cs249r_book/mlperf-edu/RELEASE_CHECKLIST.md

MLPerf EDU Release Checklist

This checklist separates packaging readiness from MLCommons endorsement readiness. A local teaching release can ship before every public-result policy question is closed, but the release notes must say that clearly.

0.1 Preview Release Bar

Status	Item	Evidence
Done	Public CLI is mlperf	pyproject.toml exposes mlperf = "mlperf.edu_cli:main"
Done	uv install path exists	INSTALL.md documents uv sync, uv tool install, and uv build
Done	Wheel has registry metadata	src/mlperf_edu/workloads.yaml is a real packaged data file
Done	Native registry has generated flat mirror	tools/export_flat_registry.py --check checks root and packaged mirrors
Done	Local audit is clean	mlperf audit passes with 0 local warnings
Done	Local validation is clean	latest recorded validate release passed 60/60 with 0 local warnings
Done	HTML/JSON/CSV/provenance artifacts are default	run, init, validate, report, and grade paths emit them
Done	Review packets exist	review_packets/ has nine current public-result candidate packets
Open	Publishable license statement for the component	Add or point to the authoritative MLPerf EDU component license before external package publication
Open	Fresh Linux package smoke in CI	GitHub Actions workflow exists; confirm it passes after pushing

1.0 Public/Endorsement Release Bar

Status	Item	Decision Needed
Open	mlperf audit --policy public passes	Resolve MovieLens-100K approval or replace the score-bearing recommender dataset
Open	Quality targets approved	Experts sign off on score-bearing and performance-bearing target rationale
Open	Public result wording approved	MLCommons decides whether MLPerf EDU can use endorsed/result language
Open	Dataset/model attribution checked	Every public asset has source, license, citation, fetch policy, and report fields
Open	Reproducibility packet reviewed	At least one external fresh-machine run checks install, fetch, run, report, grade, and verify
Open	Release notes written	Clearly distinguish MLPerf EDU teaching/research results from official competitive MLPerf submissions

Non-Blocking Follow-Ups

• Add a published package index target only after the license statement is final.
• Add optional backend extras for ONNX Runtime, MLX, and llama.cpp once the corresponding runners are stable enough to be user-facing.
• Add comparison tooling for repeated pro runs after the base install path is stable.