How the catalog is built
shipshape does not pretend its rule system fell from the sky. The catalog stands on the best open design-engineering work, and it is curated, extended and calibrated by hand. This page names both parts, because a product that flags unsubstantiated claims should not make one about its own origins.
The open sources
Eight open-source works inform the catalog. Nothing from them is redistributed to clients; they shape the private checks and server-side prompts.
- kill-ai-slopApache-2.0
the tell taxonomy and source-scan patterns behind many color, typography, layout and component checks
- impeccableApache-2.0
heuristic scoring, severity model, and the audit sections on patterns and strengths
usability principles, severity factors, and the metadata and density checks
microcopy, typography, accessibility and layout rules, and the review model behind our vetting
metadata deep checks, motion performance limits, and the proof-gate vetting doctrine
motion decision rules and the exact duration and easing values in fix directives
audit checklist rules (dark-section outliers, legal links, orphan words); its upgrade techniques are deliberately excluded
the production tell list, preservation ban list and design-read anchoring from years of building with these rules
What we add
- Selection and exclusion: what enters the catalog, and what stays out even when it is popular
- Consolidation: eight vocabularies merged into one catalog of 71 checks with one severity model
- False-positive catalogs: the documented exceptions that keep a purple brand from being flagged for being purple
- The scoring model: severity, visibility and decay weights, so many small findings cannot bury a professional page
- Calibration: the whole system is tested against reference sites (Stripe, Linear, Vercel, Apple, Basecamp) until they score high and slop demos score low, before anything ships
- The audit doctrine: evidence rules, a falsification pass, and the honesty policy that a claim we cannot verify is never reported as fabricated
The catalog itself, its wording, thresholds and prompts stay private. Naming the sources is not publishing the recipe; the work is in the curation and the calibration, and that work continues with every scan.