homelab/stack
1
0
Fork 0
Code Issues 59 Pull Requests 1 Actions Packages Projects 1 Releases Wiki Activity

fix(llm): code-cited sources ranked per code; early return on no codes; log resolved replica path (refs P48)

Browse Source
This commit is contained in:
kert
2026-09-08 22:47:08 -04:00
parent 6424742027
commit 086d35445c
2 changed files with 26 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/llm/evidence.py
View File

@@ -98,10 +98,11 @@ def valuation_evidence(question: str, cfg: LlmConfig) -> ValuationEvidence | Non
det = detect_codes(question) det = detect_codes(question)
if not det.codes: if not det.codes:
return None return None
path = _replica_path(cfg)
try: try:
con = duckdb.connect(_replica_path(cfg), read_only=True) con = duckdb.connect(path, read_only=True)
except Exception as e: # noqa: BLE001 — duckdb raises several types except Exception as e: # noqa: BLE001 — duckdb raises several types
log.warning("valuation evidence skipped (%s): %s", cfg.duckdb_replica, e) log.warning("valuation evidence skipped (%s): %s", path, e)
return None return None
try: try:
rows, unpriced = valuation(con, list(det.codes), years=cfg.valuation_years) rows, unpriced = valuation(con, list(det.codes), years=cfg.valuation_years)
@@ -114,12 +115,20 @@ def valuation_evidence(question: str, cfg: LlmConfig) -> ValuationEvidence | Non
_CITED_SQL = text( _CITED_SQL = text(
"SELECT e.document, e.cmetadata FROM langchain_pg_embedding e " "SELECT document, cmetadata FROM ( "
"JOIN langchain_pg_collection c ON c.uuid = e.collection_id " "SELECT e.document, e.cmetadata, "
"WHERE c.name = :collection " "row_number() OVER ( "
"AND string_to_array(COALESCE(e.cmetadata->>'codes', ''), ' ') && CAST(:codes AS text[]) " "PARTITION BY w.code "
"ORDER BY e.cmetadata->>'date' DESC NULLS LAST, e.cmetadata->>'item_key', e.cmetadata->>'p_id' " "ORDER BY e.cmetadata->>'date' DESC NULLS LAST, e.cmetadata->>'item_key', e.cmetadata->>'p_id' "
"LIMIT :lim" ") AS rn "
"FROM langchain_pg_embedding e "
"JOIN langchain_pg_collection c ON c.uuid = e.collection_id "
"JOIN unnest(CAST(:codes AS text[])) AS w(code) "
"ON w.code = ANY(string_to_array(COALESCE(e.cmetadata->>'codes', ''), ' ')) "
"WHERE c.name = :collection "
") t "
"WHERE rn <= :window "
"ORDER BY cmetadata->>'date' DESC NULLS LAST, cmetadata->>'item_key', cmetadata->>'p_id'"
) )
@@ -130,6 +139,8 @@ def code_cited_sources(
*codes* — at most *per_code* per code, newest rule first, deduped on *codes* — at most *per_code* per code, newest rule first, deduped on
(item_key, p_id). Scores are 0.0: these are additive, not ranked.""" (item_key, p_id). Scores are 0.0: these are additive, not ranked."""
wanted = [c.upper() for c in codes] wanted = [c.upper() for c in codes]
if not wanted:
return []
try: try:
with engine.begin() as conn: with engine.begin() as conn:
rows = conn.execute( rows = conn.execute(
@@ -137,7 +148,7 @@ def code_cited_sources(
{ {
"collection": collection, "collection": collection,
"codes": wanted, "codes": wanted,
"lim": per_code * len(wanted) * 4, "window": per_code * 2,
}, },
).fetchall() ).fetchall()
except Exception as e: # noqa: BLE001 except Exception as e: # noqa: BLE001

7
tests/llm/test_evidence.py
View File

@@ -202,8 +202,10 @@ class TestCodeCitedSources:
out = code_cited_sources(engine, ["G0556", "G0557"], per_code=2) out = code_cited_sources(engine, ["G0556", "G0557"], per_code=2)
sql = str(conn.execute.call_args.args[0]) sql = str(conn.execute.call_args.args[0])
assert "string_to_array" in sql and "langchain_pg_collection" in sql assert "string_to_array" in sql and "langchain_pg_collection" in sql
assert "PARTITION BY" in sql
assert conn.execute.call_args.args[1]["collection"] == "rules" assert conn.execute.call_args.args[1]["collection"] == "rules"
assert conn.execute.call_args.args[1]["codes"] == ["G0556", "G0557"] assert conn.execute.call_args.args[1]["codes"] == ["G0556", "G0557"]
assert conn.execute.call_args.args[1]["window"] == 4
assert [s["snippet"] for s in out] == [ assert [s["snippet"] for s in out] == [
"para one G0556", "para one G0556",
"para two G0556", "para two G0556",
@@ -217,6 +219,11 @@ class TestCodeCitedSources:
assert code_cited_sources(engine, ["G0556"], per_code=2) == [] assert code_cited_sources(engine, ["G0556"], per_code=2) == []
assert "code-cited sources skipped" in caplog.text assert "code-cited sources skipped" in caplog.text
def test_no_codes_returns_empty_without_querying(self):
engine = MagicMock()
assert code_cited_sources(engine, [], per_code=2) == []
engine.begin.assert_not_called()
class TestMergeSources: class TestMergeSources:
def test_keeps_order_and_dedupes_by_label(self): def test_keeps_order_and_dedupes_by_label(self):