fix: fail closed on generic entity matches

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

Author: Pigbibi

docs/free_source_setup.md

@@ -9,6 +9,15 @@
 - `data/live/political_watchlist.csv`: initial watchlist for the live publish path.
 - `data/live/source_items.csv`: latest raw text pulled by the scheduled RSS pipeline.
 - `data/live/source_events.csv`: events extracted deterministically from `source_items.csv`.
+
+`source_events.csv` keeps the original columns and adds the following compatible
+evidence columns: `entity_match_type`, `match_evidence`, and
+`relationship_type`. Values are `issuer`, `direct_beneficiary`,
+`industry_context`, or `unverified`. Only the first two are company-level
+evidence; industry vocabulary is retained for auditability but is not scored as
+an issuer event. Rows without a verifiable entity match are not emitted by the
+live extractor. Legacy imported rows default to `unverified` and therefore fail
+closed in downstream company-event scoring.
 - `data/live/political_events.csv`: stable Advisor input, refreshed by RSS/source pipeline or maintained after manual review.
 - `data/live/source_tracker.csv`: merged watchlist and event tracker.
 

docs/free_source_setup.zh-CN.md

@@ -9,6 +9,13 @@
 - `data/live/political_watchlist.csv`：真实发布链路的初始观察池。
 - `data/live/source_items.csv`：定时 RSS pipeline 最近一次拉取的公开原始文本。
 - `data/live/source_events.csv`：从 `source_items.csv` 确定性抽取出的事件。
+
+`source_events.csv` 保留原有字段，并增加兼容字段：
+`entity_match_type`、`match_evidence`、`relationship_type`。关系值为
+`issuer`、`direct_beneficiary`、`industry_context` 或 `unverified`。
+只有前两类可作为公司级证据；行业通用词会保留用于审计，但不会进入公司事件评分。
+live 抽取器不会输出无法验证实体关联的行；旧格式导入行默认为
+`unverified`，下游公司事件评分会 fail-closed。
 - `data/live/political_events.csv`：Advisor 稳定读取的真实事件输入，由 RSS/source pipeline 刷新，也可以人工核验后维护。
 - `data/live/source_tracker.csv`：观察池与事件合并后的 tracker。
 

src/political_event_tracking_research/event_study.py

@@ -18,6 +18,9 @@ class Event:
     confidence: str
     source_url: str
     notes: str
+    entity_match_type: str = "unverified"
+    match_evidence: str = ""
+    relationship_type: str = "unverified"
 
 
 @dataclass(frozen=True)
@@ -74,6 +77,9 @@ def load_events(path: str | Path) -> list[Event]:
                 confidence=row.get("confidence", ""),
                 source_url=row.get("source_url", ""),
                 notes=row.get("notes", ""),
+                entity_match_type=row.get("entity_match_type", "unverified") or "unverified",
+                match_evidence=row.get("match_evidence", ""),
+                relationship_type=row.get("relationship_type", "unverified") or "unverified",
             )
         )
     return events

src/political_event_tracking_research/official_event_import.py

@@ -45,6 +45,9 @@ class OfficialRecord:
     direction: str
     source_url: str
     summary: str
+    entity_match_type: str = "unverified"
+    match_evidence: str = ""
+    relationship_type: str = "unverified"
 
 
 def slug(value: str) -> str:
@@ -76,6 +79,9 @@ def load_official_records(path: str | Path) -> list[OfficialRecord]:
                 direction=row.get("direction", ""),
                 source_url=row["source_url"],
                 summary=row.get("summary", ""),
+                entity_match_type=row.get("entity_match_type", "unverified") or "unverified",
+                match_evidence=row.get("match_evidence", ""),
+                relationship_type=row.get("relationship_type", "unverified") or "unverified",
             )
         )
     return records
@@ -87,6 +93,10 @@ def validate_record(record: OfficialRecord) -> None:
         raise ValueError("record_id is required")
     if not record.symbol.strip():
         raise ValueError(f"{record.record_id}: symbol is required")
+    if record.entity_match_type not in {"issuer", "direct_beneficiary", "industry_context", "unverified"}:
+        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.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:
@@ -125,6 +135,9 @@ def normalize_records(records: list[OfficialRecord]) -> list[dict[str, str]]:
                 "confidence": confidence_for_source(record),
                 "source_url": record.source_url,
                 "notes": record.summary,
+                "entity_match_type": record.entity_match_type,
+                "match_evidence": record.match_evidence,
+                "relationship_type": record.relationship_type,
             }
         )
     rows.sort(key=lambda row: (row["event_date"], row["symbol"], row["event_id"]))
@@ -136,7 +149,19 @@ def import_official_events(input_path: str | Path, output_path: str | Path) -> l
     rows = normalize_records(records)
     write_csv_rows(
         output_path,
-        ["event_id", "event_date", "symbol", "event_type", "direction", "confidence", "source_url", "notes"],
+        [
+            "event_id",
+            "event_date",
+            "symbol",
+            "event_type",
+            "direction",
+            "confidence",
+            "source_url",
+            "notes",
+            "entity_match_type",
+            "match_evidence",
+            "relationship_type",
+        ],
         rows,
     )
     return rows

src/political_event_tracking_research/source_mention_extract.py

@@ -10,6 +10,9 @@
 from .official_event_import import OfficialRecord, normalize_records
 
 
+GENERIC_ENTITY_NAMES = frozenset({"strategy"})
+
+
 @dataclass(frozen=True)
 class RawSourceItem:
     item_id: str
@@ -24,6 +27,7 @@ class RawSourceItem:
 class MentionAlias:
     symbol: str
     aliases: tuple[str, ...]
+    name: str = ""
 
 
 def load_raw_items(path: str | Path) -> list[RawSourceItem]:
@@ -51,10 +55,13 @@ def load_aliases(path: str | Path) -> list[MentionAlias]:
     records: list[MentionAlias] = []
     for row in read_csv_rows(path):
         symbol = row["symbol"].upper()
+        name = row.get("name", "").strip()
         aliases = split_aliases(row.get("aliases", ""))
+        if name and name.casefold() not in {alias.casefold() for alias in aliases}:
+            aliases = (name, *aliases)
         if symbol not in aliases:
             aliases = (symbol, *aliases)
-        records.append(MentionAlias(symbol=symbol, aliases=aliases))
+        records.append(MentionAlias(symbol=symbol, aliases=aliases, name=name))
     return records
 
 
@@ -88,6 +95,36 @@ def match_symbols(text: str, aliases: list[MentionAlias]) -> list[str]:
     return sorted(dict.fromkeys(matches))
 
 
+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()
+
+
+def match_evidence(text: str, alias_record: MentionAlias) -> tuple[str, str] | None:
+    normalized_text = normalize_match_text(text)
+    for alias in alias_record.aliases:
+        if not alias_pattern(alias).search(normalized_text):
+            continue
+        if _is_generic_alias(alias_record, alias):
+            return "industry_context", normalize_match_text(alias)
+        relationship = "issuer"
+        if re.search(
+            rf"(?:awarded to|contract with|funding for|benefit(?:s|ed)? from)\s+(?:the\s+)?{re.escape(normalize_match_text(alias))}",
+            normalized_text,
+            re.IGNORECASE,
+        ):
+            relationship = "direct_beneficiary"
+        return relationship, normalize_match_text(alias)
+    return None
+
+
 def infer_event_type(item: RawSourceItem) -> str:
     source_type = item.source_type
     text = item.text.lower()
@@ -133,24 +170,42 @@ def extract_source_records(raw_items_path: str | Path, aliases_path: str | Path,
     aliases = load_aliases(aliases_path)
     records: list[OfficialRecord] = []
     for item in raw_items:
-        symbols = match_symbols(item.text, aliases)
-        for symbol in symbols:
+        for alias_record in aliases:
+            evidence = match_evidence(item.text, alias_record)
+            if evidence is None:
+                continue
+            entity_match_type, matched_text = evidence
             records.append(
                 OfficialRecord(
-                    record_id=f"{item.item_id}-{symbol.lower()}",
+                    record_id=f"{item.item_id}-{alias_record.symbol.lower()}",
                     record_date=item_date(item),
-                    symbol=symbol,
+                    symbol=alias_record.symbol,
                     source_type=item.source_type,
                     event_type=infer_event_type(item),
                     direction=infer_direction(item.text),
                     source_url=item.source_url,
                     summary=f"{item.author}: {item.text}".strip(": "),
+                    entity_match_type=entity_match_type,
+                    match_evidence=matched_text,
+                    relationship_type=entity_match_type,
                 )
             )
     rows = normalize_records(records)
     write_csv_rows(
         output_path,
-        ["event_id", "event_date", "symbol", "event_type", "direction", "confidence", "source_url", "notes"],
+        [
+            "event_id",
+            "event_date",
+            "symbol",
+            "event_type",
+            "direction",
+            "confidence",
+            "source_url",
+            "notes",
+            "entity_match_type",
+            "match_evidence",
+            "relationship_type",
+        ],
         rows,
     )
     return rows

src/political_event_tracking_research/tracker.py

@@ -67,6 +67,8 @@ 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"}:
+            continue
         score += EVENT_WEIGHTS.get(event.event_type, 0)
         score += CONFIDENCE_WEIGHTS.get(event.confidence, 0)
     return score

tests/test_source_mention_extract.py

@@ -2,7 +2,11 @@
 
 from pathlib import Path
 
-from political_event_tracking_research.source_mention_extract import extract_source_records, infer_direction, match_symbols
+from political_event_tracking_research.source_mention_extract import (
+    extract_source_records,
+    infer_direction,
+    match_symbols,
+)
 from political_event_tracking_research.source_mention_extract import MentionAlias
 
 
@@ -51,3 +55,91 @@ def test_extract_source_records_outputs_confidence_by_source_type(tmp_path: Path
     assert by_symbol["EVT3"]["confidence"] == "low"
     assert by_symbol["EVT4"]["event_type"] == "procurement"
     assert output.exists()
+
+
+def test_generic_aliases_are_context_only_and_keep_evidence(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"
+        "nist-1,2026-04-01T00:00:00Z,government_policy,https://www.nist.gov/example,NIST,"
+        'NIST guidance discusses nuclear reactor safety and tokenization.\n',
+        encoding="utf-8",
+    )
+    aliases = tmp_path / "aliases.csv"
+    aliases.write_text(
+        "symbol,name,aliases\n"
+        "OKLO,Oklo,Oklo|OKLO|nuclear reactor\n"
+        "COIN,Coinbase,Coinbase|COIN|tokenization\n",
+        encoding="utf-8",
+    )
+
+    rows = extract_source_records(raw_items, aliases, tmp_path / "events.csv")
+
+    by_symbol = {row["symbol"]: row for row in rows}
+    assert by_symbol["OKLO"]["entity_match_type"] == "industry_context"
+    assert by_symbol["COIN"]["entity_match_type"] == "industry_context"
+    assert by_symbol["OKLO"]["match_evidence"] == "nuclear reactor"
+    assert by_symbol["COIN"]["relationship_type"] == "industry_context"
+
+
+def test_explicit_issuer_match_is_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-1,2026-04-01T00:00:00Z,issuer_release,https://example.com/release,Issuer,"
+        'Coinbase announced a new product.\n',
+        encoding="utf-8",
+    )
+    aliases = tmp_path / "aliases.csv"
+    aliases.write_text("symbol,name,aliases\nCOIN,Coinbase,COIN|tokenization\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]["relationship_type"] == "issuer"
+    assert rows[0]["match_evidence"] == "Coinbase"
+
+
+def test_explicit_beneficiary_relation_is_preserved(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"
+        "award-1,2026-04-01T00:00:00Z,government_procurement,https://www.govinfo.gov/example,Agency,"
+        'Funding for Coinbase was announced.\n',
+        encoding="utf-8",
+    )
+    aliases = tmp_path / "aliases.csv"
+    aliases.write_text("symbol,name,aliases\nCOIN,Coinbase,COIN\n", encoding="utf-8")
+
+    rows = extract_source_records(raw_items, aliases, tmp_path / "events.csv")
+
+    assert rows[0]["entity_match_type"] == "direct_beneficiary"
+    assert rows[0]["relationship_type"] == "direct_beneficiary"
+
+
+def test_known_ticker_and_industry_collisions_never_become_issuer_events(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"
+        "mstr,2026-04-01T00:00:00Z,government_policy,https://www.nist.gov/mstr,NIST,Strategy guidance\n"
+        "oklo,2026-04-01T00:00:00Z,government_policy,https://www.nist.gov/oklo,NIST,nuclear reactor safety\n"
+        "panw,2026-04-01T00:00:00Z,government_policy,https://www.nist.gov/panw,NIST,generic cybersecurity guidance\n"
+        "coin,2026-04-01T00:00:00Z,government_policy,https://www.federalreserve.gov/coin,Fed,tokenization policy\n"
+        "intc,2026-04-01T00:00:00Z,government_policy,https://www.commerce.gov/intc,Commerce,CHIPS Act funding\n",
+        encoding="utf-8",
+    )
+    aliases = tmp_path / "aliases.csv"
+    aliases.write_text(
+        "symbol,name,aliases\n"
+        "MSTR,Strategy,Strategy|MSTR\n"
+        "OKLO,Oklo,Oklo|OKLO|nuclear reactor\n"
+        "PANW,Palo Alto Networks,Palo Alto Networks|PANW|cybersecurity\n"
+        "COIN,Coinbase,Coinbase|COIN|tokenization\n"
+        "INTC,Intel,Intel|INTC|CHIPS Act\n",
+        encoding="utf-8",
+    )
+
+    rows = extract_source_records(raw_items, aliases, tmp_path / "events.csv")
+
+    assert {row["symbol"] for row in rows} == {"MSTR", "OKLO", "PANW", "COIN", "INTC"}
+    assert {row["entity_match_type"] for row in rows} == {"industry_context"}

tests/test_tracker.py

@@ -38,3 +38,12 @@ def test_tracker_prioritizes_disclosure_plus_mention() -> None:
     assert rows[0]["latest_event_date"] == "2026-01-02"
     assert rows[1]["symbol"] == "BBB"
     assert rows[1]["trigger_status"] == "watchlist"
+
+
+def test_tracker_does_not_score_industry_context_as_company_evidence() -> 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", "industry_context")]
+
+    rows = build_tracker_rows([item], events)
+
+    assert rows[0]["priority_score"] == 0