Historical note (7 November 2026): this was written when Switch homebrew was the project's focus. The project has since moved to ARM64 verification against real hardware; see the vision. The method described here (register the expected output before the run) still applies.

Research note · 7 November 2026

Beyond the First Frame: measuring progress rather than promising compatibility

An early experimental runtime is easy to describe through a list of implemented subsystems. A stronger scientific question is: when an independently authored program runs, can its output be checked against an expectation registered before the run?

The engineering shift

Ounce-Core has evolved from CPU, memory and service models into a larger testing workflow: permitted external homebrew sources pinned to exact revisions, bounded execution with recorded guest results, first-frame pixel oracles, and verification records preserved for comparisons and diagnoses. This is a research workflow, not a guarantee that arbitrary programs work.

Why 50 registered programs are not 50 independent users

The registered research corpus contains work on 50 external homebrew programs. The set's size, its currently passing count, and the number of people outside the maintainers who have independently repeated the results are three different metrics. The latest canonical CI evidence, not this dated article, determines the pass count. Inspect the live workflow and dated plan.

What a strong proof looks like

  1. An independently maintained and legally redistributable input, locked to a revision and build environment.
  2. A registered expected output fixed before the runtime is executed.
  3. Unchanged interpreter and Tier-3 observations, with artifact hashes and raw records.
  4. A verification step that does not silently discard errors, missing observations, or failed expectations.
  5. A repeat run on a different developer's machine; discrepancies are published, not hidden.

Why this matters outside a console context

Inspecting how instructions, memory, operating-system services and graphics interact has potential educational and systems-research value. Ounce-Verify's evidence discipline is also relevant to debugging and testing methodology. These are research directions, not assertions of adopted commercial use or a finished teaching platform.

Scope and source material

Ounce-Core is experimental and unaffiliated with Nintendo or NVIDIA. Switch 2 work is parked per PLAN.md; neither original-Switch homebrew oracles nor limited experimental Switch 2 research prove broad Switch 2 compatibility or retail-game execution. Explore the oracle methodology, Ounce-Verify, the shipped ounce-diagnose v0.2.0 toolkit, and the provenance record.