arborist/tests/test_query.py
russell@unturf.com 1ef0ac753a
qa: post-LLM faithfulness verifier sets STRICT/HYBRID/VISUAL audit_mode
Adds aborist/qa/verify.py — three-strategy verifier (explicit quotes,
bullet/sentence spans, multi-word proper nouns) that lexically checks
every claim against retrieved context under norm-v1 + lowercase. The
strategy that fires is recorded as verifier_method for diagnostics.
Entity strategy gates classification via an entity_policy
(strict/hybrid/proximity/drop) so a single proper-noun match no longer
overclaims STRICT — proximity (default) requires a cluster of 3+
verified entities within 300 chars.

Wires into ask() and query(). System prompts now require verbatim
quoted spans for every factual claim, restated via a user-turn
grounding_reminder one message before sources arrive (recent
user-turn instructions outweigh decayed system-turn rules under long
context in Hermes).

providence_cache gains 5 columns (audit_mode, n_quotes, n_verified,
unverified_quotes, verifier_method) with CHECK constraints. A
connect-time _migrate_audit_mode() ALTERs legacy DBs idempotently.
Cache hits return the persisted audit_mode rather than asserting
STRICT unconditionally.

CLI:
  emergent     list VISUAL/HYBRID records; --aggregate ranks
               unverified quotes by frequency (corpus-growth signal).
  reclassify   re-run the verifier against live providence records;
               cold-source records skipped; --dry-run reports
               transitions without writing; each change writes one
               'providence_reclassify' audit event.
2026-04-28 15:45:08 -04:00

174 lines
5.1 KiB
Python

"""Multi-source corpus Q&A: search → context → cache → Hermes.
Stub client only — no network. Validates:
- FTS5 finds the right docs across a small corpus
- context_root is deterministic (sorted source roots, then Merkle)
- cache hit on identical question returns same record without calling client
- different question -> different cache_key
"""
from __future__ import annotations
import json
from typing import Iterator
from aborist.document import Document
from aborist.ingest import ingest_source
from aborist.qa import query
from aborist.qa.client import StubClient
from aborist.source import Source
from aborist.store import connect
class FakeSource(Source):
source_type = "test"
def __init__(self, docs: list[Document]):
self.docs = docs
def iter_documents(self) -> Iterator[Document]:
yield from self.docs
def _doc(uri: str, content: str) -> Document:
return Document(uri=uri, content=content, source_type="test", title=uri.rsplit("/", 1)[-1])
# Three docs with distinguishable content so FTS5 can pick winners.
DOCS = [
_doc(
"test://anarchism",
"Anarchism is a political philosophy that opposes the state. " * 12
+ "Mutual aid is central to anarchist theory. " * 8,
),
_doc(
"test://capitalism",
"Capitalism is an economic system based on private ownership. " * 12
+ "Market exchange and capital accumulation drive growth. " * 8,
),
_doc(
"test://anarcho-capitalism",
"Anarcho-capitalism combines anarchism's opposition to the state with capitalism's private property. " * 12
+ "Murray Rothbard developed many of its core ideas. " * 8,
),
]
def test_query_picks_relevant_docs_and_writes_record(tmp_path):
main_db = tmp_path / "corpus.db"
qa_db = tmp_path / "qa.db"
conn = connect(main_db)
try:
ingest_source(conn, FakeSource(DOCS))
finally:
conn.close()
# Verbatim quote from DOCS[2] → audit_mode=STRICT.
client = StubClient(
answer=(
'Per the source: '
'"Anarcho-capitalism combines anarchism\'s opposition to the state '
'with capitalism\'s private property"'
)
)
result = query(
question="What is anarcho-capitalism?",
qa_db=qa_db,
chat_client=client,
model_id="test-model",
single_db=main_db,
top_k=3,
)
assert result["status"] == "cache_miss_then_written"
assert result["audit_mode"] == "STRICT"
assert "Anarcho-capitalism" in result["answer_text"]
assert len(result["sources"]) >= 1
assert any("anarcho-capitalism" in s["document_uri"] for s in result["sources"])
# context_root is deterministic.
sorted_roots = sorted(s["document_root"] for s in result["sources"])
if len(sorted_roots) == 1:
assert result["context_root"] == sorted_roots[0]
# Repeat call with same question → cache hit, no LLM call.
n_calls_before = len(client.calls)
result2 = query(
question="What is anarcho-capitalism?",
qa_db=qa_db,
chat_client=client,
model_id="test-model",
single_db=main_db,
top_k=3,
)
assert result2["status"] == "cache_hit"
assert result2["cache_key"] == result["cache_key"]
assert len(client.calls) == n_calls_before # no new call
def test_query_different_question_different_cache(tmp_path):
main_db = tmp_path / "corpus.db"
qa_db = tmp_path / "qa.db"
conn = connect(main_db)
try:
ingest_source(conn, FakeSource(DOCS))
finally:
conn.close()
client = StubClient(answer="answer text")
r1 = query(
question="What is anarchism?",
qa_db=qa_db,
chat_client=client,
model_id="m",
single_db=main_db,
)
r2 = query(
question="What is capitalism?",
qa_db=qa_db,
chat_client=client,
model_id="m",
single_db=main_db,
)
assert r1["cache_key"] != r2["cache_key"]
assert len(client.calls) == 2 # both missed cache, both called client
def test_query_no_sources_when_empty_corpus(tmp_path):
main_db = tmp_path / "empty.db"
qa_db = tmp_path / "qa.db"
connect(main_db).close() # creates schema, no docs
result = query(
question="What is anything?",
qa_db=qa_db,
chat_client=StubClient(),
model_id="m",
single_db=main_db,
)
assert result["status"] == "no_sources"
def test_query_persists_audit_event(tmp_path):
main_db = tmp_path / "corpus.db"
qa_db = tmp_path / "qa.db"
conn = connect(main_db)
try:
ingest_source(conn, FakeSource(DOCS))
finally:
conn.close()
query(
question="What is anarchism?",
qa_db=qa_db,
chat_client=StubClient(answer="X"),
model_id="m",
single_db=main_db,
)
# qa.db should now have one providence_query event in its audit chain.
qc = connect(qa_db)
try:
events = qc.execute(
"SELECT event_type FROM audit_events ORDER BY seq"
).fetchall()
finally:
qc.close()
types = [e["event_type"] for e in events]
assert "providence_query" in types