RELEASE_CHECKLIST.md

Release Checklist

This file tracks what should be included in the public repository and what still needs to happen before publication.

Scaffolding/ is intentionally local-only and should not be committed.

Release blocker first

• Confirm .env stays untracked and never appears in any commit
• Run a final secret scan over the repo before creating the first public commit

Notes:

• .gitignore excludes .env
• .gitignore excludes Scaffolding/
• local caches such as .uv_cache/, .pytest_cache/, and __pycache__/ should remain untracked

Public-ready file set

Core repo files intended to ship:

• README.md
• HUB_README.md
• LICENSE
• RELEASE_CHECKLIST.md
• PUBLIC_FILESET.md
• pyproject.toml
• openenv.yaml
• Dockerfile
• env/
• server/
• training/
• green_agent/
• tests/
• examples/
• results/baseline_table.md
• results/baseline_table.json
• results/benchmark_comparison.png
• results/deadline_crunch_reset.png
• results/deadline_crunch_after_three_steps.png
• results/web_dashboard_reset.png
• results/web_dashboard_after_step.png
• blog_draft.md
• demo_notebook.ipynb

Files that should stay local-only:

• .env
• Scaffolding/
• .uv_cache/
• .pytest_cache/
• __pycache__/
• results/*_smoke*

1. GitHub repo

Goal: make the repository public-ready and reproducible.

Checklist:

• Initialize or reconnect the real git repository if needed
• Verify no secrets are staged
• Commit the environment, trainer, tests, examples, and docs as one coherent public baseline
• Add a short release note in the root README that explains current status
• Verify python -m pytest, openenv validate, and python -m training.generate_demo_artifacts still pass from a clean checkout

Definition of done:

• Someone can clone the repo, install dependencies, and run the benchmark from the docs

2. Hub README

Goal: produce the Hub-facing environment card.

Checklist:

• Draft HUB_README.md
• Add one /web screenshot Available local assets: results/web_dashboard_reset.png, results/web_dashboard_after_step.png
• Add explicit Hub/Space usage instructions if the environment is published there
• Link back to the GitHub repo and blog post once those are live

Definition of done:

• A Hub browser can understand the environment in under one minute

3. Blog post

Goal: finish the narrative submission.

Checklist:

• Refresh the benchmark table so it includes the current RL result
• Add one section on why this is a better OpenEnv fit than a generic coding/browser benchmark
• Add one section on failure analysis for both the LLM baseline and the learned policy
• Add screenshots or charts from results/
• Add an honest limitations/future work section

Definition of done:

• The post can stand on its own as the main submission artifact

4. Demo notebook

Goal: create a notebook that walks a reader through the environment quickly.

Checklist:

• Create demo_notebook.ipynb
• Verify every cell runs top-to-bottom in a clean environment
• Include one reset snapshot, one manual stepped rollout, and one benchmark table read
• Include heuristic and learned-policy trace comparison
• Include an optional structured-output LLM rollout path

Definition of done:

• A reader can run the notebook and understand the task without reading the full repo first

Link placeholders

Fill these once the project is published:

• GitHub repo: https://github.com/SulmanK/gpu-scheduler-openenv
• Hugging Face environment/Space: <HF_SPACE_OR_HUB_URL>
• Blog post: <BLOG_POST_URL>

Update these files after publishing:

• README.md
• HUB_README.md
• blog_draft.md or final published blog source