~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
"""Measure reference/text growth and assert the complete exported documents."""
from concurrent.futures import ThreadPoolExecutor
from dataclasses import replace
import io
import zipfile
import pytest
from local_deep_research.exporters.quarto_exporter import QuartoExporter
from .corpus import Report, perform, prepare, today, verify
@pytest.mark.parametrize("mode", ["document", "zip"])
@pytest.mark.parametrize("curve", ["references", "text"])
def test_export_growth_preserves_complete_document_and_bibliography(
quarto_workload, entry_probe, mode, curve
):
for size in (
quarto_workload.settings["size"],
quarto_workload.settings["size"] * 2,
):
report = Report(size, 4) if curve == "references" else Report(4, size)
bound = 3 if mode == "zip" else 2
def setup():
entry_probe.update(parses=0, parsed_characters=0)
return prepare(report)
def check(state, output):
verify(state, output, report, mode)
assert entry_probe["parses"] <= bound
assert entry_probe["parsed_characters"] <= bound * len(
report.content
)
return dict(
references=report.references,
paragraphs=report.paragraphs,
**entry_probe,
)
quarto_workload.measure(
f"{curve}_{mode}",
size,
setup,
lambda state: perform(state[0], report, mode),
check,
units=len(report.content),
unit="input_character",
parameters=dict(
references=report.references,
paragraphs=report.paragraphs,
mode=mode,
instrumentation=True,
corpus_sha256=report.sha256,
input_characters=len(report.content),
input_bytes=len(report.content.encode("utf-8")),
),
)
@pytest.mark.parametrize("mode", ["document", "zip"])
def test_reused_exporter_observes_changed_reports(mode):
exporter = QuartoExporter()
outputs = []
for report in [Report(2, 1), Report(3, 2), Report(2, 1)]:
state = exporter, today(), report.content
output = perform(exporter, report, mode)
verify(state, output, report, mode)
outputs.append((state, output, report))
for state, output, report in outputs:
verify(state, output, report, mode)
def test_shared_exporter_keeps_concurrent_report_results_separate():
exporter = QuartoExporter()
reports = [Report(2, 1), Report(3, 2)]
started = today()
with ThreadPoolExecutor(max_workers=2) as pool:
outputs = list(pool.map(lambda r: perform(exporter, r, "zip"), reports))
for report, output in zip(reports, outputs, strict=True):
verify((exporter, started, report.content), output, report, "zip")
@pytest.mark.parametrize("fault", ["document", "bibliography", "member"])
def test_archive_oracle_rejects_incomplete_or_changed_results(fault):
report = Report(2, 1)
state = prepare(report)
output = perform(state[0], report, "zip")
verify(state, output, report, "zip")
contents = io.BytesIO()
with zipfile.ZipFile(io.BytesIO(output.content)) as original:
with zipfile.ZipFile(contents, "w", zipfile.ZIP_DEFLATED) as changed:
for name in original.namelist():
if name == "references.bib" and fault == "member":
continue
value = original.read(name).decode("utf-8")
if fault == "document" and name.endswith(".qmd"):
value = value.replace("[@ref1]", "[@ref9999]")
if fault == "bibliography" and name.endswith(".bib"):
value = value.replace("/source/1", "/source/9999")
changed.writestr(name, value)
corrupted = replace(output, content=contents.getvalue())
with pytest.raises(AssertionError, match="Quarto .* mismatch"):
verify(state, corrupted, report, "zip")