Merge main and preserve both ECCV and SDM validation paths
Metadata integrity checks / metadata (push) Successful in 3m5s

This commit is contained in:
2026-11-17 16:13:06 +08:00
7 changed files with 519 additions and 0 deletions
+17
View File
@@ -77,6 +77,23 @@ jobs:
eccv_literal_row = next(row for row in eccv_rows if row['external_id'] == '2026/3844') eccv_literal_row = next(row for row in eccv_rows if row['external_id'] == '2026/3844')
assert eccv_literal_row['title'] == eccv_literal['gold']['visible_title'] assert eccv_literal_row['title'] == eccv_literal['gold']['visible_title']
assert '&' in eccv_literal_row['title'] assert '&' in eccv_literal_row['title']
sdm_fixture = json.loads(gzip.decompress((src/'tests/fixtures/sdm2026-source.json.gz').read_bytes()))
sdm_snapshots = root/'sdm2026-snapshots'
for name, text in sdm_fixture['files'].items():
target = sdm_snapshots/name
target.parent.mkdir(parents=True, exist_ok=True)
target.write_bytes(text.encode('utf-8'))
subprocess.run([python,'-I','-B','-m','paper_atlas','collect','sdm2026-proceedings','--snapshots',str(sdm_snapshots),'--out',str(root/'installed-sdm2026')],cwd=outside,check=True)
sdm_report = json.loads((root/'installed-sdm2026/collection-report.json').read_text())
assert sdm_report['records'] == 53 and sdm_report['missing_abstract_count'] == 53
assert not sdm_report['database_written'] and sdm_report['network_requests'] == 0
assert sdm_report['program_observations']['count'] == 53 and sdm_report['program_observations']['abstract_pages'] == 52
assert sdm_report['program_registry_author_sequence_differences'] == 20
sdm_rows = [json.loads(line) for line in pathlib.Path(sdm_report['files'][0]['file']).read_text().splitlines()]
assert len(sdm_rows) == 53 and all(row['abstract'] is None for row in sdm_rows)
sdm_observations = [json.loads(line) for line in pathlib.Path(sdm_report['program_observations']['file']).read_text().splitlines()]
assert sum(row['cancelled_presentation'] for row in sdm_observations) == 1
assert all('status' not in row and 'identifiers' not in row for row in sdm_observations)
icaps_fixture = json.loads(gzip.decompress((src/'tests/fixtures/icaps2026-source.json.gz').read_bytes())) icaps_fixture = json.loads(gzip.decompress((src/'tests/fixtures/icaps2026-source.json.gz').read_bytes()))
icaps_snapshots = root/'icaps2026-snapshots' icaps_snapshots = root/'icaps2026-snapshots'
icaps_snapshots.mkdir() icaps_snapshots.mkdir()
+2
View File
@@ -130,3 +130,5 @@ ICAPS 2026 也提供独立的 `collect icaps2026-registry`,保留 Crossref 卷
`collect corl2026-program` 将官方 687 个论坛与固定公开档案按 ID 对照,保留 34 条标题版本差异及分字段来源。不会自动接受 arXiv 匹配或推断拒稿;详情见 [CoRL 2026 使用与恢复说明](docs/corl2026-program.md)。 `collect corl2026-program` 将官方 687 个论坛与固定公开档案按 ID 对照,保留 34 条标题版本差异及分字段来源。不会自动接受 arXiv 匹配或推断拒稿;详情见 [CoRL 2026 使用与恢复说明](docs/corl2026-program.md)。
ECCV 2026 的[暂定节目采集器](docs/eccv2026-program.md)合并重复节目 ID,并以不可变缓存断点采集公开元数据;保留摘要原文、来源关键词及未分类 track,始终待身份审核,不自动入库。 ECCV 2026 的[暂定节目采集器](docs/eccv2026-program.md)合并重复节目 ID,并以不可变缓存断点采集公开元数据;保留摘要原文、来源关键词及未分类 track,始终待身份审核,不自动入库。
`collect sdm2026-proceedings` 核验 53 个 Crossref 论文章节,并单独保留 53 项日程观察及 52 页摘要。作者差异、取消展示和摘要缺失分别保留,不自动拼接来源或绑定作品;见 [SDM 2026 使用与恢复说明](docs/sdm2026-proceedings.md)。
+44
View File
@@ -0,0 +1,44 @@
# SDM 2026:论文集与日程版本
这是长期维护、按需运行的离线来源适配器。它复用固定的公开元数据快照,不请求网络、不下载论文 PDF、不连接数据库,也不会把候选版本关联当成已通过的作品去重。
```sh
paper-atlas collect sdm2026-proceedings \
--snapshots /data/paper-index/collected \
--out /data/paper-index/collected/sdm2026-normalized-pass01
```
`--snapshots` 下需保留四个目录:`sdm2026-official-research-v1`、`sdm2026-crossref-research-v1`、`sdm2026-program-research-v1`、`sdm2026-public-abstract-research-v1`。适配器固定核验各自 manifest 和全部成功响应的字节数、哈希、URL 与状态。测试 fixture 保留原始响应字节及独立盘点结果。使用全新输出目录恢复,保留失败目录中的 `failed.json`;成功报告最后以原子替换发布。来源更新时另存版本并重新审阅,不能修改旧 manifest 后冒充旧快照。
## 两种输出
[论文集](https://epubs.siam.org/doi/book/10.1137/1.9781611979329)的 ISBN 为 `9781611979329`。[Crossref 精确 ISBN 查询](https://api.crossref.org/works?filter=isbn%3A9781611979329&rows=1000)的冻结响应有 55 项:53 个论文章节、1 本论文集、1 份前言。适配器只将 53 个章节放入 `report.files` 的待身份审阅 JSONL,排除论文集和前言;核对完整章节 DOI 集合、父 DOI、容器标题、ISBN、作者顺序证据及出版年份。
正式记录保留 Crossref 作者版本与原始字段,使用 `proceedings_unclassified`,不按主题或分会场推断 track。注册发表记录表达已收入论文集的事实,不冒充原始签名录用决定。印刷与在线出版日期分别保留,不能用它们替代会议举办日期或决定日期。两个标题含 `<sup>*</sup>`:显示标题保留脚注星号并标记待审阅,原始标题完整保留;去掉这一特定脚注的文字仅用于列出待审阅候选,不用于自动改名或绑定作品。
**53 个 Crossref 章节都没有摘要字段。**主记录的 `abstract` 因此为空,报告如实记为 53 个摘要缺失,不能因另一来源有同标题摘要而直接填入。
[官方在线日程](https://meetings.siam.org/program.cfm?CONFCODE=sdm26)有 9 个研究论文分会场、53 项日程条目和 52 个公开摘要页。`program-observations.jsonl` 单独保留日程版本、公开页标题、来源报告的作者姓名、摘要、会场引用、回执与哈希;它不是入库 manifest,也不在 `report.files` 中。页面中作者行的单位、国家及联系邮箱不规范化成人物字段,原始页面留在证据文件中。
52 个公开页中,32 个作者序列与 Crossref 完全一致,20 个不同,包括姓名拼写、顺序和作者缺失。例如 *Scaffold-Aware Generative Augmentation and Reranking for Enhanced Virtual Screening* 的日程页列出 Xin Wang、Qinwen Ge,Crossref 列出 Xin Wang、Yu Wang、Yunchao Liu、Jens Meiler、Tyler Derr。它们必须保留为独立来源版本,不能仅凭标题相同拼接摘要与作者,或生成姓名别名。候选 DOI 字段明确表示待审阅关系,不是逐篇身份批准;作者列表相同也不能代替全库作品身份核查。
*A Trajectory View of Statistical Policy Comparison* 的日程展示标为取消,且没有公开摘要链接;它仍出现在冻结的 Crossref 论文章节集合中。适配器保留这两项事实,**不把取消展示转换成论文撤回、拒稿或删除发表记录**。邀请报告、博士论坛和其他非研究论文活动也不计入这里的 53 个章节。
## 公开拒稿与访问限制
[官方投稿说明](https://www.siam.org/conferences-events/siam-conferences/sdm26/submissions/)链接 OpenReview group `SIAM.org/2026/Conference_on_Data_Mining`。匿名读取 group 配置成功;按实际 `rejected_venue_id` 查询 notes 返回 HTTP 403,随后停止,没有绕过或继续查询其他负面状态。因此公开拒稿、desk-rejected 和 withdrawn 决定覆盖仍未核实,不能报告为零条或从录用名单缺席推断拒稿。
来源冻结时,服务器访问 SIAM 主站的两份说明页及出版社目录也返回 403;回执如实保留。可读取的是公开在线日程及独立公开的 Crossref 注册元数据。Crossref 记录不是受限出版社页面的重建,也不被称作已经核验的出版社摘要版本。
## 来源恢复与后续核查
| 冻结目录 | manifest SHA-256 |
| --- | --- |
| sdm2026-official-research-v1 | a39a6ee6a12b69051ba2dad7c9cbc5b0fe215af85100c7aa2f5166d44c22fe61 |
| sdm2026-crossref-research-v1 | 51d337a153a1046e4a00ebc031707faf72e733577e4fc243a31672ee93277da5 |
| sdm2026-program-research-v1 | 689a2aaaa15be532957c4c88e03e1e2e87909334b6f529be16905d8b9cc1fb77 |
| sdm2026-public-abstract-research-v1 | 8e1c29611e145cbfa54ec4afaa2efe152557b4e04e339901af5cd33abefc29a5 |
服务器恢复包:`/data/paper-index/validation/sdm2026-sources-recovery-v1.tar.gz`,SHA-256 为 `d6d14c0e4bb2c26e16a4888e3b150e44f982d689351638b1b9a28644736c4387`;本机工作记录目录有同名备份。包内保留原始来源、独立盘点及逐页版本比较记录,含义是可复现来源证据,不是已完成导入或身份批准。
下次继续时,先核对来源版本和访问状态,再做当前数据库的 DOI 绑定、标题/作者近似候选与完整投稿历史核查,逐条解决作者及摘要版本问题。批准后另建入库队列、保护基线和导入后验收;所有来源结束后仍需全库去重、覆盖、摘要、约束和实际查询验收。测试、构建和安装验证只在已授权 `ces-control:host` 远程 runner 执行。
+1
View File
@@ -25,6 +25,7 @@ COMMANDS = {
('collect','mlsys'): ('paper_atlas.collectors.proceedings', {'source':'mlsys'}, 'MLSys 官方历年标题与摘要'), ('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-registry'): ('paper_atlas.collectors.icaps_registry2026', {}, '核验 ICAPS 2026 Crossref 注册元数据并保留独立版本'),
('collect','eccv2026-program'): ('paper_atlas.collectors.eccv2026', {}, '核验ECCV2026暂定节目并断点采集公开元数据'), ('collect','eccv2026-program'): ('paper_atlas.collectors.eccv2026', {}, '核验ECCV2026暂定节目并断点采集公开元数据'),
('collect','sdm2026-proceedings'): ('paper_atlas.collectors.sdm2026', {}, '核验 SDM 2026 注册论文并单独保留公开日程版本'),
('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','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','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'), ('collect','cvf'): ('paper_atlas.collectors.vision', {'source':'cvf'}, 'CVPR / ICCV 官方开放目录'),
+313
View File
@@ -0,0 +1,313 @@
"""Frozen SDM 2026 publication records and separate public program observations."""
import argparse
from collections import Counter
from datetime import date, datetime
import hashlib
from html.parser import HTMLParser
import json
from pathlib import Path
import re
from urllib.parse import parse_qs, urljoin, urlsplit
from paper_atlas.io.html import Document, clean
SOURCE = 'crossref_sdm_registry'
TRACK = 'proceedings_unclassified'
BOOK = '10.1137/1.9781611979329'
ISBN = '9781611979329'
TITLE = 'Proceedings of the 2026 SIAM International Conference on Data Mining (SDM)'
QUERY = 'https://api.crossref.org/works?filter=isbn%3A9781611979329&rows=1000'
PROGRAM = 'https://meetings.siam.org/program.cfm?CONFCODE=sdm26'
SNAPSHOTS = {
'sdm2026-official-research-v1': ('a39a6ee6a12b69051ba2dad7c9cbc5b0fe215af85100c7aa2f5166d44c22fe61', 'sources'),
'sdm2026-crossref-research-v1': ('51d337a153a1046e4a00ebc031707faf72e733577e4fc243a31672ee93277da5', 'sources'),
'sdm2026-program-research-v1': ('689a2aaaa15be532957c4c88e03e1e2e87909334b6f529be16905d8b9cc1fb77', 'sessions'),
'sdm2026-public-abstract-research-v1': ('8e1c29611e145cbfa54ec4afaa2efe152557b4e04e339901af5cd33abefc29a5', 'records'),
}
def digest(raw):
return hashlib.sha256(raw).hexdigest()
def one(items):
if len(items) != 1:
raise ValueError('Expected exactly one reviewed source field')
return items[0]
def load_snapshot(root, name):
expected, key = SNAPSHOTS[name]
directory = root/name; path = directory/'manifest.json'
if root.is_symlink() or directory.is_symlink() or path.is_symlink() or path.stat().st_size > 256*1024:
raise ValueError('Unsafe snapshot manifest')
raw = path.read_bytes()
if digest(raw) != expected:
raise ValueError('Reviewed snapshot manifest hash changed')
manifest = json.loads(raw)
if datetime.fromisoformat(manifest['at']).utcoffset() is None:
raise ValueError('Snapshot timestamp requires timezone')
files = {'manifest.json': raw}
for receipt in manifest[key]:
if 'file' not in receipt:
# A recorded 403 is an access observation, not an empty data set.
continue
filename = receipt['file']; path = directory/filename
if Path(filename).name != filename or path.is_symlink() or path.stat().st_size > 24*1024**2:
raise ValueError('Unsafe snapshot source file')
body = path.read_bytes()
if (receipt['http_status'] != 200 or receipt['url'] != receipt['final_url']
or receipt['bytes'] != len(body) or receipt['sha256'] != digest(body)):
raise ValueError('Snapshot receipt/body disagree')
if filename in files:
raise ValueError('Repeated snapshot source file')
files[filename] = body
return manifest, files
def registry_papers(message):
items = message.get('items', [])
if message.get('total-results') != 55 or len(items) != 55:
raise ValueError('Incomplete exact-ISBN registry census')
if Counter(x.get('type') for x in items) != {'book-chapter': 53, 'book': 1, 'other': 1}:
raise ValueError('Unreviewed publication item types')
papers = []; excluded = []; seen = set()
for original in items:
doi = original.get('DOI')
if not isinstance(doi, str) or doi in seen or original.get('ISBN') != [ISBN]:
raise ValueError('Missing/repeated DOI or wrong ISBN')
seen.add(doi)
if doi in (BOOK, BOOK+'.fm'):
expected = ('book', [TITLE]) if doi == BOOK else ('other', ['Front Matter'])
if (original['type'], original.get('title')) != expected:
raise ValueError('Unreviewed excluded volume/front matter')
excluded.append({'doi': doi, 'type': original['type'], 'title': original['title']})
continue
if (original.get('type') != 'book-chapter' or original.get('container-title') != [TITLE]
or not re.fullmatch(re.escape(BOOK)+r'\.[1-9][0-9]*', doi)
or original.get('URL') != 'https://doi.org/'+doi
or original.get('publisher') != 'Society for Industrial and Applied Mathematics'
or original.get('relation', {}).get('is-part-of') != [
{'id-type': 'doi', 'id': 'https://doi.org/'+BOOK, 'asserted-by': 'object'}]):
raise ValueError('Wrong publication edition, publisher or parent DOI')
title = one(original.get('title', []))
if not isinstance(title, str) or not clean(title):
raise ValueError('Missing registry title')
flags = []
if '<' in title or '>' in title:
match = re.fullmatch(r'([^<>]+)<sup>\*</sup>\s*', title, re.S)
if not match:
raise ValueError('Unreviewed title markup')
comparison_title = clean(match[1])
flags.append('title_sup_asterisk_footnote_requires_review')
else:
comparison_title = clean(Document(title).root.text())
displayed_title = clean(Document(title).root.text())
authors = original.get('author')
if (not isinstance(authors, list) or not authors
or any(not isinstance(a.get('given'), str) or not isinstance(a.get('family'), str)
or not clean(a['given']) or not clean(a['family']) for a in authors)
or [a.get('sequence') for a in authors] != ['first']+['additional']*(len(authors)-1)):
raise ValueError('Incomplete registry authors or order evidence')
names = [clean(a['given']+' '+a['family']) for a in authors]
if any('@' in n or 'anonymous' in n.casefold() for n in names):
raise ValueError('Unreviewed registry author identity')
dates = {}
for field in ('published', 'published-print', 'published-online', 'issued'):
if field in original:
parts = one(original[field]['date-parts'])
if len(parts) != 3 or any(type(x) is not int for x in parts) or parts[0] != 2026:
raise ValueError('Unreviewed registry publication date')
dates[field] = date(*parts).isoformat()
if not dates or original.get('abstract'):
raise ValueError('Reviewed registry date/abstract availability changed')
papers.append({'original': original, 'title': displayed_title, 'comparison_title': comparison_title,
'authors': names, 'review_flags': flags, 'publication_dates': dates})
if {p['original']['DOI'] for p in papers} != {BOOK+'.'+str(n) for n in range(1, 54)} or len(excluded) != 2:
raise ValueError('Incomplete reviewed 53-chapter DOI set')
return sorted(papers, key=lambda p: int(p['original']['DOI'].rsplit('.', 1)[1])), excluded
class SessionParser(HTMLParser):
"""The source omits closing dt/dd tags; each next dt starts a new event."""
def __init__(self):
super().__init__(); self.rows = []; self.current = None; self.strong = False; self.author = False
def finish(self):
if self.current is not None:
match = re.fullmatch(r'(\d{1,2}:\d{2}-\d{1,2}:\d{2})\s+(.+)', clean(self.current.pop('timed_title')))
if not match or not clean(self.current['authors_literal']):
raise ValueError('Incomplete contributed program event')
self.current['time_literal'], self.current['title'] = match.groups()
self.current['authors_literal'] = clean(self.current['authors_literal'])
self.rows.append(self.current)
self.current = None; self.strong = False; self.author = False
def handle_starttag(self, tag, attrs):
attrs = dict(attrs)
if tag == 'dt':
self.finish()
self.current = {'timed_title': '', 'authors_literal': '', 'cancelled_presentation': False, 'abstract_links': []}
if self.current is not None:
if tag == 'strong': self.strong = True
elif tag == 'dd': self.author = True
elif tag == 's': self.current['cancelled_presentation'] = True
elif tag == 'a' and attrs.get('href', '').startswith('dsp_talk.cfm?p='):
self.current['abstract_links'].append(attrs['href'])
def handle_endtag(self, tag):
if tag == 'strong': self.strong = False
elif tag == 'dl': self.finish()
def handle_data(self, text):
if self.current is not None:
if self.strong: self.current['timed_title'] += text
elif self.author: self.current['authors_literal'] += text
elif clean(text) == 'Cancelled': self.current['cancelled_presentation'] = True
def talk_fields(raw, session_url):
nodes = list(Document(raw.decode('utf-8')).root.walk())
if one([clean(n.text()) for n in nodes if n.tag == 'h1']) != '2026 SIAM International Conference on Data Mining':
raise ValueError('Wrong program page edition')
title = clean(one([n for n in nodes if n.tag == 'div' and n.attrs.get('class') == 'ptitle']).text())
abstract_node = one([n for n in nodes if n.tag == 'div' and n.attrs.get('class') == 'abstract'])
abstract = clean(abstract_node.text())
if not title or not abstract.startswith('Abstract.') or not abstract[len('Abstract.'):].strip():
raise ValueError('Missing public program title/abstract')
reference = one([n for n in nodes if n.tag == 'div' and n.attrs.get('class') == 'sesref'])
link = one([n.attrs.get('href', '') for n in reference.walk() if n.tag == 'a'])
def session_key(url):
u = urlsplit(url)
query = {k.lower(): v for k, v in parse_qs(u.query).items()}
if u.scheme != 'https' or u.netloc != 'meetings.siam.org' or u.path != '/sess/dsp_programsess.cfm' or set(query) != {'sessioncode'}:
raise ValueError('Foreign program session link')
return query['sessioncode']
if session_key(link) != session_key(session_url):
raise ValueError('Program page belongs to a different session')
authors = []
for n in nodes:
if n.tag == 'li':
# Source rows contain name, affiliations, country and contact email.
# Only the independently checked leading name is a metadata field.
literal = clean(n.text()); name, separator, _ = literal.partition(',')
if not separator or not name or '@' in name:
raise ValueError('Unreviewed program author row')
authors.append(name)
if not authors:
raise ValueError('Missing source-reported program authors')
markup = sorted({n.tag for n in abstract_node.walk() if n.tag in ('math', 'sup', 'sub', 'svg', 'img')})
return {'title': title, 'source_reported_authors': authors, 'abstract': abstract[len('Abstract.'):].strip(),
'abstract_markup_tags': markup, 'review_flags': ['structured_abstract_markup_requires_review'] if markup else [],
'author_completeness': 'Program-reported names only; not asserted to be the complete or final publisher author sequence.'}
def program_observations(snapshots, papers):
official, official_files = snapshots['sdm2026-official-research-v1']
sessions, session_files = snapshots['sdm2026-program-research-v1']
pages, page_files = snapshots['sdm2026-public-abstract-research-v1']
main_nodes = list(Document(official_files['program.html'].decode()).root.walk())
links = {urljoin(PROGRAM, n.attrs.get('href', '')): clean(n.text()) for n in main_nodes
if n.tag == 'a' and re.match(r'CP[1-9]\s', clean(n.text()))}
if len(links) != 9 or {s['url']: s['title'] for s in sessions['sessions']} != links:
raise ValueError('Contributed session membership disagrees with main program')
receipts = {r['url']: r for r in pages['records']}
if len(receipts) != len(pages['records']) or len(receipts) != 52:
raise ValueError('Incomplete or repeated public talk receipts')
rows = []; used = set()
for session in sessions['sessions']:
parser = SessionParser(); parser.feed(session_files[session['file']].decode()); parser.finish()
for position, event in enumerate(parser.rows, 1):
urls = [urljoin(session['url'], link) for link in event.pop('abstract_links')]
event.update(record_kind='source_observation_not_import_record', session=session['title'],
session_url=session['url'], session_sha256=session['sha256'], session_position=position)
if not urls and event['cancelled_presentation']:
event.update(program_page=None, public_page_fields=None)
else:
url = one(urls)
if url not in receipts or url in used:
raise ValueError('Missing or repeated public program page')
used.add(url); receipt = receipts[url]
if receipt['session_url'] != session['url']:
raise ValueError('Public page receipt session disagrees')
fields = talk_fields(page_files[receipt['file']], session['url'])
if fields['title'] != event['title']:
fields['review_flags'].append('session_page_title_versions_differ_requires_review')
event.update(program_page=url, public_page_receipt=receipt, public_page_fields=fields)
candidates = []
for paper in papers:
if paper['comparison_title'].casefold() == event['title'].casefold():
fields = event['public_page_fields']
candidates.append({'registry_doi_candidate': paper['original']['DOI'],
'comparison': 'title_candidate_only_not_identity_approval',
'registry_title_review_flags': paper['review_flags'],
'author_sequence_exact': fields['source_reported_authors'] == paper['authors'] if fields else None})
event['registry_candidates'] = candidates
event['scope'] = 'Separate program version; cancellation is presentation status only. No original submission ID, decision, work identity or author alias inferred.'
rows.append(event)
if len(rows) != 53 or used != set(receipts) or sum(r['cancelled_presentation'] for r in rows) != 1:
raise ValueError('Incomplete reviewed program event census')
return rows
def collect(out, *, snapshots):
out = Path(out)
if out.exists():
raise ValueError('Output exists; inspect previous operation before retrying')
out.mkdir(parents=True)
try:
loaded = {name: load_snapshot(Path(snapshots), name) for name in SNAPSHOTS}
registry_manifest, registry_files = loaded['sdm2026-crossref-research-v1']
message = json.loads(registry_files['isbn-works.json'])['message']
papers, excluded = registry_papers(message)
observations = program_observations(loaded, papers)
evidence = []; directory = out/'evidence'; directory.mkdir()
for name, (manifest, files) in loaded.items():
for filename, raw in files.items():
rel = 'evidence/'+digest(raw)+Path(filename).suffix
(out/rel).write_bytes(raw)
evidence.append({'snapshot': name, 'file': filename, 'output_file': rel, 'sha256': digest(raw)})
receipt = one([r for r in registry_manifest['sources'] if r.get('file') == 'isbn-works.json'])
rows = []
for p in papers:
original = p['original']; doi = original['DOI']
rows.append({'source': SOURCE, 'external_id': doi, 'url': original['URL'], 'title': p['title'],
'authors': p['authors'], 'abstract': None, 'identifiers': {'doi': doi},
'submission_identifiers': [{'scheme': 'publication_doi', 'value': doi}],
'venue': 'SDM', 'year': 2026, 'track': TRACK, 'status': 'accepted', 'tags': [],
'observed_at': receipt['at'], 'evidence_url': QUERY,
'raw_decision': 'Registered research chapter in the exact SDM2026 proceedings ISBN; not an original signed review decision.',
'metadata': {'registry_fields': original, 'publication_dates': p['publication_dates'],
'registry_response_sha256': digest(registry_files['isbn-works.json']), 'review_flags': p['review_flags'],
'original_decision_signature_verified': False,
'abstract_availability': 'No abstract in frozen Crossref deposit; public program abstracts remain separate source versions.',
'title_transform': 'Display text with footnote asterisk retained; raw title preserved. Footnote-free text is used only to list review candidates.',
'scope': 'Requires fresh work/DOI/history review before import. No automatic program fusion, track, person alias or negative decision inference.'}})
path = out/'sdm-crossref-registry-2026-unclassified.jsonl'
payload = ''.join(json.dumps(r, ensure_ascii=False)+'\n' for r in rows).encode(); path.write_bytes(payload)
file_manifest = {'source': SOURCE, 'parser_version': 1, 'venue': 'SDM', 'year': 2026, 'track': TRACK,
'count': 53, 'file': str(path), 'file_sha256': digest(payload), 'list_url': QUERY, 'status': 'partial', 'missing_abstract_count': 53}
path.with_suffix('.manifest.json').write_text(json.dumps(file_manifest, indent=2)+'\n')
sidecar = out/'program-observations.jsonl'
raw = ''.join(json.dumps(r, ensure_ascii=False)+'\n' for r in observations).encode(); sidecar.write_bytes(raw)
report = {'status': 'collected_pending_identity_review', 'records': 53, 'files': [file_manifest],
'missing_abstract_count': 53, 'excluded_registry_items': excluded,
'program_observations': {'file': str(sidecar), 'sha256': digest(raw), 'count': 53, 'abstract_pages': 52,
'cancelled_presentations': 1, 'not_an_import_manifest': True},
'program_registry_author_sequence_differences': sum(c['author_sequence_exact'] is False for o in observations for c in o['registry_candidates']),
'evidence': evidence, 'public_rejected_metadata': 'unknown_anonymous_query_http403',
'database_written': False, 'network_requests': 0, 'pdfs_downloaded': False,
'scope': 'Registry publication records and program source observations remain separate; no work merge, metadata fusion or final coverage/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.
+142
View File
@@ -0,0 +1,142 @@
import copy
import gzip
import hashlib
import json
from pathlib import Path
import tempfile
import unittest
from paper_atlas.collectors import sdm2026 as c
class SDM2026Tests(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.fixture = json.loads(gzip.decompress((Path(__file__).parent/'fixtures/sdm2026-source.json.gz').read_bytes()))
cls.message = json.loads(cls.fixture['files']['sdm2026-crossref-research-v1/isbn-works.json'])['message']
def snapshots(self, root):
for name, text in self.fixture['files'].items():
path = root/name; path.parent.mkdir(parents=True, exist_ok=True); path.write_bytes(text.encode('utf-8'))
def test_publication_census_excludes_book_and_front_matter_not_cancelled_paper(self):
papers, excluded = c.registry_papers(self.message)
self.assertEqual(len(papers), 53)
self.assertEqual({x['doi'] for x in excluded}, {c.BOOK, c.BOOK+'.fm'})
self.assertIn(c.BOOK+'.7', {p['original']['DOI'] for p in papers})
marked = [p for p in papers if p['review_flags']]
self.assertEqual({p['original']['DOI'] for p in marked}, {c.BOOK+'.12', c.BOOK+'.35'})
for p in marked:
self.assertTrue(p['title'].endswith('*'))
self.assertIn('<sup>*</sup>', p['original']['title'][0])
def test_full_collection_preserves_separate_source_versions_and_unknown_negatives(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp); self.snapshots(root/'src')
report = c.collect(root/'out', snapshots=root/'src')
self.assertEqual((report['records'], report['missing_abstract_count']), (53, 53))
self.assertFalse(report['database_written']); self.assertEqual(report['network_requests'], 0)
self.assertEqual(report['public_rejected_metadata'], 'unknown_anonymous_query_http403')
self.assertEqual(report['program_registry_author_sequence_differences'], 20)
manifest, = report['files']; self.assertEqual(manifest['status'], 'partial')
rows = [json.loads(line) for line in Path(manifest['file']).read_text().splitlines()]
originals = {x['DOI']: x for x in self.message['items'] if x['type'] == 'book-chapter'}
for row in rows:
self.assertEqual(row['metadata']['registry_fields'], originals[row['external_id']])
self.assertIsNone(row['abstract']); self.assertFalse(row['tags'])
self.assertEqual(row['identifiers'], {'doi': row['external_id']})
self.assertEqual(row['submission_identifiers'], [{'scheme': 'publication_doi', 'value': row['external_id']}])
self.assertEqual((row['status'], row['track']), ('accepted', 'proceedings_unclassified'))
self.assertEqual(row['metadata']['publication_dates']['published-online'], '2026-11-12')
sidecar = report['program_observations']
self.assertTrue(sidecar['not_an_import_manifest'])
observations = [json.loads(line) for line in Path(sidecar['file']).read_text().splitlines()]
self.assertEqual(len(observations), 53)
self.assertEqual(sum(x['public_page_fields'] is not None for x in observations), 52)
for observation in observations:
self.assertNotIn('status', observation); self.assertNotIn('identifiers', observation)
self.assertEqual(observation['record_kind'], 'source_observation_not_import_record')
cancelled, = [x for x in observations if x['cancelled_presentation']]
self.assertEqual(cancelled['title'], 'A Trajectory View of Statistical Policy Comparison')
self.assertIsNone(cancelled['public_page_fields'])
# The real source has conflicting author sets under the same title.
conflict = next(x for x in observations if x['title'].startswith('Scaffold-Aware Generative'))
self.assertEqual(conflict['public_page_fields']['source_reported_authors'], ['Xin Wang', 'Qinwen Ge'])
candidate, = conflict['registry_candidates']; self.assertFalse(candidate['author_sequence_exact'])
registry_row = next(x for x in rows if x['external_id'] == candidate['registry_doi_candidate'])
self.assertEqual(registry_row['authors'], ['Xin Wang', 'Yu Wang', 'Yunchao Liu', 'Jens Meiler', 'Tyler Derr'])
# Every copied source and manifest remains byte-identical, including CRLF.
for evidence in report['evidence']:
body = (root/'out'/evidence['output_file']).read_bytes()
self.assertEqual(body, self.fixture['files'][evidence['snapshot']+'/'+evidence['file']].encode('utf-8'))
self.assertEqual(hashlib.sha256(body).hexdigest(), evidence['sha256'])
with self.assertRaisesRegex(ValueError, 'Output exists'):
c.collect(root/'out', snapshots=root/'src')
def test_all_program_events_match_independent_inventory(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp); self.snapshots(root)
loaded = {name: c.load_snapshot(root, name) for name in c.SNAPSHOTS}
papers, _ = c.registry_papers(self.message)
observations = c.program_observations(loaded, papers)
gold = {(x['session_url'], x['title']): x for x in self.fixture['program_gold']}
self.assertEqual({(x['session_url'], x['title']) for x in observations}, set(gold))
for item in observations:
original = gold[(item['session_url'], item['title'])]
for key in ('authors_literal', 'time_literal', 'cancelled_presentation'):
self.assertEqual(item[key], original[key])
def test_all_public_abstracts_and_source_reported_names_match_independent_gold(self):
files = self.fixture['files']
manifest = json.loads(files['sdm2026-public-abstract-research-v1/manifest.json'])
receipts = {r['url']: r for r in manifest['records']}
self.assertEqual(len(self.fixture['page_gold']), 52)
for gold in self.fixture['page_gold']:
receipt = receipts[gold['program_url']]
raw = files['sdm2026-public-abstract-research-v1/'+receipt['file']].encode('utf-8')
self.assertEqual(hashlib.sha256(raw).hexdigest(), gold['page_sha256'])
fields = c.talk_fields(raw, receipt['session_url'])
self.assertEqual(fields['title'], gold['title'])
self.assertEqual(fields['source_reported_authors'], gold['authors'])
self.assertEqual(fields['abstract'], gold['abstract'])
self.assertTrue(all('@' not in name for name in fields['source_reported_authors']))
def test_wrong_isbn_parent_doi_and_incomplete_census_rejected(self):
for kind in ('isbn', 'parent', 'repeat', 'type', 'year', 'author', 'markup', 'abstract', 'missing'):
with self.subTest(kind=kind):
message = copy.deepcopy(self.message)
row = next(x for x in message['items'] if x['type'] == 'book-chapter')
if kind == 'isbn': row['ISBN'] = ['9780000000000']
if kind == 'parent': row['relation']['is-part-of'][0]['id'] = 'https://doi.org/10.9999/other'
if kind == 'repeat': row['DOI'] = c.BOOK
if kind == 'type': row['type'] = 'other'
if kind == 'year': row['published']['date-parts'] = [[2025, 10, 31]]
if kind == 'author': row['author'][0]['sequence'] = 'additional'
if kind == 'markup': row['title'] = ['Changed <sup>2</sup> title']
if kind == 'abstract': row['abstract'] = 'Unreviewed new version'
if kind == 'missing': message['items'].pop()
with self.assertRaises(ValueError): c.registry_papers(message)
def test_foreign_session_and_changed_edition_rejected(self):
raw = self.fixture['files']['sdm2026-public-abstract-research-v1/162766.html'].encode()
url = 'https://meetings.siam.org/sess/dsp_programsess.cfm?SESSIONCODE=91511'
for changed in [raw.replace(b'sessioncode=91511', b'sessioncode=91507'),
raw.replace(b'2026 SIAM International Conference', b'2025 SIAM International Conference')]:
with self.assertRaises(ValueError): c.talk_fields(changed, url)
def test_corruption_preserves_failure_without_success_report(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp); self.snapshots(root/'src')
(root/'src/sdm2026-public-abstract-research-v1/162766.html').write_bytes(b'corrupt')
with self.assertRaisesRegex(ValueError, 'receipt/body'):
c.collect(root/'out', snapshots=root/'src')
self.assertTrue((root/'out/failed.json').exists())
self.assertFalse((root/'out/collection-report.json').exists())
def test_source_symlink_rejected(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp); self.snapshots(root/'src')
p = root/'src/sdm2026-crossref-research-v1/isbn-works.json'
original = p.read_bytes(); p.unlink(); (root/'data.json').write_bytes(original); p.symlink_to(root/'data.json')
with self.assertRaisesRegex(ValueError, 'Unsafe'):
c.collect(root/'out', snapshots=root/'src')