~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
"""Explicit Source separators reuse the existing citation parser (Refs #7527)."""
import pytest
from local_deep_research.utilities.search_utilities import (
_MAX_RANGE_ENDPOINT_DIGITS,
_MAX_RANGE_SPAN,
_MAX_TOTAL_EXPANSION,
extract_cited_indices_and_urls,
filter_cited_links,
)
@pytest.fixture
def links():
return [
{"index": str(i), "url": f"https://source{i}.example/Article"}
for i in range(1, 6)
]
@pytest.mark.parametrize("word", ["Source", "Sources", "source", "SOURCES"])
@pytest.mark.parametrize(
"separator", [":", "#", ": ", "# ", " :", " #", " : ", " # "]
)
@pytest.mark.parametrize(
"numbers", ["2, 3", "2 and 3", "2 & 3", "2-3", "2–3", "2—3"]
)
def test_punctuation_extraction(word, separator, numbers):
assert extract_cited_indices_and_urls(f"{word}{separator}{numbers}.") == (
{2, 3},
set(),
False,
)
@pytest.mark.parametrize("phrase", ["Sources: 2, 3", "Source #2 and 3"])
@pytest.mark.parametrize("mode", ["strict", "fallback"])
def test_punctuation_mixed_filter(links, phrase, mode):
prose = f"Claim [1]. Further evidence {phrase}."
kept = filter_cited_links(links, prose, mode=mode)
assert kept == links[:3]
assert all(actual is original for actual, original in zip(kept, links))
@pytest.mark.parametrize("phrase", ["Sources: 2, 3", "Source #2"])
def test_disabled_preserves_original_object(links, phrase):
assert filter_cited_links(links, phrase, mode="disabled") is links
@pytest.mark.parametrize(
"prose", ["Source 2", "Sources 2 and 3", "[2, 3]", "[2–3]"]
)
def test_control_existing_signals(links, prose):
expected = {2} if prose == "Source 2" else {2, 3}
assert extract_cited_indices_and_urls(prose) == (expected, set(), False)
for mode in ("strict", "fallback"):
assert filter_cited_links(links, prose, mode=mode) == [
link for link in links if int(link["index"]) in expected
]
@pytest.mark.parametrize(
"phrase",
[
"Source2",
"Sources2",
"resource2",
"resource:2",
"Source/2",
"Source=2",
"Source;2",
"Source-2",
"Source::2",
"Source##2",
"Source :#2",
],
)
def test_control_requires_explicit_delimiter(links, phrase):
prose = f"[1] alongside {phrase}."
assert extract_cited_indices_and_urls(prose) == ({1}, set(), False)
for mode in ("strict", "fallback"):
assert filter_cited_links(links, prose, mode=mode) == links[:1]
@pytest.mark.parametrize("fence", ["```", "~~~"])
def test_control_fences_and_url_identity(links, fence):
prose = (
f"{fence}\nSources: 2, 3; Source #4\n{fence}\n"
"[1] https://SOURCE2.example/Article"
)
assert extract_cited_indices_and_urls(prose) == (
{1},
{links[1]["url"]},
False,
)
for mode in ("strict", "fallback"):
assert filter_cited_links(links, prose, mode=mode) == links[:2]
@pytest.mark.parametrize("separator", [" ", ": ", " #"])
@pytest.mark.parametrize("budget", ["range", "digits", "aggregate"])
def test_shared_limits_fail_safe(links, separator, budget):
if budget == "range":
numbers = f"2-{2 + _MAX_RANGE_SPAN + 1}"
elif budget == "digits":
numbers = "9" * (_MAX_RANGE_ENDPOINT_DIGITS + 1)
else:
# Repeated legal ranges still consume the shared expansion budget.
numbers = ", ".join(
[f"2-{2 + _MAX_RANGE_SPAN}"]
* (_MAX_TOTAL_EXPANSION // (_MAX_RANGE_SPAN + 1) + 1)
)
prose = f"[1] Sources{separator}{numbers}"
indices, urls, indeterminate = extract_cited_indices_and_urls(prose)
assert 1 in indices
assert urls == set()
assert indeterminate
for mode in ("strict", "fallback"):
assert filter_cited_links(links, prose, mode=mode) is links