What the research is trying to hold true

Research goals

These are the commitments WebCompute is trying to keep as the research moves forward, not a schedule and not a list of features. Each one is written as a design commitment or an open question, because that is what the project can honestly support today. None of them describe something WebCompute has already achieved.

1. Goal 1

   Portable computation across existing Web capabilities

   Describe a piece of computation once, independently of a specific execution mechanism, so that a runtime could choose among JavaScript, WebAssembly, and WebGPU on a per-device basis instead of the application choosing in advance.

   Investigate whether that choice can be made with low enough overhead to be worth having. This is a research question, not an assumption the project starts from.

2. Goal 2

   Incremental research and implementation

   Proceed in small, falsifiable steps: a canonical repository and explainer, foundation reports, a scoped first experiment, and only then a decision about runtime implementation.

   No stage should be skipped to reach a demo faster, and no stage should be treated as final before later evidence has tested it.

3. Goal 3

   Integration with the existing Web platform

   Build on JavaScript, WebAssembly, WebGPU, Web Workers, and the capabilities browsers already expose, rather than introducing a new language, a new runtime environment, or a new hardware abstraction layer.

   Any future API should read as an addition to the platform a developer already knows, not as a parallel platform next to it.

4. Goal 4

   Safety, privacy, and no new hardware exposure

   Stay inside the security and privacy boundaries the Web platform already enforces.

   Do not expose raw hardware information, introduce new fingerprinting surface, or enumerate device capabilities beyond what a browser already reveals to a page.

5. Goal 5

   Open, testable, evidence-backed development

   Publish the research reports, the explainer, and, eventually, the experiment results in the open, including results that do not support the hypothesis.

   Treat every claim as testable. A claim with no way to falsify it does not belong in the research record.

These goals can change as the foundation reports are written. A goal that turns out to be wrong, or impossible to satisfy alongside the others, should be revised in the open rather than quietly dropped.

See also the FAQ for direct answers about what WebCompute does and does not replace, and the canonical repository for the requirements and non-goals report these goals are drawn from.