Engineering · from the codebase · no fluff
Notes from building it all in the browser.
Everything on this site — the editor, the SQL engine, the dataset library — runs client-side on a plain static host. That constraint forces interesting designs. These are the write-ups, with the real code and the bugs left in.
The editor
A million-row CSV editor that never copies the data
An edits-overlay Map, an operation-log undo stack, and ~30 DOM rows pretending to be a million. Plus the focus() bug that made double-click land on the wrong row.
Read the design →The SQL engine
Real SQL on a static host, without breaking the ads
Why we ship single-threaded DuckDB-wasm on purpose: SharedArrayBuffer wants COOP/COEP, cross-origin isolation kills ad iframes, and a tab full of CSV doesn't need threads anyway.
Read the tradeoff →The dataset library
One seed, two runtimes, the same bytes
The sample datasets generate in your browser and must match the committed files byte-for-byte. Per-row mulberry32 seeding, a warm-up bug, and the Playwright test that diffs the bytes.
Read the postmortem →
The CSV parser behind these tools is open source: csvtd-parser on GitHub — zero dependencies, quote-correct, with the delimiter auto-detection described in these posts. See what shipped recently in the changelog, or just use the tools.