[PandaHUN777]

What and why

Closes #7

Adds two synthetic dataset-label draft examples so users can start from the compact draft format instead of a full record:

• dataset_label_reference.yaml contains the required curated_label and sample_link evidence for reference_annotation and is admitted.
• dataset_label_training_unreviewed.yaml contains the same evidence for training_data but deliberately has no human acceptance, so it is rejected with BEV010.

Each draft header includes the build/validate commands and expected exit code. The draft test suite pins both outcomes and validates both examples against the generated dataset-label draft schema. The installed-wheel smoke check now builds and validates the examples too.

Admission boundary

No validation rule changes. This adds examples that exercise the existing dataset-label admission boundary: reference annotation does not require human acceptance, while training data does.

• No change to admission decisions
• Changes admission decisions (explained above; tests on both sides of the boundary)

Checklist

• uv run --frozen ruff check ., uv run --frozen mypy and uv run --frozen pytest --cov pass
• Committed benchmark results regenerated if report contents changed — not applicable: validation reports are unchanged
• CHANGELOG.md updated if users will notice
• New third-party data has its license, attribution and exact version recorded — not applicable: examples are synthetic only

Full repository CI is left to the upstream pull-request checks.

[NingyuSUN]

Thanks for #7 — nicely done.

Verified: both drafts match the dataset-label profile (predicate has_label, sample → class_label, required curated_label + sample_link); reference_annotation is admitted and training_data is rejected with BEV010 for missing human acceptance — exactly the boundary the issue asked to show. The workflow and test glob change (*claim.yaml) is necessary and minimal, and those were the only two globs over examples/drafts/. The smoke test covers both new drafts. CI passes on all jobs.

One small change before merging: CHANGELOG placement. 0.7.0 was released after you opened this and its "Unreleased" section became "0.7.0", so your line would land inside the published 0.7.0 section without any conflict. Please move it under the new ## Unreleased heading at the top of CHANGELOG.md (added in #16).

Happy to merge once that's in.

[PandaHUN777]

Thanks for the review, @NingyuSUN!

I have moved the dataset-label draft examples entry under the top-level ## Unreleased section in CHANGELOG.md.

Validated locally with:

• uv run --frozen ruff check .
• uv run --frozen mypy
• uv run --frozen pytest --cov --cov-report=term (234 tests passed, 98.09% coverage)
• uv build
• uv run --isolated --no-project --with ./dist/*.whl python tools/check_distribution.py
• git diff --check