Files
paper-atlas/tests/test_mm_publications2026.py
T
codex-bot 9864632774
Metadata integrity checks / metadata (push) Failing after 5m46s
Recover MM program evidence through reviewed title formatting
2026-11-18 04:14:41 +08:00

274 lines
20 KiB
Python

from contextlib import redirect_stdout
import io
import os
import uuid
import gzip
import hashlib
import json
from pathlib import Path
import re
import tempfile
import unittest
from paper_atlas.collectors import mm_publications2026 as c
class MMPublications2026Tests(unittest.TestCase):
@classmethod
def setUpClass(cls):
fixtures=Path(__file__).parent/'fixtures'
cls.fixture=json.loads(gzip.decompress((fixtures/'mm2026-publications.json.gz').read_bytes()))
cls.program_fixture=json.loads(gzip.decompress((fixtures/'mm2026-source.json.gz').read_bytes()))
cls.format_gold=json.loads((fixtures/'mm2026-title-formatting-review.json').read_text())
overrides={g['doi']:g for g in [*cls.format_gold['gold'],cls.format_gold['held_gold']]}
cls.gold=[overrides.get(g['doi'],g) for g in cls.fixture['independent_gold']]
cls.temp=tempfile.TemporaryDirectory();cls.root=Path(cls.temp.name)
cls.materialize(cls.root/'registry',cls.fixture['files'])
cls.materialize(cls.root/'program',cls.program_fixture['files'])
cls.out=cls.root/'out'
cls.report=c.collect(cls.out,registry_snapshots=cls.root/'registry',program_snapshots=cls.root/'program',parent_snapshot=cls.root/'registry/parent.json')
cls.rows=[json.loads(line) for f in cls.report['files'] for line in Path(f['file']).read_bytes().splitlines()]
cls.by_doi={r['external_id']:r for r in cls.rows}
@classmethod
def tearDownClass(cls):
cls.temp.cleanup()
@staticmethod
def materialize(root,files):
for name,body in files.items():
path=root/name;path.parent.mkdir(parents=True,exist_ok=True);path.write_bytes(body.encode('utf-8'))
def test_all1918_publications_against_independent_registry_program_gold(self):
self.assertEqual(len(self.rows),1918);self.assertEqual(len(self.by_doi),1918)
self.assertEqual(set(self.by_doi),{g['doi'] for g in self.gold})
for g in self.gold:
r=self.by_doi[g['doi']];m=r['metadata'];p=m['program_fields']
self.assertEqual((r['title'],r['authors'],r['track']),(g['title'],g['authors'],g['track']))
self.assertEqual(m['program_match_kind'],g['kind'])
self.assertEqual(m['registry_source_evidence']['json_pointer'],'/message/items/'+str(g['source_position']))
self.assertEqual(m['registry_source_evidence']['file'],'evidence/registry/'+g['source_page'])
self.assertEqual(m['registry_source_evidence']['canonical_object_json_sha256'],g['registry_object_sha256'])
self.assertEqual(p['paper_id'] if p else None,g['program_id'])
self.assertEqual([x['source_pointer'] for x in p['contexts']] if p else [],g['program_context_pointers'])
def test_abstract_source_is_separate_and_verbatim_not_fabricated_registry_abstract(self):
count=0
for g in self.gold:
r=self.by_doi[g['doi']];m=r['metadata']
self.assertFalse(m['registry_abstract_available']);self.assertFalse(m['registry_fields'].get('abstract'))
if g['abstract_sha256'] is not None:
count+=1;self.assertEqual(hashlib.sha256(r['abstract'].encode()).hexdigest(),g['abstract_sha256'])
self.assertEqual(r['abstract'],m['program_fields']['abstract'])
self.assertEqual(m['abstract_source'],'official_program_reviewed_title_formatting_correspondence' if g['kind'].startswith('unique_program_format_') else 'official_program_exact_metadata_correspondence')
else:
self.assertIsNone(r['abstract']);self.assertEqual(m['abstract_source'],'not_available_from_reviewed_sources')
self.assertEqual(count,1743);self.assertEqual(self.report['output_missing_abstract_count'],175)
def test_all_program_tracks_and_4326_contexts_survive_in_matches_or_unmatched(self):
programs={r['metadata']['program_fields']['paper_id']:r['metadata']['program_fields'] for r in self.rows if r['metadata']['program_fields']}
for p in self.report['unmatched_program']:
self.assertNotIn(p['paper_id'],programs);programs[p['paper_id']]=p
self.assertEqual(len(programs),2143);self.assertEqual(self.report['unmatched_program_count'],391)
gold=self.program_fixture['independent_inventory']['inventory']
self.assertEqual(set(programs),{g['paper_id'] for g in gold})
for g in gold:
p=programs[g['paper_id']]
self.assertEqual((p['title'],p['authors'],p['track']),(g['title'],g['authors'],g['track_literal']))
self.assertEqual([x['source_pointer'] for x in p['contexts']],g['locations'])
self.assertEqual(sum(len(p['contexts']) for p in programs.values()),4326)
self.assertEqual(len({p['track'] for p in programs.values()}),28)
def test_no_person_alias_original_submission_decision_or_research_type_inference(self):
for r in self.rows:
doi=r['external_id'];m=r['metadata']
self.assertEqual(r['identifiers'],{'doi':doi})
self.assertEqual(r['submission_identifiers'],[{'scheme':'publication_doi','value':doi}])
self.assertEqual((r['source'],r['venue'],r['year']),(c.SOURCE,'MM',2026))
self.assertEqual(r['status'],'accepted');self.assertIn('not an original signed decision',r['raw_decision'])
self.assertFalse(m['original_decision_signature_verified']);self.assertFalse(m['research_type_classified'])
self.assertFalse(m['scientific_work_identity_approved'])
self.assertTrue({'decided_at','decision_date','source_decision_id','paper_type'}.isdisjoint(r))
self.assertIn({'namespace':c.SOURCE+':crossref:type','value':'proceedings-article'},r['tags'])
self.assertTrue({x['namespace'] for x in r['tags']} <= {c.SOURCE+':crossref:type',c.SOURCE+':mm_program:track'})
for track in ('interactive_art','industry_talk','demo_and_video'):
self.assertTrue(any(r['track']==track for r in self.rows))
variants=[r for r in self.rows if r['track']=='proceedings_unclassified']
self.assertEqual(len(variants),166)
self.assertTrue(all(r['metadata']['program_fields'] is None and r['abstract'] is None for r in variants))
def test_original_bibliography_evidence_kept_without_duplicating_reference_payload(self):
authors_by_page={}
for name,body in self.fixture['files'].items():
actual=self.out/'evidence/registry'/('parent.json' if name=='parent.json' else Path(name).name)
self.assertEqual(actual.read_bytes(),body.encode())
if re.fullmatch(r'page-[0-9]+\.json',Path(name).name):
authors_by_page[str(actual.relative_to(self.out))]=[row['author'] for row in json.loads(body)['message']['items']]
for name,body in self.program_fixture['files'].items():
self.assertEqual((self.out/'evidence/program'/Path(name).name).read_bytes(),body.encode())
for r in self.rows:
m=r['metadata'];self.assertNotIn('reference',m['registry_fields'])
self.assertEqual(m['registry_fields']['title'],[r['title']])
self.assertEqual(m['registry_fields']['author'],authors_by_page[m['registry_source_evidence']['file']][int(m['registry_source_evidence']['json_pointer'].rsplit('/',1)[-1])])
for f in self.report['files']:
raw=Path(f['file']).read_bytes();self.assertEqual(hashlib.sha256(raw).hexdigest(),f['file_sha256'])
self.assertEqual(len(raw.splitlines()),f['count']);self.assertEqual(f['status'],'partial')
def test_eight_single_name_fields_preserved_without_splitting(self):
n=0
for g in self.gold:
r=self.by_doi[g['doi']];flags=r['metadata']['review_flags']
actual=[x['author_position'] for x in flags if x['kind']=='single_registered_name_field_not_split']
self.assertEqual(actual,g['single_name_positions']);n+=len(actual)
for i in actual:
self.assertEqual(r['authors'][i-1],r['metadata']['registry_fields']['author'][i-1]['family'].strip())
self.assertEqual(n,8)
def test_structured_title_markup_retained_and_flagged_without_flattening(self):
count=0
for g in self.gold:
if re.search(r'<[^>]+>|&[A-Za-z#0-9]+;',g['title']):
count+=1;r=self.by_doi[g['doi']]
self.assertEqual(r['title'],g['title'])
self.assertIn({'kind':'literal_title_markup_or_entity_retained_requires_review'},r['metadata']['review_flags'])
self.assertGreater(count,0);self.assertEqual(count,self.report['registered_title_markup_flags'])
def test_formatting_gold_preserves_raw_fields_and_copies_exact_program_evidence(self):
self.assertEqual(len(self.format_gold['gold']),26)
for gold in self.format_gold['gold']:
row=self.by_doi[gold['doi']];m=row['metadata'];p=m['program_fields']
self.assertEqual((row['title'],row['authors']),(gold['title'],gold['authors']))
self.assertEqual((p['paper_id'],row['track']),(gold['program_id'],gold['track']))
self.assertEqual(hashlib.sha256(row['abstract'].encode()).hexdigest(),gold['abstract_sha256'])
comparison=m['program_title_formatting_comparison']
self.assertTrue(comparison['raw_titles_preserved']);self.assertFalse(comparison['work_identity_approved'])
self.assertEqual(set(comparison['program_comparison_texts']),{p['paper_id']})
self.assertFalse(m['scientific_work_identity_approved'])
held=self.by_doi[self.format_gold['author_disagreement_hold']['doi']]
self.assertEqual(held['metadata']['program_title_candidates'],['mfp7801'])
self.assertEqual(held['metadata']['program_exact_candidates'],[])
self.assertEqual(held['metadata']['program_match_kind'],'program_format_author_version_requires_review')
self.assertIsNone(held['abstract']);self.assertEqual(held['track'],'proceedings_unclassified')
def test_reviewed_formatting_keeps_superscript_subscript_and_decodes_entities_once(self):
self.assertEqual(c.formatting_text('<i>A</i> H<sup>2</sup> B<sub>1</sub> &amp; C'),'A H2 B1 & C')
self.assertEqual(c.formatting_text('A &amp;amp; B'),'A &amp; B')
def test_unknown_attributes_unbalanced_or_encoded_markup_not_discarded(self):
for title in ['A <model>B</model>','A <i class="x">B</i>','A <i>B','A </i>B',
'A <sup/>B','A <!--hidden-->B','<!DOCTYPE html>A','A &lt;i&gt;B&lt;/i&gt;',
'A <sup','A <i>B</sup>','<![CDATA[A]]>']:
with self.subTest(title=title):self.assertIsNone(c.formatting_text(title))
def test_ambiguous_program_and_author_multiset_differences_remain_held(self):
def observation(pid,authors):
return {'paper_id':pid,'title':'M3 Model','authors':authors,'track':'Main Track'}
rows=[{'title':'M<sup>3</sup> Model','authors':['A','B']}]
kinds,_=c.match_program(rows,[observation('p1',['A','B']),observation('p2',['B','A'])])
self.assertEqual(kinds,{'multiple_program_ids_same_title_authors_hold':1})
self.assertNotIn('program',rows[0])
rows=[{'title':'M<sup>3</sup> Model','authors':['A','A','B']}]
c.match_program(rows,[observation('p1',['A','B'])])
self.assertEqual(rows[0]['kind'],'program_format_author_version_requires_review')
self.assertNotIn('program',rows[0])
def test_literal_title_author_disagreement_does_not_use_formatting_bypass(self):
rows=[{'title':'M<sup>3</sup> Model','authors':['A','B']}]
observations=[{'paper_id':'literal','title':'M<sup>3</sup> Model','authors':['Other'],'track':'Main Track'},
{'paper_id':'formatted','title':'M3 Model','authors':['A','B'],'track':'Main Track'}]
c.match_program(rows,observations)
self.assertEqual(rows[0]['kind'],'program_author_version_requires_review')
self.assertEqual(rows[0]['program_title_candidates'],['literal']);self.assertNotIn('program',rows[0])
def test_corrupt_registry_parent_or_program_keeps_failure_not_complete_output(self):
for kind in ('registry_manifest','registry_page','parent','program'):
with tempfile.TemporaryDirectory() as tmp:
root=Path(tmp);self.materialize(root/'registry',self.fixture['files']);self.materialize(root/'program',self.program_fixture['files'])
path={'registry_manifest':root/'registry'/c.SNAPSHOT/'manifest.json','registry_page':root/'registry'/c.SNAPSHOT/'page-001.json','parent':root/'registry/parent.json','program':root/'program'/c.program.SNAPSHOT/'program-data.json'}[kind]
path.write_bytes(path.read_bytes()+b' ')
with self.subTest(kind=kind),self.assertRaises(ValueError):
c.collect(root/'out',registry_snapshots=root/'registry',program_snapshots=root/'program',parent_snapshot=root/'registry/parent.json')
self.assertTrue((root/'out/failed.json').exists());self.assertFalse((root/'out/collection-report.json').exists())
def test_symlink_source_rejected_and_no_previous_output_overwrite(self):
with tempfile.TemporaryDirectory() as tmp:
root=Path(tmp);self.materialize(root/'registry',self.fixture['files'])
path=root/'registry/parent.json';original=root/'registry/actual.json';path.rename(original);path.symlink_to(original)
with self.assertRaisesRegex(ValueError,'Unsafe'):
c.collect(root/'out',registry_snapshots=root/'registry',program_snapshots=self.root/'program',parent_snapshot=path)
with self.assertRaisesRegex(ValueError,'Output exists'):
c.collect(self.out,registry_snapshots=self.root/'registry',program_snapshots=self.root/'program',parent_snapshot=self.root/'registry/parent.json')
def test_complete_report_retains_source_partitions_not_import_or_final_dedup_claim(self):
self.assertEqual(self.report['program_match_counts'],{'unique_program_title_fullauthor_agreement_not_identity_approval':1726,'unique_program_format_title_fullauthor_agreement_not_identity_approval':26,'program_author_version_requires_review':122,'program_format_author_version_requires_review':1,'no_exact_program_title_requires_review':43})
self.assertEqual(self.report['registry_missing_abstract_count'],1918)
self.assertEqual(self.report['abstracts_from_exact_program_correspondence'],1743)
self.assertEqual(self.report['abstracts_from_literal_program_correspondence'],1717)
self.assertEqual(self.report['abstracts_from_reviewed_title_formatting'],26)
self.assertEqual(self.report['unmatched_program_count'],391)
self.assertEqual(sum(f['count'] for f in self.report['files']),1918)
self.assertEqual(self.report['approved_for_import'],0);self.assertFalse(self.report['database_written'])
self.assertFalse(self.report['pdfs_downloaded']);self.assertEqual(self.report['network_requests'],0)
self.assertIn('unknown',self.report['public_negative_availability'])
def assert_native_ingestion(self, pg_service=None):
# An isolated synthetic database exercises the writer contract, not a
# production identity approval. All source rows/track partitions run.
from paper_atlas.db.store import Store
from paper_atlas.db.backend import open_store
from paper_atlas.ingestion.records import ingest_file
from paper_atlas.postgres.client import connect
from paper_atlas.postgres.migration import migrate
from paper_atlas.postgres.schema import quote
with tempfile.TemporaryDirectory() as tmp:
root=Path(tmp);db=root/'seed.sqlite';seed=self.rows[0]
store=Store(db)
paper,record=store.record('arxiv','synthetic-mm-anchor','https://example.org/arxiv',
seed['title'],'Existing canonical abstract',seed['authors'],seed['identifiers'])
store.decision(paper,record,'ICLR',2025,'main','synthetic-old-forum',
'https://example.org/decision','rejected','Reject')
before=tuple(store.conn.execute('SELECT * FROM papers WHERE id=?',(paper,)).fetchone())
store.conn.commit();store.conn.close()
schema='pa_mm_pub_'+uuid.uuid4().hex[:12] if pg_service else None
try:
with redirect_stdout(io.StringIO()):
if pg_service:migrate(db,pg_service,root/'migration.json',schema=schema,min_free_gib=0)
for f in self.report['files']:
ingest_file(None if pg_service else db,f['file'],pg_service=pg_service,pg_schema=schema)
store=open_store(None if pg_service else db,pg_service=pg_service,pg_schema=schema)
try:
conn=store.conn
self.assertEqual(tuple(conn.execute('SELECT * FROM papers WHERE id=?',(paper,)).fetchone()),before)
self.assertEqual(conn.execute('SELECT status FROM decisions WHERE submission_id=1').fetchone()[0],'rejected')
actual={r[0]:r for r in conn.execute("SELECT sr.external_id,sr.paper_id,sr.title,sr.abstract,sr.authors_json,sr.metadata_json,sr.id FROM source_records sr JOIN sources s ON s.id=sr.source_id WHERE s.name=?",(c.SOURCE,))}
self.assertEqual(set(actual),set(self.by_doi))
self.assertEqual(actual[seed['external_id']][1],paper)
for doi,row in self.by_doi.items():
got=actual[doi];self.assertEqual((got[2],got[3],json.loads(got[4])),(row['title'],row['abstract'],row['authors']))
meta=json.loads(got[5]);self.assertEqual(meta.pop('record_identifiers'),{'doi':doi,c.SOURCE:doi});self.assertEqual(meta,row['metadata'])
tags=conn.execute("SELECT t.namespace,t.value,count(*) FROM paper_tags pt JOIN tags t ON t.id=pt.tag_id GROUP BY t.namespace,t.value").fetchall()
self.assertEqual(sum(r[2] for r in tags if r[0]==c.SOURCE+':crossref:type'),1918)
self.assertEqual(sum(r[2] for r in tags if r[0]==c.SOURCE+':mm_program:track'),1752)
self.assertEqual({r[0] for r in tags},{c.SOURCE+':crossref:type',c.SOURCE+':mm_program:track'})
decisions=conn.execute("SELECT d.status,d.decision_date,count(*) FROM decisions d JOIN submissions s ON s.id=d.submission_id JOIN source_records r ON r.id=s.source_record_id JOIN sources src ON src.id=r.source_id WHERE src.name=? GROUP BY d.status,d.decision_date",(c.SOURCE,)).fetchall()
self.assertEqual([tuple(r) for r in decisions],[('accepted',None,1918)])
self.assertEqual(conn.execute("SELECT count(*) FROM submission_identifiers WHERE scheme='publication_doi'").fetchone()[0],1918)
coverage=conn.execute("SELECT es.status,sum(es.imported_count) FROM edition_sources es JOIN sources s ON s.id=es.source_id WHERE s.name=? GROUP BY es.status",(c.SOURCE,)).fetchall()
self.assertEqual([tuple(r) for r in coverage],[('partial',1918)])
finally:store.conn.close()
finally:
if pg_service:
with connect(pg_service) as pg:
for name, in pg.execute('SELECT nspname FROM pg_namespace WHERE starts_with(nspname,%s)',(schema,)).fetchall():
pg.execute('DROP SCHEMA '+quote(name)+' CASCADE')
def test_all1918_native_sqlite_source_tags_and_prior_rejection_survive(self):
self.assert_native_ingestion()
@unittest.skipUnless(os.environ.get('PAPER_ATLAS_TEST_PG_SERVICE'),'Requires approved remote PostgreSQL test service')
def test_all1918_native_postgres_source_tags_and_prior_rejection_survive(self):
self.assert_native_ingestion(os.environ['PAPER_ATLAS_TEST_PG_SERVICE'])