fix: InternalCorpusProvider last_error; None novelty/cf scores crash report agent

memory/graph_memory.py

@@ -344,7 +344,7 @@ class GraphMemory(BaseMemory):
         concerns = r.get("concerns", []) or []
         strengths = r.get("strengths", []) or []
         lines = [
-            f"Overall counterfactuality score: {r.get('overall_score', 0):.2f}/1.0",
+                f"Overall counterfactuality score: {(r.get('overall_score') or 0):.2f}/1.0",
             f"Most common type: {r.get('common_type', 'factual')}",
             f"Summary: {r.get('summary', '')}",
         ]
@@ -369,7 +369,7 @@ class GraphMemory(BaseMemory):
             for cr in claim_rows:
                 lines.append(
                     f"  - [{cr.get('cf_type', 'unknown')}] "
-                    f"(score={cr.get('cf_score', 0):.2f}) {cr.get('text', '')}"
+                    f"(score={(cr.get('cf_score') or 0):.2f}) {cr.get('text', '')}"
                 )
 
         return format_context_block("Counterfactuality Analysis", "\n".join(lines))
@@ -389,7 +389,7 @@ class GraphMemory(BaseMemory):
             return ""
         r = rows[0]
         lines = [
-            f"Novelty score: {r.get('novelty_score', 0):.1f}/10 ({r.get('verdict', 'UNKNOWN')})",
+            f"Novelty score: {(r.get('novelty_score') or 0):.1f}/10 ({r.get('verdict', 'UNKNOWN')})",
             f"Problem statement: {r.get('problem_statement', '')}",
             f"Rationale: {r.get('rationale', '')}",
         ]

similarity/corpus/internal.py

@@ -45,6 +45,7 @@ class InternalCorpusProvider:
         self._job_id = job_id
         self._k = k or settings.kgram_size
         self._w = w or settings.window_size
+        self.last_error: str | None = None
 
     async def search_phrase(self, phrase: str, limit: int) -> list[Candidate]:
         if not settings.enable_internal:

tests/similarity/test_privacy.py

@@ -55,6 +55,13 @@ def test_neutral_label_without_date():
     assert _submitted_label(None) == "CitationEdge Corpus · submitted unknown-date"
 
 
+def test_internal_provider_exposes_last_error(internal_provider):
+    """Regression: pipeline reads provider.last_error; InternalCorpusProvider
+    must define it like every other provider (prod crash: AttributeError)."""
+    provider = internal_provider(doc_id="doc_new")
+    assert provider.last_error is None
+
+
 async def test_indexed_document_is_found_by_internal_provider(fake_mongo):
     assert await _index(fake_mongo) is True
 

tests/test_graph_memory.py

@@ -0,0 +1,49 @@
+"""Regression tests for GraphMemory report-context formatting.
+
+Incident: job_829f9c99 - when the novelty agent's LLM call failed under the
+Groq quota storm it stored novelty_score=None, and get_novelty_context raised
+"unsupported format string passed to NoneType.__format__", crashing the report
+agent and stopping the whole pipeline (no PDF, no email).
+"""
+
+from unittest.mock import MagicMock
+
+import pytest
+
+from memory.graph_memory import GraphMemory
+
+
+@pytest.fixture
+def graph_memory(mock_neo4j, mock_embedder):
+    return GraphMemory(neo4j=mock_neo4j, embedder=mock_embedder)
+
+
+def test_novelty_context_with_none_score(graph_memory):
+    """novelty_score=None (LLM failure) must render as 0.0, not crash."""
+    graph_memory.neo4j.run.return_value = [
+        {"novelty_score": None, "verdict": "UNKNOWN",
+         "problem_statement": "ps", "rationale": "r", "similar_works": []}
+    ]
+    ctx = graph_memory.get_novelty_context("doc_x")
+    assert "Novelty score: 0.0/10 (UNKNOWN)" in ctx
+
+
+def test_novelty_context_with_real_score(graph_memory):
+    graph_memory.neo4j.run.return_value = [
+        {"novelty_score": 7.3, "verdict": "NOVEL",
+         "problem_statement": "ps", "rationale": "r", "similar_works": []}
+    ]
+    ctx = graph_memory.get_novelty_context("doc_x")
+    assert "Novelty score: 7.3/10 (NOVEL)" in ctx
+
+
+def test_counterfactuality_context_with_none_scores(graph_memory):
+    """None overall_score/cf_score (LLM failure) must not crash the report."""
+    graph_memory.neo4j.run.side_effect = [
+        [{"overall_score": None, "common_type": "factual", "concerns": [],
+          "strengths": [], "summary": "s"}],
+        [{"text": "c1", "cf_type": "unsupported", "cf_score": None}],
+    ]
+    ctx = graph_memory.get_counterfactuality_context("doc_x")
+    assert "Overall counterfactuality score: 0.00/1.0" in ctx
+    assert "(score=0.00) c1" in ctx