OPERATOR'S LOG / EXPERIMENT 001

Every generation, on the record.

This is the inspectable evidence archive for Design Skill Benchmark Baseline 001: 32 generated websites, six models, three design systems, and the grades behind the case study.

Records: 32
Models: 6
Skills: 3
Frames: 64

What this is: a static snapshot of every scored generation, its desktop and mobile review frames, score breakdown, review note, violations, and a sanitized source archive. What it is not: a live evaluator or a model leaderboard beyond this dataset.

Download Baseline 001 (.zip)

JSON ↓ CSV ↓

DIMENSION 01 / DESIGN SKILLS

Three design systems, strict rules.

Each model was tested against specific design skill definitions to evaluate style fidelity, layout grammar, and whether negative constraints were respected.

BLUEPRINT 91.3 AVG · 5 OUTPUTS

Blueprint System

Technical and architectural diagrammatic language: high-density grids, technical monospace typography, dimension markings, schematic drawing frames, and blueprint cyan accents.

Key requirements: Spec-sheet hierarchy, title block/schedule, engineering notes, dimension markers.
Prohibitions: No marketing gradients, no rounded cartoon pills, no generic hero banners.

FIELD MANUAL 90.9 AVG · 12 OUTPUTS

Field Manual System

Government, military, and institutional operational handbooks: numbered procedure lists, tactical callouts, classification stamps, warning blocks, and utilitarian olive/gold accents.

Key requirements: Strict section numbering, procedure tables, classification banners, emergency callouts.
Prohibitions: No SaaS glassmorphism, no marketing fluff, no non-functional interactive gimmicks.

SOLARI 89.2 AVG · 15 OUTPUTS

Solari System

Mechanical split-flap displays and transit timetable boards: dark airport/station departure aesthetics, high-contrast amber/yellow glow, and tabular multi-column schedule matrices.

Key requirements: Split-flap typography/character cells, board header status, status codes (ON TIME, DELAYED).
Prohibitions: No light mode, no fluid unconstrained cards that break character grid alignment.

DIMENSION 02 / PROMPT CATALOG

Eleven real-world scenario prompts.

Prompts combined dense domain requirements, data structures, and responsive layouts to test whether models followed instructions or took shortcuts.

Prompt Scenario	Design Skill	Domain & Complexity	Outputs
RELAY — AI Agent Deployment & Operations	Field Manual	Multi-agent operations handbook, orchestrator routing, telemetry tables	4 runs
Municipal Trail Operations	Field Manual	Park trail maintenance, incident status report, ranger log	3 runs
Northstar Observatory	Field Manual	Telescope optics schedule, observation logs, aperture coordinates	3 runs
Municipal League Baseball	Solari	Civic baseball split-flap scoreboard, standings board, pitch count matrix	6 runs
Terminal 17 — Regional Freight Exchange	Solari	Intermodal rail/truck timetable board, bay assignments, cargo status	4 runs
Night Shift 91.3 FM	Solari	Late-night radio program board, broadcast queue, audio track schedule	3 runs
ATLAS — Distributed Intelligence Architecture	Blueprint	High-density multi-node diagram, memory architecture, execution topology	3 runs
Halloway Instruments HX-4	Field Manual	Environmental sensor field manual, calibration chart, sensor readings	2 runs
Aeroline R7 — Endurance Cycle Platform	Blueprint	Endurance engineering spec sheet, frame stress diagrams, torque tolerances	1 run
Northline Waterworks	Blueprint	Regional water authority engineering sheet, pipeline pressure monitors	1 run
Analog Cinema Club	Solari	Repertory theater 35mm schedule, screening room split-flap board	2 runs

DIMENSION 03 / EVALUATED MODELS

Six leading LLMs on the test bench.

Scoreboard of model performance across 32 runs, measured on 6 independent grading dimensions.

Model	Outputs	Overall Score	Responsive (10)	Design (20)	Key Observation
Grok 4.6	4	93.4	7.5 / 10	17.9 / 20	Highest consistency; tightest adherence to negative constraints and typography.
GPT-5.6 Sol	11	93.0	7.9 / 10	17.3 / 20	Strongest responsive behavior; excellent split-flap and table translations.
Gemini 3.7 Flash	10	90.4	6.0 / 10	17.2 / 20	High design polish and prompt fidelity; occasional mobile viewport overflow.
MiMo V2.5	2	84.6	5.7 / 10	14.5 / 20	Solid visual styling; missed a subset of explicit table requirements.
Sonnet 4.6 Thinking	3	84.0	3.2 / 10	15.3 / 20	Strong desktop layouts, but severe mobile responsive failure (fixed widths).
Kimi K2.6	2	82.4	2.6 / 10	15.0 / 20	Creative layouts, but widespread horizontal scrolling on screens under 1024px.