**Rationale:** Tests two orthogonal capabilities in one pass — factual web research (grounding, recency, no hallucination) and aesthetic judgment (current design trends vs generic template output). A model can fail either axis independently, which is the point: design quality alone must not mask a fabricated-sources fail. **Expected spread:** Frontier models with reliable search tool-use should clear GT1–GT3 easily; the real spread will show up on GT4 (design originality) and on models that either skip the search tool or hallucinate plausible-sounding articles despite having it available. **Known cheats/risks:** - Model ignores the search tool entirely and invents 3 plausible articles (titles/sources that sound real but don't exist) — must be caught manually, this is the highest-value failure to detect. - Model searches but grabs stale articles (>14 days) without checking dates. - Model produces a visually polished page but with broken/dead article links. - Model pads with lorem-ipsum-style filler design elements to look "trendy." **GT checklist:** - **GT1 (blocking gate):** All 3 articles are real and verifiable — correct source, correct URL resolves, headline matches actual content. Manual check required. Any fabricated article → automatic Fail tier regardless of GT2–GT6. - **GT2:** All 3 articles published within 14 days of the run date (check actual publish date vs run timestamp, not vs today). - **GT3:** Summaries are accurate — no factual distortion vs source content. - **GT4:** Design quality — Playwright screenshot review. Score against: distinctive typography, non-default layout, deliberate color system, absence of generic Bootstrap/Tailwind-default look. - **GT5:** Technical execution — renders with no console errors, single self-contained file, responsive at 2+ viewport widths. - **GT6:** No placeholder/lorem-ipsum content, no broken images/missing alt. **Scoring:** GT1 is a hard blocking gate (Fail tier if violated, independent of everything else). Otherwise: S = GT2–GT6 all pass with strong GT4; A = all pass, average GT4; B = GT2/GT3 pass but GT4/GT5 weak; C = partial GT2/GT3 issues (e.g. 1 stale article) but no fabrication; Fail = GT1 violated or degenerate/non-rendering output.