Prepress · browser-local screen geometry · 2026 · Solo — product framing, test-first screen geometry, interaction design, accessibility, responsive verification, production deployment
Rosette Proof
A prepress instrument that makes four-screen angular interference inspectable before plate-making.
The problem
Prepress operators and hands-on printmakers checking a four-process halftone set must compare six repeating angle pairs, understand which separation is tightest, and preserve the exact configuration for a physical proof. A list of C/M/Y/K values hides the closest pair, while an unrelated visual can disagree with the arithmetic. Rosette Proof keeps that bounded review causal: set one ruling and four angles, watch line fields rotate from those values, isolate any pair directly from the six-pair clearance ledger, and copy the tightest gap into a deterministic press note. Risk bands are review heuristics, and the equal-frequency beat period is a geometric reference rather than a prediction of a visible press defect. The tool does not model a RIP, stochastic screening, dot shape, registration, gain, substrate, ink behavior, output scale, or color. Settings stay in ephemeral browser state; there is no upload, account, persistence, paid API, tracking SDK, outbound email, advertising, payment, or destructive operation. Production deployment dpl_EdvxLyaer2rLhfDTEwKTPat3x54H reached READY before independent Foundry artifact, HTTPS deploy, and seven-step behavior checks passed at the stable alias.
Architecture
Key decisions
Make interference the operating surface
The central rosette is not decorative artwork. Each SVG line field rotates from its actual ink angle, and selecting a ledger row suppresses the other two inks so the operator can inspect one mathematical pair without changing the registered set.
Keep risk language transparent
Rosette Proof exposes its critical, watch, and clear thresholds beside the controls and calls them review heuristics. It carries the tightest pair forward without claiming that screen geometry alone predicts press quality.
Keep the prepress claim narrow
The core models equal-frequency line-angle relationships only. The interface repeatedly requires an output-size proof on intended stock and names the RIP, dot, gain, registration, substrate, and ink variables outside its boundary.
Use one disciplined Configure bench
IBM Plex, an 8-pixel rhythm, flat paper layers, sharp rectangular controls, process-ink marks, and one interface blue create set-to-rosette-to-note causality without feature tiles, center stacking, blur, icon toppers, or dashboard metrics.
Let failed gates improve the product
The first behavior walk exposed a timer race that erased a copied receipt; a test-first state-settlement helper fixed it. A second walk exposed unnecessary smooth scrolling that destabilized off-viewport automation. Both were repaired before deployment.