First experimental validation
The first experiment is being scoped to test the research hypothesis, not to act as a finished product. Before any API design is finalized, the core claim should be tested on a deliberately small surface: can a planning layer select execution strategies with low enough overhead and low enough regret to provide meaningful performance portability?
- Map, reduce, stencil, and chained map work
- JavaScript, WebAssembly, and WebGPU
- execution latency, transfer cost, planning overhead, oracle regret, and cross-device portability
The key metric is oracle regret: how much slower the planning layer’s chosen strategy is than the best strategy available on that device, chosen with perfect hindsight. Low regret across heterogeneous devices would validate the hypothesis; high regret would refute it. Either result will be published with the same prominence.
As the experiments run, this page will grow to document the benchmark methodology, the device and browser matrix, the graphs and datasets, the planner results, and the raw data, with links to reproducible code in the canonical repository.
Experiment design and code will live in simplbase/webcompute.