fix: tighten entity evidence remediation

Co-Authored-By: Codex <noreply@openai.com>

Author: Pigbibi

src/political_event_tracking_research/official_event_import.py

@@ -97,6 +97,8 @@ def validate_record(record: OfficialRecord) -> None:
         raise ValueError(f"{record.record_id}: unsupported entity_match_type {record.entity_match_type!r}")
     if record.relationship_type not in {"issuer", "direct_beneficiary", "industry_context", "unverified"}:
         raise ValueError(f"{record.record_id}: unsupported relationship_type {record.relationship_type!r}")
+    if record.entity_match_type != record.relationship_type:
+        raise ValueError(f"{record.record_id}: entity_match_type and relationship_type must agree")
     if record.event_type not in ALLOWED_EVENT_TYPES:
         raise ValueError(f"{record.record_id}: unsupported event_type {record.event_type!r}")
     if record.source_type in GOVERNMENT_SOURCE_TYPES:

src/political_event_tracking_research/source_mention_extract.py

@@ -10,7 +10,23 @@
 from .official_event_import import OfficialRecord, normalize_records
 
 
-GENERIC_ENTITY_NAMES = frozenset({"strategy"})
+GENERIC_ALIAS_DENYLIST = frozenset(
+    {
+        "advanced nuclear",
+        "bitcoin treasury",
+        "chips act",
+        "crypto assets",
+        "cybersecurity",
+        "digital assets",
+        "energy-related infrastructure",
+        "foundry",
+        "nuclear energy",
+        "nuclear reactor",
+        "semiconductor manufacturing",
+        "strategy",
+        "tokenization",
+    }
+)
 
 
 @dataclass(frozen=True)
@@ -99,12 +115,7 @@ def _is_generic_alias(alias_record: MentionAlias, alias: str) -> bool:
     normalized = normalize_match_text(alias).strip()
     if normalized.upper() == alias_record.symbol.upper():
         return False
-    if normalized.casefold() == alias_record.name.casefold():
-        return normalized.casefold() in GENERIC_ENTITY_NAMES
-    if alias_record.name:
-        return True
-    # Lower-case phrases are intentionally treated as industry vocabulary, not entities.
-    return normalized == normalized.lower() or len(normalized.split()) >= 3 and not normalized.istitle()
+    return normalized.casefold() in GENERIC_ALIAS_DENYLIST
 
 
 def match_evidence(text: str, alias_record: MentionAlias) -> tuple[str, str] | None:

src/political_event_tracking_research/tracker.py

@@ -67,7 +67,7 @@ def latest_event(events: list[Event]) -> Event | None:
 def event_score(events: list[Event]) -> int:
     score = 0
     for event in events:
-        if event.entity_match_type not in {"issuer", "direct_beneficiary"}:
+        if event.relationship_type not in {"issuer", "direct_beneficiary"}:
             continue
         score += EVENT_WEIGHTS.get(event.event_type, 0)
         score += CONFIDENCE_WEIGHTS.get(event.confidence, 0)

tests/test_official_event_import.py

@@ -76,3 +76,21 @@ def test_community_lead_records_are_not_in_stable_source_set() -> None:
 
     with pytest.raises(ValueError, match="unsupported source_type"):
         normalize_records([record])
+
+
+def test_import_rejects_mismatched_entity_and_relationship_types() -> None:
+    record = OfficialRecord(
+        record_id="mismatched-relation",
+        record_date="2026-01-10",
+        symbol="BAD",
+        source_type="issuer_release",
+        event_type="public_mention",
+        direction="neutral",
+        source_url="https://example.com/release",
+        summary="Mismatched evidence fields.",
+        entity_match_type="issuer",
+        relationship_type="industry_context",
+    )
+
+    with pytest.raises(ValueError, match="must agree"):
+        normalize_records([record])

tests/test_source_mention_extract.py

@@ -100,6 +100,23 @@ def test_explicit_issuer_match_is_company_level(tmp_path: Path) -> None:
     assert rows[0]["match_evidence"] == "Coinbase"
 
 
+def test_proper_noun_alternate_alias_remains_company_level(tmp_path: Path) -> None:
+    raw_items = tmp_path / "source_items.csv"
+    raw_items.write_text(
+        "item_id,published_at,source_type,source_url,author,text\n"
+        "issuer-2,2026-04-01T00:00:00Z,issuer_release,https://example.com/release,Issuer,"
+        'Former Brand announced a new product.\n',
+        encoding="utf-8",
+    )
+    aliases = tmp_path / "aliases.csv"
+    aliases.write_text("symbol,name,aliases\nTEST,Current Company,Current Company|Former Brand\n", encoding="utf-8")
+
+    rows = extract_source_records(raw_items, aliases, tmp_path / "events.csv")
+
+    assert rows[0]["entity_match_type"] == "issuer"
+    assert rows[0]["match_evidence"] == "Former Brand"
+
+
 def test_explicit_beneficiary_relation_is_preserved(tmp_path: Path) -> None:
     raw_items = tmp_path / "source_items.csv"
     raw_items.write_text(

tests/test_tracker.py

@@ -47,3 +47,12 @@ def test_tracker_does_not_score_industry_context_as_company_evidence() -> None:
     rows = build_tracker_rows([item], events)
 
     assert rows[0]["priority_score"] == 0
+
+
+def test_tracker_uses_relationship_type_as_scoring_fact() -> None:
+    item = WatchlistItem("AAA", "Alpha", "watchlist", "watchlist", "seed", "https://example.com")
+    events = [Event("e1", date(2026, 1, 1), "AAA", "public_mention", "bullish", "high", "https://example.com", "", "industry_context", "cybersecurity", "issuer")]
+
+    rows = build_tracker_rows([item], events)
+
+    assert rows[0]["priority_score"] > 0