repo-scoping/tests/test_quality_gates.py

from repo_scoping.acceptance import (
    blocking_quality_gate_outcomes,
    evaluate_candidate_capability_quality,
    evaluate_candidate_graph_quality,
    quality_gate_outcome_dicts,
)
from repo_scoping.core.models import (
    AnalysisRun,
    CandidateAbility,
    CandidateCapability,
    CandidateFeature,
    CandidateGraph,
    Repository,
    SourceReference,
)
from repo_scoping.core.service import RegistryService
from repo_scoping.repo_ingestion.git import GitIngestionService
from repo_scoping.storage.sqlite import RegistryStore


def source_ref(path="src/app.py", kind="source"):
    return SourceReference(
        fact_id=1,
        path=path,
        kind=kind,
        name=path,
        line=1,
    )


def provider_routing_capability():
    return CandidateCapability(
        id=10,
        name="Route LLM Requests Across Providers",
        description="Routes provider requests.",
        inputs=[],
        outputs=[],
        confidence=0.9,
        status="candidate",
        source_refs=[source_ref("src/providers.py")],
        confidence_label="high",
        primary_class="llm-integration",
        attributes=["utility-owned"],
        features=[
            CandidateFeature(
                id=20,
                name="HTTP API surface",
                type="API",
                location="src/app.py",
                confidence=0.8,
                status="candidate",
                source_refs=[source_ref("src/app.py")],
                confidence_label="high",
                primary_class="API",
            )
        ],
    )


def test_quality_gates_flag_known_provider_routing_failure():
    outcomes = evaluate_candidate_capability_quality(provider_routing_capability())

    outcome_ids = {outcome.criterion_id for outcome in outcomes}
    assert {"RREG-QC-002", "RREG-QC-003"} <= outcome_ids
    assert all(outcome.outcome != "approve" for outcome in outcomes)
    assert blocking_quality_gate_outcomes(outcomes)


def test_quality_gates_flag_circular_scope_evidence():
    capability = CandidateCapability(
        id=11,
        name="Map Repository Scope",
        description="Uses generated scope.",
        inputs=[],
        outputs=[],
        confidence=0.8,
        status="candidate",
        source_refs=[source_ref("SCOPE.md", "generated-scope")],
        confidence_label="high",
        primary_class="scope-generation",
        attributes=["utility-owned"],
    )

    outcomes = evaluate_candidate_capability_quality(capability)

    assert outcomes[0].criterion_id == "RREG-QC-005"
    assert outcomes[0].outcome == "requires_review"


def test_quality_gates_flag_scope_derived_candidates_for_review():
    capability = CandidateCapability(
        id=12,
        name="Application workload deployment",
        description="Extracted from SCOPE.md.",
        inputs=[],
        outputs=[],
        confidence=0.6,
        status="candidate",
        source_refs=[source_ref("SCOPE.md", "scope")],
        confidence_label="medium",
        primary_class="infrastructure",
        attributes=["scope-derived", "review-required-scope"],
    )

    outcomes = evaluate_candidate_capability_quality(capability)

    outcome_ids = {outcome.criterion_id for outcome in outcomes}
    assert {"RREG-QC-005"} <= outcome_ids
    assert all(outcome.outcome == "requires_review" for outcome in outcomes)


def test_quality_gates_flag_template_contaminated_abilities():
    graph = CandidateGraph(
        repository=Repository(
            id=1,
            name="Ops Warden",
            url=".",
            description=None,
            branch="main",
            status="analyzed",
        ),
        analysis_run=AnalysisRun(
            id=1,
            repository_id=1,
            snapshot_id=None,
            status="completed",
            started_at="2026-05-15T00:00:00Z",
            completed_at="2026-05-15T00:00:01Z",
            error_message=None,
            scanner_version="deterministic-v1",
        ),
        abilities=[
            CandidateAbility(
                id=1,
                name="A Git Repository Template To Bootstrap Coulomb Projects",
                description="Derived from repo-seed README boilerplate.",
                confidence=0.7,
                status="candidate",
                source_refs=[source_ref("README.md", "documentation")],
            )
        ],
    )

    outcomes = evaluate_candidate_graph_quality(graph)

    assert outcomes[0].criterion_id == "RREG-QC-007"
    assert outcomes[0].outcome == "downgraded"


def test_quality_gate_outcomes_are_serializable_for_assessment_artifacts():
    graph = CandidateGraph(
        repository=Repository(
            id=1,
            name="Repo",
            url=".",
            description=None,
            branch="main",
            status="indexed",
        ),
        analysis_run=AnalysisRun(
            id=1,
            repository_id=1,
            snapshot_id=None,
            status="completed",
            started_at="2026-05-15T00:00:00Z",
            completed_at="2026-05-15T00:00:01Z",
            error_message=None,
            scanner_version="deterministic-v1",
        ),
        abilities=[
            CandidateAbility(
                id=1,
                name="Support Repo",
                description="Support repo.",
                confidence=0.8,
                status="candidate",
                source_refs=[],
                capabilities=[provider_routing_capability()],
            )
        ],
    )

    payload = quality_gate_outcome_dicts(evaluate_candidate_graph_quality(graph))

    assert payload
    assert payload[0]["criteria_version"] == "repo-scoping-quality-criteria/v1"


def test_legacy_trusted_auto_approval_skips_quality_gate_blocked_capability(tmp_path):
    store = RegistryStore(tmp_path / "registry.sqlite3")
    store.initialize()
    service = RegistryService(store, ingestion=GitIngestionService(tmp_path / "checkouts"))

    safe, reason = service._trusted_auto_approve_capability_decision(
        provider_routing_capability()
    )

    assert safe is False
    assert "quality gates require review" in reason
    assert "RREG-QC-002" in reason


def test_analysis_records_deterministic_gate_review_decision(tmp_path):
    source = tmp_path / "provider-repo"
    source.mkdir()
    (source / "README.md").write_text("# Provider Repo\n", encoding="utf-8")
    (source / "providers.py").write_text(
        "provider_registry = {'openrouter': OpenRouterAdapter}\n",
        encoding="utf-8",
    )
    store = RegistryStore(tmp_path / "registry.sqlite3")
    store.initialize()
    service = RegistryService(store, ingestion=GitIngestionService(tmp_path / "checkouts"))
    repository = service.register_repository(name="Provider Repo", url=str(source))

    summary = service.analyze_repository(repository.id, use_llm_assistance=False)

    decisions = service.list_review_decisions(repository.id, summary.analysis_run.id)
    gate_decision = next(
        decision for decision in decisions if decision.action == "quality_gate_evaluation"
    )
    assert gate_decision.reviewer_type == "deterministic-gate"
    assert "RREG-QC-002" in gate_decision.criterion_ids
    assert gate_decision.criteria_version == "repo-scoping-quality-criteria/v1"
    assert "without approving registry truth" in gate_decision.rationale
    assert service.ability_map(repository.id).abilities == []

    override = service.record_quality_gate_override(
        repository.id,
        summary.analysis_run.id,
        criterion_id="RREG-QC-002",
        element_type="capability",
        element_id=10,
        reason="Curator confirmed this repo now owns provider routing.",
        notes="Future criteria update may be needed.",
    )
    assert override.action == "quality_gate_override"
    assert override.reviewer_type == "human"
    assert override.decision_kind == "override"
    assert override.criterion_ids == ["RREG-QC-002"]
    assert override.rationale == "Curator confirmed this repo now owns provider routing."