About this library

The website is a view of the public repository. An experiment is authored in ideas/<category>/<slug>/idea.md; contributors do not maintain a second website list. The site build generates experiment pages, catalog pages, search shards, counts, field notes, open questions and the scoreboard from those records.

Reading and discovery

The homepage starts with twelve server-rendered experiments. Category pages and the complete archive work without JavaScript. Interactive search reads compact public records in files of at most 100 items, with at most three requests in flight. Unfiltered browsing loads only the pages needed for the requested result page. A search or filter scans all search shards before reporting a complete count; full idea bodies and the private intake inventory are never search payloads. Search covers titles, verdicts, tags and lessons. Twelve results render at a time.

Search, filters, sorting and page number are reflected in the URL so a reader can share a view or return to it. Failed requests preserve the last readable results and offer retry and a link to the static archive. Scores are labeled as calculated library scores; they are not endorsements or independent evidence.

Design decisions and research

The visual direction uses editorial hierarchy, open result rows, and a physical archive object. The custom still-life was modeled and rendered in Blender using original procedural geometry, materials and lighting, without stock assets. The distributed WebP is covered by the repository's MIT license.

The research informed the decisions; it does not establish that visitors prefer this particular implementation. No user study has been run on the redesign.

distinguishes useful visual richness from irrelevant decoration. The primary browsing controls remain prominent; typography and composition carry the design.

supports visible selections, removable filters and clear result counts. Its commerce findings are applied here as a design inference.

criticizes repeated templates, decorative modules and unnecessary reveal effects. This is practitioner opinion, not a representative survey.

Library of Congress search guidance demonstrate search alongside topic browsing. These are interface references, not evidence that popularity follows from a particular layout.

• Nielsen Norman Group: aesthetic and minimalist design
• Baymard: applied filters
• Webdesign discussion: preventing the AI slop look
• Europeana's exploration interface and

The site uses system fonts, explicit focus states, reduced-motion support and no analytics scripts. Experiment and category pages share the same visual language. No new paid service or visitor account is required.

Checks before publication

Run npm run check. It validates source records, scans public output, checks deterministic generation and exercises additions, escaping and withdrawals. Browser review must cover search, combined filters, clearing, pagination, returning from a record, unanswered questions and contribution navigation. Test narrow and wide layouts, as well as the static fallback and failure message.

Cloudflare Pages builds every main-branch repository update with npm run check. This setting was verified directly in the Pages project configuration, replacing the earlier generation-only command that skipped validation and privacy gates. Only the generated site/dist directory is public; source inventory stays ignored and outside the deployment output. Versioned CSS and JavaScript URLs prevent the browser from retaining the previous design after a deploy.