docs: add dataset-label draft examples

.github/workflows/ci.yml

@@ -70,7 +70,7 @@ jobs:
         id: drafts
         uses: ./
         with:
-          files: examples/drafts/*.yaml
+          files: examples/drafts/*claim.yaml
           profile: literature-claim
           format: draft
           fail-on: rejected

CHANGELOG.md

@@ -2,6 +2,9 @@
 
 ## Unreleased
 
+- Add admitted and rejected `dataset-label` draft examples, covered by draft and
+  installed-wheel checks.
+
 ## 0.7.0 — Expert review, quality checks, community and documentation site
 
 - Add `bioevidence review` (`check`, `agreement`, `adjudication-sheet`, `score`, `freeze`)

examples/drafts/dataset_label_reference.yaml

@@ -0,0 +1,35 @@
+# Synthetic dataset-label draft admitted for reference annotation.
+# Build:    bioevidence build examples/drafts/dataset_label_reference.yaml --output record.json
+# Validate: bioevidence validate record.json --profile dataset-label   (exit 0: admitted)
+profile: dataset-label
+uses: [reference_annotation]
+
+statement:
+  subject: {id: "SYN:SAMPLE_A", label: Synthetic sample A, type: sample}
+  predicate: has_label
+  object: {id: "SYN:LABEL_A", label: Synthetic class A, type: class_label}
+  scope: ["cohort:synthetic"]
+
+sources:
+  - id: labels
+    title: Synthetic label table; not real biological evidence
+    type: dataset_snapshot
+    uri: https://example.org/synthetic-labels
+    version: synthetic-v1
+    retrieved_at: "2026-09-25T00:00:00Z"
+    sha256: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+
+evidence:
+  - source: labels
+    locator: row-1-label
+    text: Synthetic sample A has synthetic class A.
+    type: curated_label
+    method: manual_curation
+    scope: ["cohort:synthetic"]
+
+  - source: labels
+    locator: row-1-link
+    text: Synthetic sample A is linked to row 1 of the label table.
+    type: sample_link
+    method: deterministic_parser
+    scope: ["cohort:synthetic"]

examples/drafts/dataset_label_training_unreviewed.yaml

@@ -0,0 +1,35 @@
+# Synthetic dataset-label draft rejected for training use without human acceptance.
+# Build:    bioevidence build examples/drafts/dataset_label_training_unreviewed.yaml --output record.json
+# Validate: bioevidence validate record.json --profile dataset-label   (exit 1: rejected)
+profile: dataset-label
+uses: [training_data]
+
+statement:
+  subject: {id: "SYN:SAMPLE_A", label: Synthetic sample A, type: sample}
+  predicate: has_label
+  object: {id: "SYN:LABEL_A", label: Synthetic class A, type: class_label}
+  scope: ["cohort:synthetic"]
+
+sources:
+  - id: labels
+    title: Synthetic label table; not real biological evidence
+    type: dataset_snapshot
+    uri: https://example.org/synthetic-labels
+    version: synthetic-v1
+    retrieved_at: "2026-09-25T00:00:00Z"
+    sha256: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+
+evidence:
+  - source: labels
+    locator: row-1-label
+    text: Synthetic sample A has synthetic class A.
+    type: curated_label
+    method: manual_curation
+    scope: ["cohort:synthetic"]
+
+  - source: labels
+    locator: row-1-link
+    text: Synthetic sample A is linked to row 1 of the label table.
+    type: sample_link
+    method: deterministic_parser
+    scope: ["cohort:synthetic"]

tests/test_draft.py

@@ -39,6 +39,19 @@ def test_reviewed_draft_is_admitted_for_both_uses():
     assert overall == "admitted" and uses == {"research_summary": "admitted", "knowledge_base": "admitted"}
 
 
+def test_dataset_label_reference_draft_is_admitted():
+    overall, uses, _ = statuses(draft("dataset_label_reference.yaml"))
+    assert overall == "admitted"
+    assert uses == {"reference_annotation": "admitted"}
+
+
+def test_dataset_label_training_draft_requires_human_acceptance():
+    overall, uses, report = statuses(draft("dataset_label_training_unreviewed.yaml"))
+    assert overall == "rejected"
+    assert uses == {"training_data": "rejected"}
+    assert {f["rule_id"] for f in report["findings"]} == {"BEV010"}
+
+
 def test_build_expands_structure_and_hashes_local_file():
     record = build(draft())
     source, = record["source_artifacts"]
@@ -138,9 +151,17 @@ def test_draft_schema_is_valid_json_schema(profile):
     assert schema["properties"]["profile"] == {"const": profile}
 
 
-@pytest.mark.parametrize("name", ["llm_claim.yaml", "reviewed_claim.yaml"])
-def test_example_drafts_match_their_profile_schema(name):
-    Draft202012Validator(draft_json_schema("literature-claim")).validate(draft(name))
+@pytest.mark.parametrize(
+    "name,profile",
+    [
+        ("llm_claim.yaml", "literature-claim"),
+        ("reviewed_claim.yaml", "literature-claim"),
+        ("dataset_label_reference.yaml", "dataset-label"),
+        ("dataset_label_training_unreviewed.yaml", "dataset-label"),
+    ],
+)
+def test_example_drafts_match_their_profile_schema(name, profile):
+    Draft202012Validator(draft_json_schema(profile)).validate(draft(name))
 
 
 def test_draft_schema_turns_profile_allowlists_into_enums():

tests/test_github_action.py

@@ -22,7 +22,7 @@ def run(monkeypatch, tmp_path, *args):
 
 @pytest.mark.parametrize("fail_on,expected", [("review", 1), ("rejected", 0)])
 def test_drafts_fail_on_review_only_when_asked(monkeypatch, tmp_path, fail_on, expected):
-    code, summary, counts = run(monkeypatch, tmp_path, "--files", "examples/drafts/*.yaml",
+    code, summary, counts = run(monkeypatch, tmp_path, "--files", "examples/drafts/*claim.yaml",
                                 "--profile", "literature-claim", "--format", "draft", "--fail-on", fail_on)
     assert code == expected
     assert counts == {"admitted": 1, "review_required": 1, "rejected": 0, "error": 0}

tools/check_distribution.py

@@ -44,10 +44,15 @@ def run(*args, expected=0):
     run("generate-schema", "--output", output)
     assert "BioEvidenceRecord" in json.loads(output.read_text(encoding="utf-8"))["$defs"]
 with tempfile.TemporaryDirectory() as directory:
-    for draft, expected in [("llm_claim.yaml", 2), ("reviewed_claim.yaml", 0)]:
+    for draft, profile, expected in [
+        ("llm_claim.yaml", "literature-claim", 2),
+        ("reviewed_claim.yaml", "literature-claim", 0),
+        ("dataset_label_reference.yaml", "dataset-label", 0),
+        ("dataset_label_training_unreviewed.yaml", "dataset-label", 1),
+    ]:
         output = Path(directory) / (draft + ".json")
         run("build", root / "examples/drafts" / draft, "--output", output)
-        run("validate", output, "--profile", "literature-claim", expected=expected)
+        run("validate", output, "--profile", profile, expected=expected)
     schema = json.loads(run("draft-schema", "--profile", "literature-claim").stdout)
     assert schema["properties"]["profile"] == {"const": "literature-claim"}
 record = json.loads((root / "examples/general/curated_assertion.json").read_text(encoding="utf-8"))