction laws. ~150 general/management principles left unmined by design. |
| (later) jlreq / clreq, Typotheque | — | not staged | CJK layout — mostly agent-track rules; see [[wiki/Design/Design Expansion — Reading & Resources |
How to run the extraction (for future-me)
- Read one source via the
pdfskill. Go chapter by chapter. - For each discrete technique, apply the classification test and write it into the correct track note (or split it).
- Keep Agent-Track entries atomic and prescriptive; keep Human-Track entries example-anchored and about judgment, not rules.
- Update this queue’s Status column and bump
source-counton each track note.
Open questions
- Final form of the Agent Track: prose catalog, or a structured skill / ruleset that
cosand tsumugu agents load directly? (Lean toward extracting as prose first, hardening into a skill once stable — don’t reinvent; check what Hermes/Grok already consume.) - Universal Principles spans more than UI (it’s general design). Scope to UI-relevant principles, or catalog whole? Default: UI-relevant first, flag the rest.
Related Pages
- Master Scorecard — all 250 items graded on the lens.
- Human vs AI Capability Lens — the scoring model.
- Design Expansion — Reading & Resources
- Front-End Web Design
- Design, Condensed