Launching vizcrush: Three Beliefs My Benchmarks Killed

On the eve of vizcrush's public launch, two files were open side by side. The left file contained the launch copy: the JavaScript core outperformed the most popular npm downsampling package by a factor of 32, and WASM added another 5-10x. Meanwhile, the right file showed the benchmark control run, where wasm/js was approximately 1.00×.

To verify the measurements behind the claim, the author examined the results. Half of them held true, like the 32× JavaScript comparison (1.72ms vs. 55.52ms) and the control run contradicting the claimed 5-10× WASM improvement. Consequently, every performance claim either received a measurement or was removed from the launch. Three beliefs were debunked, each receiving a public retraction in the form of an ADR (After-Death Record) within the repository.

Vizcrush is a collection of data primitives for browser visualization, written in Rust and compiled to WebAssembly with a pure-JS fallback. It was open-sourced with 11 packages available on npm. This launch story revolves around turning benchmark results into product policy, where claims, documentation, and WebGPU policy align with measurements, while WASM dispatch remains availability-based pending further investigation.

Keep in mind that each result is workload-specific: LTTB (Largest-Triangle-Three-Buckets) is downsampling, the stats kernel is a reduction, and bin2d is histogramming. The performance gains observed were algorithm- and engine-dependent, and none of the findings imply a general WASM-versus-JS verdict. In fact, these specific workloads and engines were measured, with claims and documentation following each measurement.