● Unified Evaluation Artifact

ColdStart × Slop-Catcher — Combined Demo

One target, two lenses: design QA + zero-shot generalization.

⚠ Offline demo: scripted action sequence against a mock page — verifies plumbing, not model capability.
Design-QA scans in this report are MOCK / dry-run — not a live audit (VLM score is a mock fixture; metrics are self-reported by the scanned app).
Human-Grade Baseline

Clean Design (/)

PASS
4/100
Slop Score • Passed modern design audit
Visual Hint • High-Craft Element

Sky-600 button with crisp contrast, 8px grid alignment, and clean typography stack.

Contrast Ratio
7.2:1
Spacing Variance
0px
Design QA Audit
✓ No design flaws or AI slop detected
Synthetic Degradation

AI Slop (/?slop=1)

WARN
40/100
Slop Score • Triggers Design QA gatekeeper
Visual Hint • Synthetic Slop Element

Purple gradient hero with low-contrast gray-on-gray button and random off-grid margins.

Contrast Ratio
2.1:1
Spacing Variance
15px
Design QA Flags
  • Low contrast ratio detected (2.1:1 < 4.5:1 WCAG AA)
  • Off-grid spacing variance detected (15px deviation)

Evaluation Matrix

Target Slop Score Design Status Task Completed Verifier Checks (D1/D2/D3)
Clean Landing Page / 4 PASS YES PASS (D1, D2, D3)
Slop Landing Page /?slop=1 40 WARN N/A N/A

ColdStart Agent Execution & Ground Truth Verifier

Plumbing Verification (Local, Mock Page) — real HTTP + SQLite locally, scripted agent; not a live cold model run.

Task Completion
DONE
Steps Taken
6 steps
Final Page Title
Access Requested
Ground Truth Verifier
PASSED (3/3)
Agent Task Instruction
Open early access page at http://localhost:13122/, enter name "ColdStart Agent" and email "agent@demo.solari", submit the form, and report done.
✓ Check D1: SQLite Row Existence
D1 PASSED: Signup row found (id=1, name="ColdStart Agent")
✓ Check D2: Target Email Match
D2 PASSED: Email matches expected "agent@demo.solari"
✓ Check D3: Run Window Validity
D3 PASSED: created_at (2026-09-03T14:48:33.551Z) is within window [2026-09-03T14:48:33.537Z - 2026-09-03T14:48:38.559Z]