fix: services research lists junk no-content pages as cited sources (#1669)

This commit is contained in:
Afonso Coutinho
2026-06-03 06:22:58 +01:00
committed by GitHub
parent eae8797e08
commit 7f80d33210
2 changed files with 92 additions and 3 deletions

View File

@@ -14,6 +14,8 @@ import time
from pathlib import Path from pathlib import Path
from typing import Optional, Dict from typing import Optional, Dict
from src.research_utils import is_low_quality
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
RESEARCH_DATA_DIR = Path("data/deep_research") RESEARCH_DATA_DIR = Path("data/deep_research")
@@ -179,13 +181,14 @@ class ResearchHandler:
@staticmethod @staticmethod
def _extract_sources(findings: list) -> list: def _extract_sources(findings: list) -> list:
"""Extract deduplicated [{url, title}] from findings.""" """Extract deduplicated [{url, title}] from findings, filtering low-quality ones."""
seen = set() seen = set()
sources = [] sources = []
for f in findings: for f in findings:
url = f.get("url", "") url = f.get("url", "")
title = f.get("title", "") or url title = f.get("title", "") or url
if url and url not in seen: summary = f.get("summary", "") or f.get("evidence", "")
if url and url not in seen and not is_low_quality(summary):
seen.add(url) seen.add(url)
sources.append({"url": url, "title": title}) sources.append({"url": url, "title": title})
return sources return sources
@@ -346,7 +349,8 @@ class ResearchHandler:
for f in findings: for f in findings:
url = f.get("url", "") url = f.get("url", "")
title = f.get("title", "") or url title = f.get("title", "") or url
if url and url not in seen_urls: summary = f.get("summary", "") or f.get("evidence", "")
if url and url not in seen_urls and not is_low_quality(summary):
seen_urls.add(url) seen_urls.add(url)
source_lines.append(f"- [{title}]({url})") source_lines.append(f"- [{title}]({url})")
if source_lines: if source_lines:

View File

@@ -0,0 +1,85 @@
"""services/research _extract_sources must gate low-quality findings.
The src/research_handler.py copy filters findings whose summary is junk
boilerplate (via research_utils.is_low_quality) before listing them as
cited sources. The services/research copy diverged and had no gate, so
"the page does not contain relevant information" URLs showed up as
sources, and a junk finding seen first suppressed the good title for the
same URL. services/research/service.py imports this handler, so it is the
live path.
"""
import importlib.util
import sys
import types
import pytest
@pytest.fixture
def handler_cls(monkeypatch):
"""Load services.research.research_handler from its file path so the
heavy services/__init__.py (httpx etc.) is never imported."""
pkg = types.ModuleType("services")
pkg.__path__ = []
sub = types.ModuleType("services.research")
sub.__path__ = []
monkeypatch.setitem(sys.modules, "services", pkg)
monkeypatch.setitem(sys.modules, "services.research", sub)
name = "services.research.research_handler"
monkeypatch.delitem(sys.modules, name, raising=False)
spec = importlib.util.spec_from_file_location(
name, "services/research/research_handler.py"
)
mod = importlib.util.module_from_spec(spec)
monkeypatch.setitem(sys.modules, name, mod)
spec.loader.exec_module(mod)
return mod.ResearchHandler
JUNK = "The page does not contain relevant information"
def test_low_quality_summary_is_not_a_source(handler_cls):
out = handler_cls._extract_sources([{"url": "http://a", "title": "T", "summary": JUNK}])
assert out == []
def test_good_summary_is_kept(handler_cls):
out = handler_cls._extract_sources(
[{"url": "http://a", "title": "T", "summary": "Detailed statistics about the topic"}]
)
assert out == [{"url": "http://a", "title": "T"}]
def test_junk_first_no_longer_suppresses_the_good_finding(handler_cls):
out = handler_cls._extract_sources(
[
{"url": "http://a", "title": "Bad", "summary": JUNK},
{"url": "http://a", "title": "Good", "summary": "Real data about the topic"},
]
)
assert out == [{"url": "http://a", "title": "Good"}]
def test_evidence_is_checked_when_summary_missing(handler_cls):
out = handler_cls._extract_sources(
[{"url": "http://a", "title": "T", "evidence": "Concrete evidence text"}]
)
assert out == [{"url": "http://a", "title": "T"}]
def test_report_sources_section_gates_junk(handler_cls):
h = object.__new__(handler_cls)
report = h._format_research_report(
"q",
"full report",
{},
1.0,
findings=[
{"url": "http://junk", "title": "Junk", "summary": JUNK},
{"url": "http://good", "title": "Good", "summary": "Useful content here"},
],
)
assert "http://good" in report
assert "- [Junk](http://junk)" not in report