Two versions, one set of tested examples
GitHub Actions builds each interpreter at its pinned source commit and runs the same three Scarp programs. Each must exit successfully, produce exactly the checked-in stdout fixture, and pass its version-appropriate source validation command without unexpected output.
| Compiler | Run program | Validate source | Checked result |
|---|---|---|---|
| v0.0.5-alpha.1 | scarp run hello.srp | scarp check hello.srpLegacy evaluating check | Verified fixtures |
| v0.0.6-alpha.3 | scarp exec hello.srp | scarp verify hello.srpStatic-only verification | Verified fixtures |
Important: the historical check command in v0.0.5-alpha.1 can evaluate program statements; it is not the same as the static-only verify command shipped since alpha.3.
Changing CLI versions? The version-by-version migration guide explains which spellings are available in each published alpha and why check is not interchangeable with static-only verify.
Exactly what was tested
These are the files rendered in the homepage code explorer. The newer compiler passed the existing fixtures; that does not imply that all Scarp syntax, old CLI commands, libraries or other projects are compatible.
Published alpha.6-only numerical coverage
Additional examples for vector rms, deterministic topk ranking, and temperature-aware softmax_temp/logsoftmax_temp run only with v0.0.6-alpha.3, not with the older baseline. Their exact stdout is checked, and inspect symbols and inspect references JSON outputs are checked for schema version 1 and declared kinds. These tests do not certify the whole symbol-index API or numerical library.
Evidence and reproducibility
The two exact compiler source revisions, checked on an Ubuntu GitHub Actions runner, are:
- v0.0.5-alpha.1:
afe8b5380c9d6c4774162e51c6c9bc8b90901671 - v0.0.6-alpha.3:
ceff07d9014f466974e0c9e02fc5d23573f55777
See the automated CI workflow and recent runs, versioned fixture directory, and test runner. Future changes to examples or compiler versions must pass these checks before deployment.
What these results do not prove
- They do not exercise every language feature or source module.
- They run in Linux CI; separate compiler release packaging covers other platforms.
- They do not test GPU execution, native code generation, or a stable package ecosystem.
- They do not make prerelease syntax, manifests, or APIs stable.
The site's historical documentation and example baseline remains v0.0.5-alpha.1. The separately published v0.0.6-alpha.3 has also passed these three fixtures. Read the release comparison or download the version you intend to use.