8 October 2026. These are small synthetic checks performed for AI News and Updates, with runnable code and recorded results. They are not cloud, performance or purchasing benchmarks.
Python 3.12.14, macOS 26.3.1, ARM64. An isolated Python environment used packages installed from PyPI. Ray used two local CPU workers and no dashboard or GPU. W&B used offline mode with synthetic values, no login and no cloud sync.
| Package | Version | Passing checks |
|---|---|---|
| ray | 2.59.0 | eight CPU task results; actor state retained across calls; task exception reaches caller |
| wandb | 0.30.0 | offline mode selected; synthetic configuration and summary accepted; nonempty local run stream contains metric and artifact identifiers |
Ray returned eight expected square values, retained actor state across three increments and propagated an intentional task exception. The first attempt encountered the macOS Unix socket path-length limit; a shorter temporary directory resolved that startup issue. No multiple-node scaling, GPU, Ray Serve or production recovery claim follows from this run.
W&B accepted synthetic configuration and a final-loss summary of 0.25. Its nonempty local run stream contained the submitted metric and artifact identifiers. The check did not independently reconstruct the complete metric history or test cloud retrieval, dashboards, collaboration, Registry or Weave.
In a new Python 3.12 environment install ray==2.59.0 wandb==0.30.0, download the script, then run python review-smoke-tests.py ray ./ray-output and python review-smoke-tests.py wandb ./wandb-output. The script writes JSON results, including failures. Later environments or package versions may behave differently.