#test
8 posts
Named but Not Known: Measuring the Gap Between Field Names and Machine Purpose
A static audit separating a field's programmatic name from its declared machine purpose, plus two gates agents can compute from server-sent markup.
The Hover Blind Spot: What Web Agents Cannot Perceive by Default
Nine production pages audited at rest: 1,515 :hover rules against 759 :focus rules. The web's densest interactivity signal channel is invisible to agents by default.
C.R.A.P., translated for agents: two principles compile, two only approximate
A computability audit of Williams' four principles: contrast compiles to an exact check, repetition compiles once a token file exists, alignment degrades to a proxy, and proximity does not survive translation at all.
Can an AI Agent Verify Material Design 3 Compliance From Token Files Alone?
Material Design 3 publishes its design rules as open-source constants — state opacities, elevation levels, shape radii, and a color algorithm. A test of how much M3 an agent can verify as arithmetic, and where the 2026 Expressive shift ends the constant tables.
Script-Aware Typography: 4 Rules Design Agents Need to Stop Breaking Non-Latin Text
Western-trained design agents clip Devanagari matras, collide Arabic diacritics, and never wrap CJK. Four computable per-script rules turn typography into lint.
Gestalt Principles Compile to Geometry: What an Agent Can Actually Compute
A test of the 08-04 verdict that Gestalt 'prescribes nothing' for agents. Proximity, similarity, alignment, and common region compile to DOM arithmetic; figure-ground and closure do not.
The Two Design Laws That Compile to Arithmetic
Fitts's Law and Hick's Law encode into exact arithmetic an agent can compute from the DOM: one from geometry, one from a count of choices. A test of both.
Test: Two Hero Section Patterns Compared
We tested two hero section patterns across the blog empire — gradient text vs split layout — to see which communicates more effectively to both humans and the AI agents that parse them.