How this site was made.

The seam

The line down every page is an engine's evaluation bar. The paper side is White's share of the position on screen: 55.9% on the home page, after 10…Bg4.

Type

Archivo for the words, JetBrains Mono for moves and numbers. Roman is fact; italic is voice. Both are self-hosted subsets under the SIL Open Font License 1.1.

The engine

Move generation, alpha-beta search, the handcrafted evaluation and a learned NNUE evaluation are written from scratch. The net was trained on CC0 Lichess evaluations with quantisation-aware training and runs through a small WebAssembly module. It starts only when you ask, runs in a Web Worker and pauses when the tab is hidden.

The move generator's receipt, from the start position: depth 1, 20; depth 2, 400; depth 3, 8,902.depth 120depth 2400depth 38,902

Made for this site

Every piece, board, grain and light is made in code for this site: no third-party models, scans or environment maps. The four sound cues are synthesized, with no samples. The screenshots are the projects' own.

Publishing

Every word and number lives in one content file in the repository. The pages, the résumé and the share cards are built from it. There is no content management system.

Tested

Every number on the site is registered in one claims ledger, with an evidence type, an owner and a date. Every move shown is legal under the engine's own move generator, and the home page plays the recorded career line. Every page has no automated accessibility violations at desktop and phone widths, with and without reduced motion, and no sideways scrolling at 320 px. The end-to-end suite runs in Chromium, Firefox and WebKit. The learned net was measured against the handcrafted one before it could play: it lost by −143.3 ±35.4 Elo over 128 games, and the Lab publishes the loss.