Preserve separate ICAPS registry publication versions and abstracts
Metadata integrity checks / metadata (push) Successful in 2m51s

This commit is contained in:
2026-11-17 00:30:14 +08:00
parent e19cb83cd6
commit 9093a8ce25
7 changed files with 253 additions and 0 deletions
+7
View File
@@ -60,6 +60,13 @@ jobs:
assert icaps_report['records'] == 90 and icaps_report['previously_published_presentations'] == 10
assert icaps_report['missing_abstract_count'] == 90 and not icaps_report['database_written']
assert icaps_report['files'][0]['track'] == 'main_tracks_unclassified'
registry_snapshots = root/'icaps2026-registry-snapshots'
registry_snapshots.mkdir()
(registry_snapshots/'crossref-filtered.json').write_bytes(gzip.decompress((src/'tests/fixtures/icaps2026-crossref.json.gz').read_bytes()))
subprocess.run([python,'-I','-B','-m','paper_atlas','collect','icaps2026-registry','--snapshots',str(registry_snapshots),'--out',str(root/'installed-icaps2026-registry')],cwd=outside,check=True)
registry_report = json.loads((root/'installed-icaps2026-registry/collection-report.json').read_text())
assert registry_report['records'] == 90 and registry_report['missing_abstract_count'] == 0
assert len(registry_report['excluded_volume35_dois']) == 1 and not registry_report['database_written']
rss_fixture = json.loads(gzip.decompress((src/'tests/fixtures/rss2026-source.json.gz').read_bytes()))
rss_snapshots = root/'rss2026-snapshots'
rss_snapshots.mkdir()
+2
View File
@@ -122,3 +122,5 @@ CI 在已授权的远程 runner 上运行测试、构建 wheel,并从源码目
RSS 2026 已独立适配官方论文集与录用表;使用方法和与 2025 demo 分类的区别见 [RSS 2026 来源说明](docs/rss2026-proceedings.md)。
ICAPS 2026 的[官方名单采集器](docs/icaps2026-program.md)区分 90 篇主会轨道组录用与 10 篇既有论文展示;逐篇 track、摘要和稳定标识尚未解决,输出均待审核,不能直接视作已完成入库。
ICAPS 2026 也提供独立的 `collect icaps2026-registry`,保留 Crossref 卷 36 的 DOI、作者与摘要版本;不覆盖节目表文字,不把卷 35 或版本差异自动并入。配置与边界见同一份[ICAPS 来源说明](docs/icaps2026-program.md)。
+16
View File
@@ -22,3 +22,19 @@ paper-atlas collect icaps2026-program \
生产导入之前仍需当前数据库候选核查、独立身份审核、来源及历史保护基线、导入后逐项验收。所有来源导入完成后还必须另做全库去重、覆盖、摘要和约束查询审计。正常使用不会自动执行这些后续写入。
测试及安装后 CLI 验证只在已授权的 `ces-control:host` runner 上运行。冻结 fixture 含完整 100 条官方条目与独立盘点结果,用于检查两组边界、作者顺序、缺页、重复、来源篡改和恢复行为。
## 独立注册元数据版本
```sh
paper-atlas collect icaps2026-registry \
--snapshots /data/paper-index/collected/icaps2026-crossref-research-v1 \
--out /data/paper-index/collected/icaps2026-registry-normalized-v1
```
该命令只读固定 `crossref-filtered.json`,与官方节目表输出分开。来源是 Crossref 的 ICAPS 期刊注册元数据(ISSN 2334-0843),查询出版日期为 2026 年的结果共 91 条,其中卷 36、期 1 有 90 条,另有卷 35 的一条必须排除。不能仅凭出版日期归属会议届次。来源 URL、抓取时间、原响应哈希和字段过滤快照均保留。
卷 36 的 90 条均带 DOI 和摘要,输出保持注册标题、作者顺序及摘要版本一致;保留原始 JATS 摘要,只移除已核验的单个段落包装并规范空白,遇到其他标记(尤其公式、上下标)拒绝无声扁平化。作者个人 ID、机构、引用与文件链接没有进入过滤快照。原始注册类型 `journal-article` 留在 metadata 中,不据此给论文添加研究类型标签。
这里的 accepted 表示注册源记录了正式发表,不代表采到原始签名录用决定;publication_doi 标识发表对应的投稿事件,不是原始投稿号。具体 track 保持待分类。与官方 90 篇名单的来源核查发现:88 个规范标题一致,其中 78 个完整作者名单一致;另有 2 个标题和 10 个作者文字差异待审核。两份输出不会自动建立作品同一性、复制人名别名或覆盖文字。它们也不应未经审核同时导入,避免重复作品或投稿。
此来源仍标记 partial,不包含公开拒稿。生产导入前需核对 DOI 归属、既有作品及跨会议历史;已知 DOI 也可能存在错误绑定,不能代替最终全库去重验收。
+1
View File
@@ -23,6 +23,7 @@ COMMANDS = {
('collect','colm'): ('paper_atlas.collectors.proceedings', {'source':'colm'}, 'COLM 官方论文集'),
('collect','neurips'): ('paper_atlas.collectors.proceedings', {'source':'neurips'}, 'NeurIPS 官方论文集'),
('collect','mlsys'): ('paper_atlas.collectors.proceedings', {'source':'mlsys'}, 'MLSys 官方历年标题与摘要'),
('collect','icaps2026-registry'): ('paper_atlas.collectors.icaps_registry2026', {}, '核验 ICAPS 2026 Crossref 注册元数据并保留独立版本'),
('collect','icaps2026-program'): ('paper_atlas.collectors.icaps2026', {}, '核验 ICAPS 2026 官方录用名单并分开既有论文展示'),
('collect','rss2026-proceedings'): ('paper_atlas.collectors.rss2026', {}, '核验 RSS 2026 官方论文集与独立录用表'),
('collect','cvf'): ('paper_atlas.collectors.vision', {'source':'cvf'}, 'CVPR / ICCV 官方开放目录'),
@@ -0,0 +1,131 @@
"""Frozen Crossref ICAPS volume 36 metadata, kept separate from program versions."""
import argparse
from datetime import datetime, date
import hashlib
from html import unescape
import json
from pathlib import Path
import re
from paper_atlas.io.html import clean
SOURCE = 'crossref_icaps_registry'
SNAPSHOT_SHA = 'c7480a09f71680d457a16cc1eb35a479742b7510eaad5f236ce54c96e8a29a82'
RAW_RESPONSE_SHA = 'ac5268333b1ecdfa2e385680e2e02b5f6639939c3193738d213e66bb53e42a54'
URL = 'https://api.crossref.org/journals/2334-0843/works?filter=from-pub-date%3A2026-01-01%2Cuntil-pub-date%3A2026-12-31&rows=200'
CONTAINER = 'Proceedings of the International Conference on Automated Planning and Scheduling'
TRACK = 'main_tracks_unclassified'
def digest(raw):
return hashlib.sha256(raw).hexdigest()
def abstract_text(raw):
if not isinstance(raw, str):
raise ValueError('Missing registry abstract')
match = re.fullmatch(r'<jats:p>(.*?)</jats:p>', raw, re.S)
if not match or re.search(r'<[^>]*>', match[1]):
raise ValueError('Unreviewed abstract markup; preserve rather than flatten math')
text = clean(unescape(match[1]))
if not text:
raise ValueError('Empty registry abstract')
return text
def registry_papers(snapshot):
observed = snapshot['at']
if (snapshot.get('status') != 'registry_snapshot_not_identity_approved_not_imported'
or snapshot.get('url') != URL or snapshot.get('raw_response_sha256') != RAW_RESPONSE_SHA
or snapshot.get('database_written') is not False
or datetime.fromisoformat(observed.replace('Z', '+00:00')).utcoffset() is None
or snapshot.get('total_results') != 91 or len(snapshot.get('records', [])) != 91):
raise ValueError('Unreviewed registry snapshot receipt')
papers = []; excluded = []; seen = set()
for row in snapshot['records']:
doi = row.get('DOI')
if not isinstance(doi, str) or doi in seen:
raise ValueError('Missing or repeated registry DOI')
seen.add(doi)
if row.get('volume') == '35':
excluded.append(doi)
continue
if (row.get('volume') != '36' or row.get('issue') != '1'
or row.get('container-title') != [CONTAINER] or row.get('type') != 'journal-article'
or not re.fullmatch(r'10\.1609/icaps\.v36i1\.[0-9]+', doi)
or row.get('URL') != 'https://doi.org/'+doi
or not isinstance(row.get('title'), list) or len(row['title']) != 1
or not isinstance(row['title'][0], str) or not clean(row['title'][0])):
raise ValueError('Unreviewed publication volume, DOI, type or title')
day, = row['published']['date-parts']
if len(day) != 3 or day[0] != 2026 or any(type(x) is not int for x in day):
raise ValueError('Unreviewed publication date')
publication_date = date(*day).isoformat()
authors = row.get('authors')
if (not isinstance(authors, list) or not authors
or any(not isinstance(a, dict) or set(a)-{'given','family','name','sequence'}
or not a.get('given') or not a.get('family') for a in authors)):
raise ValueError('Incomplete registry authors')
names = [clean(a['given']+' '+a['family']) for a in authors]
if (len(set(n.casefold() for n in names)) != len(names)
or any('@' in n or 'anonymous' in n.casefold() for n in names)
or [a.get('sequence') for a in authors] != ['first']+['additional']*(len(authors)-1)):
raise ValueError('Unreviewed registry author order or identity')
papers.append({'original': row, 'title': row['title'][0], 'authors': names,
'abstract': abstract_text(row.get('abstract')), 'publication_date': publication_date})
if len(papers) != 90 or len(excluded) != 1:
raise ValueError('Incomplete reviewed volume census')
return sorted(papers, key=lambda p: p['original']['DOI']), excluded
def collect(out, *, snapshots):
out = Path(out)
if out.exists():
raise ValueError('Output exists; inspect previous operation before retrying')
out.mkdir(parents=True)
try:
path = Path(snapshots)/'crossref-filtered.json'
if path.is_symlink() or path.stat().st_size > 4*1024**2:
raise ValueError('Unsafe or oversized registry snapshot')
raw = path.read_bytes()
if digest(raw) != SNAPSHOT_SHA:
raise ValueError('Reviewed registry snapshot hash changed')
snapshot = json.loads(raw); papers, excluded = registry_papers(snapshot)
directory = out/'evidence'; directory.mkdir(); rel = f'evidence/{digest(raw)}.json'; (out/rel).write_bytes(raw)
evidence = {'file': rel, 'sha256': digest(raw), 'url': URL}; rows = []
for paper in papers:
original = paper['original']; doi = original['DOI']
rows.append({'source': SOURCE, 'external_id': doi, 'url': original['URL'],
'title': paper['title'], 'authors': paper['authors'], 'abstract': paper['abstract'],
'identifiers': {'doi': doi}, 'submission_identifiers': [{'scheme': 'publication_doi', 'value': doi}],
'venue': 'ICAPS', 'year': 2026, 'track': TRACK, 'status': 'accepted',
'raw_decision': 'publication recorded in Crossref ICAPS volume 36 issue 1; not an original signed acceptance decision',
'evidence_url': URL, 'observed_at': snapshot['at'], 'tags': [],
'metadata': {'registry_fields': original, 'source_evidence': evidence,
'abstract_transform': 'Remove one reviewed jats:p wrapper, decode character entities and normalize whitespace; original JATS retained.',
'publication_date': paper['publication_date'], 'original_decision_signature_verified': False,
'track_assignment': 'Unresolved main/special-track membership; no per-paper track inferred from topic or publication order.',
'program_relationship': 'Separate registry version; not substituted into official-listing records. Title/author variants require review.',
'scope': 'Registered publication metadata, not original review decision or current live publisher page. Requires current work-identity and history review before import.'}})
path = out/'icaps-crossref-registry-2026-main-tracks-unclassified.jsonl'
payload = ''.join(json.dumps(r, ensure_ascii=False)+'\n' for r in rows).encode(); path.write_bytes(payload)
manifest = {'source': SOURCE, 'parser_version': 1, 'venue': 'ICAPS', 'year': 2026, 'track': TRACK,
'count': 90, 'file': str(path), 'file_sha256': digest(payload), 'list_url': URL, 'status': 'partial', 'missing_abstract_count': 0}
path.with_suffix('.manifest.json').write_text(json.dumps(manifest, indent=2)+'\n')
report = {'status': 'collected_pending_identity_review', 'records': 90, 'files': [manifest],
'excluded_volume35_dois': excluded, 'missing_abstract_count': 0, 'source_evidence': evidence,
'database_written': False, 'pdfs_downloaded': False, 'public_rejected_availability': 'unknown',
'scope': 'Crossref publication versions only; no automatic official-program linkage, person aliases, work merges or complete public-negative/final-dedup claim.'}
part = out/'collection-report.json.part'; part.write_text(json.dumps(report, ensure_ascii=False, indent=2)+'\n'); part.replace(out/'collection-report.json')
return report
except Exception as error:
(out/'failed.json').write_text(json.dumps({'status': 'failed_not_complete', 'error_type': type(error).__name__, 'database_written': False})+'\n')
raise
def main(argv=None, *, prog=None):
parser = argparse.ArgumentParser(prog=prog, description=__doc__)
parser.add_argument('--snapshots', required=True)
parser.add_argument('--out', required=True)
args = parser.parse_args(argv)
print(json.dumps(collect(args.out, snapshots=args.snapshots), ensure_ascii=False))
Binary file not shown.
+96
View File
@@ -0,0 +1,96 @@
import copy
import gzip
import json
from pathlib import Path
import tempfile
import unittest
from paper_atlas.collectors import icaps_registry2026 as c
class ICAPSRegistry2026Test(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.raw = gzip.decompress((Path(__file__).parent/'fixtures/icaps2026-crossref.json.gz').read_bytes())
cls.snapshot = json.loads(cls.raw)
def test_exact_edition_not_publication_year_alone(self):
papers, excluded = c.registry_papers(self.snapshot)
self.assertEqual(len(papers), 90); self.assertEqual(len(excluded), 1)
self.assertTrue(all(p['original']['volume'] == '36' for p in papers))
self.assertEqual(excluded, [r['DOI'] for r in self.snapshot['records'] if r['volume'] == '35'])
self.assertTrue(all(p['abstract'] and p['authors'] for p in papers))
def test_full_collection_keeps_registered_versions_and_publication_identity(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp); src = root/'src'; src.mkdir(); (src/'crossref-filtered.json').write_bytes(self.raw)
report = c.collect(root/'out', snapshots=src)
self.assertEqual((report['records'], report['missing_abstract_count']), (90, 0))
self.assertFalse(report['database_written']); self.assertFalse(report['pdfs_downloaded'])
self.assertEqual(report['public_rejected_availability'], 'unknown')
manifest, = report['files']; self.assertEqual(manifest['status'], 'partial')
rows = [json.loads(l) for l in Path(manifest['file']).read_text().splitlines()]
original = {r['DOI']: r for r in self.snapshot['records'] if r['volume'] == '36'}
self.assertEqual({r['external_id'] for r in rows}, set(original))
for r in rows:
doi = r['external_id']; o = original[doi]
self.assertEqual(r['metadata']['registry_fields'], o)
self.assertEqual(r['title'], o['title'][0])
self.assertEqual(r['authors'], [a['given']+' '+a['family'] for a in o['authors']])
self.assertEqual(r['identifiers'], {'doi': doi})
self.assertEqual(r['submission_identifiers'], [{'scheme': 'publication_doi', 'value': doi}])
self.assertFalse(r['tags']); self.assertFalse(r['metadata']['original_decision_signature_verified'])
self.assertEqual(r['track'], 'main_tracks_unclassified')
self.assertEqual(r['metadata']['publication_date'], '2026-06-08')
changed = next(r for r in rows if r['external_id'] == '10.1609/icaps.v36i1.42860')
self.assertEqual(changed['title'], 'Analyzing Planner Design Trade-Offs for MAPF Under ADG-Based Realistic Execution')
self.assertIn('Stephen F. Smith', changed['authors'])
with self.assertRaisesRegex(ValueError, 'Output exists'):
c.collect(root/'out', snapshots=src)
def test_math_markup_is_not_flattened_as_plain_numbers(self):
self.assertEqual(c.abstract_text('<jats:p>A &lt; B and C &amp; D.\n Next line.</jats:p>'), 'A < B and C & D. Next line.')
for raw in ['<jats:p>10<sup>6</sup></jats:p>', '<jats:p>x</jats:p><jats:p>y</jats:p>', '<jats:p> </jats:p>', None]:
with self.assertRaises(ValueError):
c.abstract_text(raw)
def test_invalid_or_repeated_doi_and_edition_rejected(self):
for kind in ('doi', 'repeat', 'volume', 'container', 'year'):
with self.subTest(kind=kind):
s = copy.deepcopy(self.snapshot); rows = [r for r in s['records'] if r['volume'] == '36']; r = rows[0]
if kind == 'doi': r['DOI'] = '10.9999/foreign'
if kind == 'repeat': rows[1]['DOI'] = r['DOI']
if kind == 'volume': r['volume'] = '37'
if kind == 'container': r['container-title'] = ['Other Conference']
if kind == 'year': r['published']['date-parts'] = [[2025, 6, 8]]
with self.assertRaises(ValueError):
c.registry_papers(s)
def test_missing_authors_and_changed_order_evidence_rejected(self):
for kind in ('missing', 'sequence', 'profile'):
s = copy.deepcopy(self.snapshot); r = next(r for r in s['records'] if r['volume'] == '36')
if kind == 'missing': r['authors'][0].pop('family')
if kind == 'sequence': r['authors'][0]['sequence'] = 'additional'
if kind == 'profile': r['authors'][0]['ORCID'] = 'unreviewed-profile'
with self.assertRaises(ValueError):
c.registry_papers(s)
def test_receipt_scope_and_timezone_rejected(self):
for key, value in [('at', '2026-11-17T00:00:00'), ('url', 'https://example.org/works'), ('raw_response_sha256', 'other')]:
s = copy.deepcopy(self.snapshot); s[key] = value
with self.assertRaises(ValueError):
c.registry_papers(s)
def test_tamper_failure_retained_and_success_not_published(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp); src = root/'src'; src.mkdir(); (src/'crossref-filtered.json').write_bytes(self.raw+b' ')
with self.assertRaisesRegex(ValueError, 'hash changed'):
c.collect(root/'out', snapshots=src)
self.assertTrue((root/'out/failed.json').exists())
self.assertFalse((root/'out/collection-report.json').exists())
def test_symlink_rejected(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp); src = root/'src'; src.mkdir(); (root/'data.json').write_bytes(self.raw); (src/'crossref-filtered.json').symlink_to(root/'data.json')
with self.assertRaisesRegex(ValueError, 'Unsafe'):
c.collect(root/'out', snapshots=src)