fix+test: ensemble missing-key crash and eligibility boundary condition

src/openamp_foundry/scoring/ensemble.py

@@ -1,9 +1,18 @@
 from __future__ import annotations
 
+import warnings
+
 
 def ensemble_score(scores: dict[str, float], weights: dict[str, float]) -> float:
+    missing = [name for name in weights if name not in scores]
+    if missing:
+        warnings.warn(
+            f"ensemble_score: weight keys missing from scores, defaulting to 0.0: {missing}",
+            UserWarning,
+            stacklevel=2,
+        )
     total_weight = sum(weights.values()) or 1.0
-    value = sum(scores[name] * weight for name, weight in weights.items()) / total_weight
+    value = sum(scores.get(name, 0.0) * weight for name, weight in weights.items()) / total_weight
     return round(max(0.0, min(1.0, value)), 4)
 
 

tests/test_pipeline_filters.py

@@ -130,6 +130,40 @@ def test_novelty_filter_excludes_near_duplicates(tmp_path):
     assert rows[0]["scores"]["novelty"] == 0.0
 
 
+def test_eligibility_novelty_at_exact_minimum_passes(tmp_path):
+    """A candidate with novelty exactly = min_novelty (>= boundary) must be selected.
+
+    Uses a custom config with min_novelty=0.0 so that even a candidate identical to
+    the reference (novelty=0.0) passes the filter.  Confirms the filter uses >=, not >.
+    """
+    config_path = tmp_path / "permissive.yaml"
+    config_path.write_text(
+        "pipeline_version: '0.1.0'\n"
+        "filters:\n  min_length: 8\n  max_length: 35\n"
+        "  allowed_amino_acids: 'ACDEFGHIKLMNPQRSTVWY'\n"
+        "weights:\n  activity: 0.40\n  safety: 0.25\n  synthesis: 0.15\n  novelty: 0.20\n"
+        "selection:\n  top_n: 100\n  min_novelty: 0.0\n  max_safety_risk: 1.0\n"
+    )
+    seq = "KWKLFKKIGAVLKVL"
+    csv = tmp_path / "candidates.csv"
+    refs = tmp_path / "refs.csv"
+    # Candidate is identical to the reference → novelty=0.0
+    csv.write_text(f"id,sequence,source\nTEST-001,{seq},test\n")
+    refs.write_text(f"id,sequence,source\nREF-001,{seq},reference\n")
+    out = tmp_path / "ranked.jsonl"
+    run_ranking_pipeline(
+        candidate_path=csv,
+        reference_path=refs,
+        out_path=out,
+        config_path=config_path,
+    )
+    rows = [json.loads(line) for line in out.read_text().splitlines() if line.strip()]
+    assert len(rows) == 1
+    assert rows[0]["scores"]["novelty"] == 0.0
+    # With min_novelty=0.0, novelty=0.0 satisfies 0.0 >= 0.0 → should be selected
+    assert rows[0]["selected"] is True
+
+
 def test_phase3_config_has_stricter_safety_filter_than_pipeline():
     """phase3.yaml max_safety_risk=0.40 (safety>=0.60) vs pipeline.yaml 0.70 (safety>=0.30).
 

tests/test_scoring_ensemble.py

@@ -56,6 +56,22 @@ def test_returns_rounded_to_4dp(self):
         result = ensemble_score(scores, {"activity": 1.0, "safety": 0.0})
         assert result == round(result, 4)
 
+    def test_missing_score_key_defaults_to_zero_and_warns(self):
+        # weights has "safety" but scores doesn't: should default to 0.0 AND emit UserWarning
+        # A missing key could be a scorer failure or a config typo — the warning makes it auditable.
+        # weighted = (1.0*0.40 + 0.0*0.30) / (0.40+0.30) ≈ 0.5714
+        import pytest as _pytest
+        with _pytest.warns(UserWarning, match="safety"):
+            result = ensemble_score({"activity": 1.0}, {"activity": 0.40, "safety": 0.30})
+        assert abs(result - 0.5714) < 0.001
+
+    def test_extra_score_key_not_in_weights_is_ignored(self):
+        # boman_activity is in scores but not in weights — should not affect result.
+        # This pre-existing behavior (only weighted keys matter) is unchanged by the .get() fix.
+        result_without = ensemble_score({"activity": 1.0}, {"activity": 1.0})
+        result_with = ensemble_score({"activity": 1.0, "boman_activity": 0.0}, {"activity": 1.0})
+        assert result_without == result_with
+
 
 class TestSelectionReasons:
     def _scores(