additive candidate confidence scoring

src/repo_registry/candidate_graph/generator.py

@@ -68,7 +68,14 @@ class CandidateGraphGenerator:
         ability = CandidateAbilityDraft(
             name=f"Review {repository.name} Repository Usefulness",
             description=self._ability_description(chunks),
-            confidence=0.55 if docs else 0.35,
+            confidence=self._ability_confidence(
+                docs=docs,
+                interfaces=interfaces,
+                tests=tests,
+                examples=examples,
+                frameworks=frameworks,
+                languages=languages,
+            ),
             source_refs=self._source_refs(ability_sources),
             capabilities=[],
         )
@@ -88,7 +95,12 @@ class CandidateGraphGenerator:
                     ),
                     inputs=[],
                     outputs=["repository structure summary"],
-                    confidence=0.6,
+                    confidence=self._structure_confidence(
+                        manifests=manifests,
+                        frameworks=frameworks,
+                        languages=languages,
+                        docs=docs,
+                    ),
                     source_refs=self._source_refs(manifests + frameworks + languages),
                     evidence=self._evidence(tests, examples, docs),
                 )
@@ -127,7 +139,12 @@ class CandidateGraphGenerator:
             description=self._interface_description(chunks),
             inputs=[],
             outputs=["callable interface"],
-            confidence=0.65,
+            confidence=self._interface_confidence(
+                interfaces=interfaces,
+                tests=tests,
+                examples=examples,
+                docs=docs,
+            ),
             source_refs=self._source_refs(interfaces),
             features=features,
             evidence=self._evidence(tests, examples, docs),
@@ -177,6 +194,73 @@ class CandidateGraphGenerator:
             return "API"
         return "interface"
 
+    def _ability_confidence(
+        self,
+        *,
+        docs: list[ObservedFact],
+        interfaces: list[ObservedFact],
+        tests: list[ObservedFact],
+        examples: list[ObservedFact],
+        frameworks: list[ObservedFact],
+        languages: list[ObservedFact],
+    ) -> float:
+        return self._confidence(
+            0.25,
+            [
+                (0.20, bool(docs)),
+                (0.15, bool(interfaces)),
+                (0.15, bool(tests)),
+                (0.10, bool(examples)),
+                (0.10, bool(frameworks)),
+                (0.05, bool(languages)),
+            ],
+        )
+
+    def _interface_confidence(
+        self,
+        *,
+        interfaces: list[ObservedFact],
+        tests: list[ObservedFact],
+        examples: list[ObservedFact],
+        docs: list[ObservedFact],
+    ) -> float:
+        return self._confidence(
+            0.30,
+            [
+                (0.20, bool(interfaces)),
+                (0.15, bool(tests)),
+                (0.10, bool(examples)),
+                (0.10, bool(docs)),
+                (0.05, len(interfaces) > 1),
+            ],
+        )
+
+    def _structure_confidence(
+        self,
+        *,
+        manifests: list[ObservedFact],
+        frameworks: list[ObservedFact],
+        languages: list[ObservedFact],
+        docs: list[ObservedFact],
+    ) -> float:
+        return self._confidence(
+            0.25,
+            [
+                (0.20, bool(manifests)),
+                (0.15, bool(frameworks)),
+                (0.10, bool(languages)),
+                (0.05, bool(docs)),
+            ],
+        )
+
+    def _confidence(
+        self,
+        base: float,
+        factors: list[tuple[float, bool]],
+    ) -> float:
+        score = base + sum(weight for weight, applies in factors if applies)
+        return min(1.0, round(score, 2))
+
     def _ability_description(self, chunks: list[ContentChunk]) -> str:
         doc_summary = self._document_summary(chunks)
         if doc_summary: