What does the Coffee roaster — origin and roast meter prompt produce?

A coffee website with Origin map, Roast-level meter, Brew guides, Subscription lane. The vibe target is "roast-fresh, kraft-warm, specific" and the prompt body is printed in full on the page.

How well does it score?

90% average fidelity over 4 recorded runs. Best model: Claude 4.6 Sonnet. Weakest recorded run: 88% on GLM-4.6 (CN).

Are failures included?

No build failures are recorded for this prompt, and the log says so rather than implying every model succeeded elsewhere.

Concept render — composed live from this prompt's palette & layout. Reproducible with the library, not a stock screenshot.

Concept scene: Star Motes — open in library →

CoffeeVerifiedby Motif Studiopublished 2026-09-10

Coffee roaster — origin and roast meter

vibe: roast-fresh, kraft-warm, specific

starter — 4 blocks, 2 stacks≈ 700 tokens to run (estimate)

prompt.txt — copy-paste ready

A coffee roaster's page where origin and roast date do the talking.
• Origin map: a quiet SVG map (no embedded script) with the current single origin marked; beside it the lot's story in four short lines — farm, elevation, process, why it tastes the way it does.
• Roast-level meter: every bag card shows a roast meter (light to dark) and the roast date in mono — 'roasted Tue' beats 'freshly roasted' every time.
• Brew guides: one guide per method (v60, aeropress, french press) with grams, ratio and a small animated pour demonstration — no PDF walls.
• Subscription lane: frequency pills (weekly / fortnightly / monthly) with honest pause-anytime copy; the roast-date promise extends to every box.
• Tasting notes: written as description, not poetry ('stone fruit up front, cacao finish, low bitterness') with altitude and variety in the spec line.
• Palette: warm kraft, ink, one roaster-orange accent.

Stacks shipped:ReactHTMLBlocks:Origin mapRoast-level meterBrew guidesSubscription lane

Community:▶ Re-run this prompt ⤴ Submit a remix of it ★ Your saved listRe-runs are a labelled simulation; the run log on this page is the real record.

⚠ Before you run it, know this

Lowest score was GLM-4.6 (CN) at 88/100 — Tasting notes vivid; map needed one accessibility label pass.
All models built cleanly on the first pass.