llm_extraction integration by adding the bridge into candidate graph drafts

2026-04-26 03:08:55 +02:00
parent 7e66c57350
commit 3aa0c08ab9
4 changed files with 261 additions and 2 deletions
--- a/tests/test_llm_extraction_mapper.py
+++ b/tests/test_llm_extraction_mapper.py
@@ -0,0 +1,110 @@
+from repo_registry.core.models import ContentChunk, ObservedFact
+from repo_registry.llm_extraction import (
+    ExtractedAbility,
+    ExtractedCapability,
+    ExtractedEvidence,
+    ExtractedFeature,
+    LLMExtractionMapper,
+)
+
+
+def fact(id, kind, name, path, line=None):
+    metadata = {}
+    if line is not None:
+        metadata["line"] = line
+    return ObservedFact(
+        id=id,
+        repository_id=1,
+        analysis_run_id=1,
+        snapshot_id=1,
+        kind=kind,
+        path=path,
+        name=name,
+        value="",
+        metadata=metadata,
+    )
+
+
+def chunk(id, kind, path, start_line=1):
+    return ContentChunk(
+        id=id,
+        repository_id=1,
+        analysis_run_id=1,
+        snapshot_id=1,
+        path=path,
+        kind=kind,
+        start_line=start_line,
+        end_line=start_line + 2,
+        text="source text",
+    )
+
+
+def test_llm_extraction_mapper_builds_candidate_drafts_with_source_refs():
+    extracted = [
+        ExtractedAbility(
+            name="Business Email Routing",
+            description="Routes email.",
+            source_paths=["README.md"],
+            capabilities=[
+                ExtractedCapability(
+                    name="Classify Incoming Email",
+                    description="Classifies email.",
+                    inputs=["email body"],
+                    outputs=["intent"],
+                    source_paths=["README.md"],
+                    features=[
+                        ExtractedFeature(
+                            name="POST /classify",
+                            type="REST endpoint",
+                            location="app.py",
+                            source_paths=["app.py"],
+                        )
+                    ],
+                    evidence=[
+                        ExtractedEvidence(
+                            type="test",
+                            reference="tests/test_app.py",
+                            strength="strong",
+                            source_paths=["tests/test_app.py"],
+                        )
+                    ],
+                )
+            ],
+        )
+    ]
+    facts = [
+        fact(1, "documentation", "README", "README.md"),
+        fact(2, "interface", "python route decorator", "app.py", line=4),
+        fact(3, "test", "test_app.py", "tests/test_app.py"),
+    ]
+
+    candidates = LLMExtractionMapper().map(extracted, facts, [])
+
+    ability = candidates[0]
+    capability = ability.capabilities[0]
+    assert ability.name == "Business Email Routing"
+    assert ability.confidence == 0.7
+    assert ability.source_refs[0].fact_id == 1
+    assert capability.inputs == ["email body"]
+    assert capability.outputs == ["intent"]
+    assert capability.features[0].source_refs[0].line == 4
+    assert capability.evidence[0].source_refs[0].kind == "test"
+
+
+def test_llm_extraction_mapper_can_use_chunk_refs_without_facts():
+    extracted = [
+        ExtractedAbility(
+            name="Readme Ability",
+            source_paths=["README.md:1-3"],
+        )
+    ]
+
+    candidates = LLMExtractionMapper().map(
+        extracted,
+        facts=[],
+        chunks=[chunk(1, "documentation", "README.md", start_line=1)],
+    )
+
+    assert candidates[0].source_refs[0].fact_id is None
+    assert candidates[0].source_refs[0].path == "README.md"
+    assert candidates[0].source_refs[0].line == 1