[TohruSu]

Make the repository easier to assess and run: the README now explains the single-worker scheduling problem, the rule-selection workflow, and the observed experimental tradeoffs before the setup commands.

• Add a concise Mermaid workflow, the committed scheduling-cost figure, and a compact table showing Priority alongside all three independent RL runs. Distinguish primary weighted scheduling cost from secondary metrics and link to the full analysis and CSVs.
• Preserve the technical documentation, clarifying the upper bound on state combinations and the difference between pooled evaluation metrics and benchmark episode averages.
• Declare MIT using SPDX metadata, include the existing license text in distributions, and add repository/documentation package URLs. Raise the setuptools build minimum to support this metadata.
• Extend the existing Python 3.10/3.12 CI matrix with wheel build/install/CLI smoke checks and explicit read-only repository permissions. Ignore the local model produced by the README example.

Validation: all 81 tests, Ruff lint/format, and strict mypy passed on Python 3.10.22 and 3.12.14. Wheels built on both versions, included the MIT license, and ran both CLI entry points outside the checkout. A fresh core installation ran without Matplotlib. Documented training, evaluation, help, and benchmark commands ran successfully; the temporary benchmark reproduced all nine committed reference artifacts byte-for-byte. All local documentation links/anchors and README result values were checked against committed files.

The six-file change contains documentation and packaging/CI settings. Scheduling code, reference outputs, and saved Q-tables retain their existing contents. The repository remains private.