Compare commits
11 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 7d9c693761 | |||
| 98ef415912 | |||
| 90fd789a17 | |||
| f982851560 | |||
| 88aeb4d2eb | |||
| 83f3ebd26b | |||
| 13536d6268 | |||
| 7e63035570 | |||
| 44f4146f3e | |||
| b0bd9b69dc | |||
| 09a2d4ed20 |
@@ -19,8 +19,9 @@ jobs:
|
||||
# Public repository archive: no owner credentials or runner SSH keys.
|
||||
url = 'https://gitea.entroia.com/Zayden-Jung/paper-atlas/archive/' + sha + '.tar.gz'
|
||||
with urllib.request.urlopen(url, timeout=60) as response:
|
||||
archive = response.read(16*1024*1024+1)
|
||||
if len(archive)>16*1024*1024:
|
||||
# Full publication-source fixtures retain original bibliographic evidence.
|
||||
archive = response.read(24*1024*1024+1)
|
||||
if len(archive)>24*1024*1024:
|
||||
raise SystemExit('Unexpectedly large source archive')
|
||||
with tempfile.TemporaryDirectory(prefix='paper-atlas-ci-') as tmp:
|
||||
root = pathlib.Path(tmp)
|
||||
@@ -34,7 +35,7 @@ jobs:
|
||||
venv = root / 'venv'
|
||||
subprocess.run(['python3','-B','-m','venv','--system-site-packages',str(venv)],check=True)
|
||||
python = str(venv / 'bin/python')
|
||||
subprocess.run([python,'-B','-m','pip','install','--only-binary=:all:','--no-compile','psycopg[binary]==3.3.8'],check=True)
|
||||
subprocess.run([python,'-B','-m','pip','install','--only-binary=:all:','--no-compile','psycopg[binary]==3.3.8','pyarrow==26.0.1'],check=True)
|
||||
testenv = os.environ.copy()
|
||||
service_file = pathlib.Path('/etc/paper-atlas/pg_service.conf')
|
||||
if service_file.is_file():
|
||||
@@ -66,6 +67,33 @@ jobs:
|
||||
assert hashlib.sha256(row['abstract'].encode()).hexdigest() == gold['abstract_sha256']
|
||||
assert [c['source_pointer'] for c in row['contexts']] == gold['locations']
|
||||
assert not row['approved_for_import'] and not row['final_decision_verified'] and 'emails' not in row
|
||||
mm_pub_fixture = json.loads(gzip.decompress((src/'tests/fixtures/mm2026-publications.json.gz').read_bytes()))
|
||||
mm_pub_registry = root/'mm2026-publication-registry'
|
||||
for name,body in mm_pub_fixture['files'].items():
|
||||
path=mm_pub_registry/name
|
||||
path.parent.mkdir(parents=True,exist_ok=True)
|
||||
path.write_bytes(body.encode('utf-8'))
|
||||
subprocess.run([python,'-I','-B','-m','paper_atlas','collect','mm2026-publications','--registry-snapshots',str(mm_pub_registry),'--program-snapshots',str(mm_snapshots),'--parent-snapshot',str(mm_pub_registry/'parent.json'),'--out',str(root/'installed-mm2026-publications')],cwd=outside,check=True)
|
||||
mm_pub_report=json.loads((root/'installed-mm2026-publications/collection-report.json').read_text())
|
||||
assert (mm_pub_report['records'],mm_pub_report['unique_registered_dois'],mm_pub_report['abstracts_from_exact_program_correspondence'],mm_pub_report['output_missing_abstract_count'])==(1918,1918,1717,201)
|
||||
assert mm_pub_report['unmatched_program_count']==417 and not mm_pub_report['database_written'] and mm_pub_report['approved_for_import']==0
|
||||
mm_pub_rows={r['external_id']:r for f in mm_pub_report['files'] for r in map(json.loads,pathlib.Path(f['file']).read_text().splitlines())}
|
||||
assert len(mm_pub_rows)==1918
|
||||
for gold in mm_pub_fixture['independent_gold']:
|
||||
row=mm_pub_rows[gold['doi']]
|
||||
assert (row['title'],row['authors'],row['track'])==(gold['title'],gold['authors'],gold['track'])
|
||||
assert (hashlib.sha256(row['abstract'].encode()).hexdigest() if row['abstract'] is not None else None)==gold['abstract_sha256']
|
||||
assert row['metadata']['program_match_kind']==gold['kind'] and not row['metadata']['scientific_work_identity_approved'] and not row['metadata']['research_type_classified']
|
||||
if service_file.is_file():
|
||||
subprocess.run([python,'-I','-B','-m','paper_atlas','audit','source-identities','--files',*[f['file'] for f in mm_pub_report['files']],'--pg-service','foundry-research','--pg-schema','research','--out',str(root/'installed-mm2026-publication-preflight')],cwd=outside,env=testenv,check=True)
|
||||
mm_pub_preflight=json.loads((root/'installed-mm2026-publication-preflight/postgresql-preflight.json').read_text())
|
||||
assert mm_pub_preflight['records']==1918 and mm_pub_preflight['transaction_read_only'] and not mm_pub_preflight['database_written'] and mm_pub_preflight['approved_for_import']==0
|
||||
subprocess.run([python,'-I','-B','-m','paper_atlas','audit','source-identities','--mm-program-report',str(root/'installed-mm2026/collection-report.json'),'--pg-service','foundry-research','--pg-schema','research','--out',str(root/'installed-mm2026-preflight')],cwd=outside,env=testenv,check=True)
|
||||
mm_preflight = json.loads((root/'installed-mm2026-preflight/postgresql-preflight.json').read_text())
|
||||
assert mm_preflight['records'] == 2143 and len(mm_preflight['inputs']) == 28
|
||||
assert mm_preflight['transaction_read_only'] and not mm_preflight['database_written'] and mm_preflight['approved_for_import'] == 0
|
||||
assert sum(len(r['program_contexts']) for r in mm_preflight['entries']) == 4326
|
||||
assert all(r['status'] == 'unknown' and not r['identifiers'] and not r['contribution_type_verified'] and not r['publication_identity_verified'] and not r['final_decision_verified'] for r in mm_preflight['entries'])
|
||||
subprocess.run([str(venv/'bin/paper-atlas'),'--help'],cwd=outside,check=True)
|
||||
subprocess.run([python,'-I','-B','-m','paper_atlas','collect','cvf','--help'],cwd=outside,check=True)
|
||||
subprocess.run([python,'-I','-B','-m','paper_atlas','collect','rss2026-proceedings','--help'],cwd=outside,check=True)
|
||||
|
||||
@@ -0,0 +1,88 @@
|
||||
name: ReviewArena metadata projection
|
||||
on:
|
||||
workflow_dispatch:
|
||||
inputs:
|
||||
split:
|
||||
description: 'Pinned public mirror family; metadata columns only'
|
||||
required: true
|
||||
default: emnlp
|
||||
projection_name:
|
||||
description: 'Optional existing metadata output directory name for offline replay'
|
||||
required: false
|
||||
default: ''
|
||||
projection_sha256:
|
||||
description: 'Exact cached collection-report.json SHA256, required for offline replay'
|
||||
required: false
|
||||
default: ''
|
||||
jobs:
|
||||
project:
|
||||
runs-on: ces-control
|
||||
env:
|
||||
ARCHIVE_SHA: ${{ github.sha }}
|
||||
MIRROR_SPLIT: ${{ inputs.split }}
|
||||
MIRROR_PROJECTION: ${{ inputs.projection_name }}
|
||||
MIRROR_PROJECTION_SHA: ${{ inputs.projection_sha256 }}
|
||||
PYTHONDONTWRITEBYTECODE: '1'
|
||||
steps:
|
||||
- name: Project permitted metadata on the approved remote runner
|
||||
run: |
|
||||
python3 -B - <<'PY'
|
||||
import hashlib, io, json, os, pathlib, re, subprocess, tarfile, tempfile, urllib.request
|
||||
sha = os.environ['ARCHIVE_SHA']
|
||||
split = os.environ['MIRROR_SPLIT']
|
||||
cached = os.environ.get('MIRROR_PROJECTION','')
|
||||
cached_sha = os.environ.get('MIRROR_PROJECTION_SHA','')
|
||||
if not re.fullmatch(r'[0-9a-f]{40}', sha) or split not in ('colm','corl','emnlp','iclr','icml','neurips','tmlr'):
|
||||
raise SystemExit('Explicit approved source commit/split required')
|
||||
base = pathlib.Path('/data/paper-index')
|
||||
if not (base/'collected').is_dir() or not (base/'validation').is_dir():
|
||||
raise SystemExit('Confirmed ces-control host dataset paths unavailable; do not use another host')
|
||||
out = base/'collected'/('reviewarena-c2978add-'+split+'-'+sha[:8])
|
||||
if out.exists():
|
||||
raise SystemExit('Inspect existing output; no overwrite or automatic retry')
|
||||
extra=[]
|
||||
if cached:
|
||||
if not re.fullmatch('reviewarena-c2978add-'+split+r'-[0-9a-f]{8}',cached) or not re.fullmatch('[0-9a-f]{64}',cached_sha):
|
||||
raise SystemExit('Explicit permitted cached projection name and report hash required')
|
||||
extra=['--from-projection',str(base/'collected'/cached/'collection-report.json'),'--projection-sha256',cached_sha]
|
||||
elif cached_sha:
|
||||
raise SystemExit('Cached projection name required')
|
||||
url = 'https://gitea.entroia.com/Zayden-Jung/paper-atlas/archive/'+sha+'.tar.gz'
|
||||
with urllib.request.urlopen(url,timeout=60) as response:
|
||||
archive = response.read(24*1024*1024+1)
|
||||
if len(archive)>24*1024*1024:
|
||||
raise SystemExit('Oversized source archive')
|
||||
with tempfile.TemporaryDirectory(prefix='paper-atlas-mirror-') as tmp:
|
||||
root = pathlib.Path(tmp)
|
||||
with tarfile.open(fileobj=io.BytesIO(archive),mode='r:gz') as tar:
|
||||
tar.extractall(root,filter='data')
|
||||
candidates=list(root.glob('*/pyproject.toml'))
|
||||
if len(candidates)!=1:
|
||||
raise SystemExit('Unexpected repository archive')
|
||||
src=candidates[0].parent
|
||||
venv=root/'venv'
|
||||
subprocess.run(['python3','-B','-m','venv','--system-site-packages',str(venv)],check=True)
|
||||
python=str(venv/'bin/python')
|
||||
subprocess.run([python,'-B','-m','pip','install','--only-binary=:all:','--no-compile','pyarrow==26.0.1'],check=True)
|
||||
wheels=root/'wheels'
|
||||
subprocess.run([python,'-B','-m','pip','wheel','--no-deps','--wheel-dir',str(wheels),str(src)],check=True)
|
||||
wheel,=wheels.glob('paper_atlas-*.whl')
|
||||
subprocess.run([python,'-B','-m','pip','install','--no-deps',str(wheel)],check=True)
|
||||
outside=root/'outside';outside.mkdir()
|
||||
subprocess.run([python,'-I','-B','-m','paper_atlas','collect','reviewarena-metadata','--split',split,'--out',str(out),*extra],cwd=outside,check=True)
|
||||
report=json.loads((out/'collection-report.json').read_bytes())
|
||||
assert report['database_written'] is False and report['approved_for_import']==0
|
||||
receipt={'source_commit':sha,'source_archive_sha256':hashlib.sha256(archive).hexdigest(),'wheel_sha256':hashlib.sha256(wheel.read_bytes()).hexdigest(),'dataset_commit':report['commit'],'split':split,'records':report['records'],'metadata_bytes_read':report['metadata_bytes_read'],'cached_projection_report_sha256':report.get('cached_projection_report_sha256'),'raw_parquet_saved':False,'database_written':False}
|
||||
(out/'runner-receipt.json').write_text(json.dumps(receipt,indent=2)+'\n')
|
||||
backup=base/'validation'/(out.name+'-recovery.tar.gz')
|
||||
if backup.exists():
|
||||
raise SystemExit('Recovery archive already exists; inspect')
|
||||
with tarfile.open(backup,'w:gz') as tar:
|
||||
tar.add(out,arcname=str(out.relative_to(base)))
|
||||
backup.chmod(0o444)
|
||||
for path in out.iterdir():
|
||||
if path.is_file():path.chmod(0o444)
|
||||
result={'status':'metadata_projection_backed_up_not_imported','archive':str(backup),'bytes':backup.stat().st_size,'sha256':hashlib.sha256(backup.read_bytes()).hexdigest(),**receipt}
|
||||
(base/'validation'/(out.name+'-recovery.json')).write_text(json.dumps(result,indent=2)+'\n')
|
||||
print(json.dumps(result),flush=True)
|
||||
PY
|
||||
@@ -2,6 +2,8 @@
|
||||
|
||||
**长期维护、按需运行的论文元数据与公开投稿历史工具。**
|
||||
|
||||
其他会议的公开评审镜像可按 [ReviewArena 元数据投影流程](docs/reviewarena-metadata.md) 核查:仅取固定版本的标题、摘要、作者及决定等索引字段,保留镜像声明和稳定行定位,不下载正文/评审数据页,也不将镜像 ID 自动提升为官方论坛身份。
|
||||
|
||||
Paper Atlas 将 arXiv、AI 会议论文集及公开投稿决定整理到可追溯的数据库中,用于查询和后续研究分析。生产数据使用 PostgreSQL,支持从既有 SQLite 索引完整迁移;SQLite 保留离线核验和小型样本用途。它按批次运行,不需要常驻 Web 服务;采集标题、摘要、标识、标签及决定证据,不下载论文全文。
|
||||
|
||||
项目仍在建设中。候选范围包含 43 个会议(含 ACM MM、MLSys),尚未完成全部历史数据和最终去重复核。实际进度以数据库、来源清单和覆盖报告为准,而不是仓库文件数量。
|
||||
@@ -142,3 +144,5 @@ ECCV 2026 的[暂定节目采集器](docs/eccv2026-program.md)合并重复节目
|
||||
`collect recsys2026-program` 核验九个 track 的 263 项官方录用贡献观察(260 个不同 DOI),保留摘要、原始作者和重复 DOI 的来源版本。见 [RecSys 2026 使用与恢复说明](docs/recsys2026-program.md)。
|
||||
|
||||
`collect mm2026-program` 从已冻结官方快照恢复全部 28 个来源 track 的 2,143 个节目 ID,保留 4,326 次展示的来源位置与场次上下文,单列 216 个摘要缺失项;输出供贡献类型、出版及作品身份核查使用,不是可直接导入的论文/决定队列。见 [MM 2026 使用、来源和恢复说明](docs/mm2026-source-status.md)。
|
||||
|
||||
`collect mm2026-publications` 核验完整论文集的 1,918 个注册出版 DOI,并单独关联节目分轨、展示上下文和摘要。1,717 份节目摘要明确标注来源,192 项标题/作者版本差异保留为待核;所有节目及未匹配项均保留。注册文章类型不等于研究类型,输出尚需数据库身份/历史审查。见 [MM 2026 出版条目采集与恢复](docs/mm2026-publications.md)。
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
# 会议覆盖与公开决定核查
|
||||
|
||||
最新生产进展和剩余工作见 [2026-11-17 补漏状态](backfill-status-2026-11-17.md):固定 278 个会议年份槽位中,270 个有记录、8 个仍为空;这不代表完整覆盖,第 14 步仍未完成。
|
||||
最新生产验收与剩余工作见 [2026-11-17 补漏状态](backfill-status-2026-11-17.md):当日 `14:06:28 UTC` 固定范围快照中,278 个会议年份槽位有 272 个含记录、6 个仍为空;这不代表完整覆盖,第 14 步仍未完成。后续 MM 规范化、只读候选审计与出版核查不等于生产导入。
|
||||
|
||||
**当前工作范围:2020 年起主动补漏。** 2000 年起已发现资料保留,恢复工作从 [补漏范围与恢复清单(2026-11-10)](backfill-handoff-2026-11-10.md) 开始;下文旧日期快照不改写,较早年份排除不算补齐。
|
||||
|
||||
@@ -72,7 +72,7 @@ PGSERVICEFILE=/etc/paper-atlas/pg_service.conf paper-atlas coverage \
|
||||
|
||||
## 优先缺口
|
||||
|
||||
- **UAI 2024–2026**:官方均允许作者选择公开拒稿,当前主库拒稿仍为零。原 notes 接口要求 Challenge 验证,公开镜像尚无已核验来源,必须保留这个真实缺口;不能根据 PMLR 录用名单推断拒稿。参见 [2025 投稿指南](https://www.auai.org/uai2025/submission_instructions)、JSON 内其他年份官方证据。
|
||||
- **UAI 2024–2026**:官方均允许作者选择公开拒稿,首份生产快照拒稿为零。原 notes 接口要求 Challenge 验证;新核验镜像的 2022–2025 年 902 个论坛来自录用子集,不能填补公开拒稿缺口。2024 年的元评审也不能直接转成最终决定,见 [2026-11-17 原始来源与回复结构核查](uai-public-negative-source-status-2026-11-17.md)。不能根据 PMLR 或镜像录用名单反推拒稿;政策参见 [2025 投稿指南](https://www.auai.org/uai2025/submission_instructions)、JSON 内其他年份官方证据。
|
||||
- **EMNLP 2023**:官方允许公开 opt-in 拒稿,目前只导入 9 条。需继续核验实际公开全集。ARR 评审、conference commitment 与最终会议决定分开;[ARR 作者说明](https://aclrollingreview.org/authors) 明确其自愿捐赠数据与 EMNLP 2023 直接投稿发布的范围不同。
|
||||
- **ICLR**:固定快照已观察 2017 与 2020–2026;快照之后,2013/2018/2019 的 1,654 条来源记录已完成独立 PostgreSQL 验收(报告时间 `2026-11-07T19:52:28.378Z`),不回写旧快照计数,也不据此宣称逐届全集已齐。快照中的 10,208 次 unknown 不等于 rejected,近期补采与冲突核对仍在进行;评审推荐不能当最终决定。
|
||||
- **MM / MLSys**:MM 11,200 条书目记录保留 `proceedings_unspecified`;MLSys 的 448 次官方投稿与 313 条 DBLP 记录重叠。2026 MLSys 的研究/工业轨道在 [官方 CFP](https://mlsys.org/Conferences/2026/CallForResearchPapers) 中分开,仍须核对逐篇轨道证据,不能仅凭会议名全部归 main。
|
||||
|
||||
@@ -33,7 +33,7 @@ SDM 新增事件使 SeBA 和 AdamG 同时保有较早 ICLR 拒稿与后来 SDM
|
||||
| ECCV | 2,834 个不同初步节目 ID 及公开摘要已独立采集;1,997 个只读候选历史证据包已保存,尚未入库 | 初步日程不等于 Springer 最终论文集;继续作品/版本和出版核查。见 [来源状态](../eccv2026-source-status.md) |
|
||||
| ICDM | 324 项线下节目观察,明确研究241/应用83;演讲长168/短156,摘要均缺失;规范化和80个候选历史包已核验 | 展示长短不能当成论文类型;需审查作品与正式出版版本,尚未导入主库。见 [采集器说明](../icdm2026-program.md) |
|
||||
| ISMIR | 140 项节目和正式名单均已核验,140份摘要,101组自愿公开评审;两处标题、三处作者名单差异保留,78个候选历史包已核验 | 数字 miniconf ID 不是原始 OpenReview ID;评审意见不转成最终拒稿。继续作品身份/历史关联,尚未入库。采集器见 [PR #63](https://gitea.entroia.com/Zayden-Jung/paper-atlas/pulls/63) |
|
||||
| MM | 官方10.8MB日程JSON已冻结,4,326次观察对应2,143个节目ID、28个来源track;重复展示核心字段一致,216个ID缺摘要,其中两项主轨同时缺作者 | 不是2,143篇已核验科研论文。先区分各轨贡献与活动、保留全部上下文并单列缺失;尚未规范化/入库。见 [来源与访问证据](../mm2026-source-status.md) |
|
||||
| MM | 全28轨2143个节目ID/4326展示位置已规范化;814个只读候选历史证据包已核验,24篇候选涉及24次负面投稿/37条来源观察;67项原始arXiv DOI中66项注册元数据一致、1项404暂缓 | 不是2143篇已核验科研论文。继续逐轨贡献/出版依据与作品审查,0身份批准、0入库。ISBN查询与父XML均只返回父记录,不作逐篇清单。见 [来源、审计与恢复证据](../mm2026-source-status.md) |
|
||||
| CIKM | 官方录用/日程入口与五个分轨页面已保存;五个分轨链接实际是CFP/政策,不是逐篇录用名单 | 查找实际节目或出版社完整目录;TBD和没有静态表不能当作没有论文 |
|
||||
| IROS | 官方论文索引入口的原始HTTP403已保存 | 保留访问限制,不重试绕过;寻找可合法读取、能独立核验的出版元数据 |
|
||||
|
||||
@@ -43,6 +43,8 @@ SDM 新增事件使 SeBA 和 AdamG 同时保有较早 ICLR 拒稿与后来 SDM
|
||||
|
||||
MM 官方群组元数据可匿名读取,但根据其给出的 submission invitation 和 rejected/withdrawn/desk-rejected venue ID,四次限1条的匿名 notes 核查均为403。查询、时间、原文和哈希已保留,没有认证重试或绕过权限。访问限制不等于零条负面记录;详见 [MM来源证据](../mm2026-source-status.md)。ISMIR公开的101组评审来自录用节目,不能记成101条拒稿。
|
||||
|
||||
UAI新增固定镜像已核对2022–2025的902个原始论坛及3053个回复的元数据/字段结构。整理版均标录用,2022保存230条明确Accept决定;2024仅有评审、作者回复、元评审,2023/2025未保存回复,不能据此补拒稿。来源限制和恢复见 [逐年原始核查](uai-public-negative-source-status-2026-11-17.md)。这不是UAI没有公开拒稿的结论,缺口保持未补齐。
|
||||
|
||||
第14步仍需逐会议、年份、track完成公开拒稿/desk-reject/withdrawn来源核查和可得数据采集;继续核查UAI近年自愿公开范围、EMNLP实际覆盖范围及其余已列明缺口。不能从未见录用、评分、评审推荐、缺少摘要或接口不可访问制造负面决定。
|
||||
|
||||
之后仍需完成跨会作品/历次投稿与类型标签关联(第15步),所有导入结束后的独立全库去重、标识冲突、近似标题作者、摘要/约束/查询及覆盖审计(第16步),以及长期运行与恢复文档最终整理(第17步)。阶段性来源入库、测试通过、某年度有记录均不能替代这些验收。
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
# UAI 公开负面决定来源:2026-11-17 核查
|
||||
|
||||
**公开拒稿缺口仍未补齐,总体仍为 13/17。** 本页保存新来源的实际范围与恢复线索,避免下次把录用子集误当作全部投稿。主动补漏范围仍从2020年开始;本次镜像只含2022–2025,不能替代其他年份或其他会议的核查。
|
||||
|
||||
## 政策与访问边界
|
||||
|
||||
已有逐年核查记录保存在 [覆盖清单](inventory-2026-11-08.json) 的 UAI `policy_checks`:2023–2026 官方说明允许决定后公开录用及作者选择公开的拒稿,默认私密拒稿不属于公开可采集范围。政策并不证明任何镜像完整。
|
||||
|
||||
- [2023 投稿说明](https://www.auai.org/uai2023/submission_instructions)
|
||||
- [2024 投稿说明](https://auai.org/uai2024/submission_instructions)
|
||||
- [2025 投稿说明](https://www.auai.org/uai2025/submission_instructions)
|
||||
- [2026 投稿说明](https://www.auai.org/uai2026/submission_instructions)
|
||||
|
||||
先前匿名 notes 请求遇到 ChallengeRequiredError/403,仍作为访问限制保留。本轮没有认证、重试或绕过这些入口;读取的是整理者公开发布的固定版本镜像。复制的 `readers`、`signatures` 和邀请字段仍只是镜像内的来源断言,不是对当前原始接口可见性或签名真实性的验证。
|
||||
|
||||
## 实际镜像范围
|
||||
|
||||
来源:[latkes/ai-research-index-iclr-openreview](https://huggingface.co/datasets/latkes/ai-research-index-iclr-openreview/tree/4c9e09931a7814d7500630ffd064b74f1c01bcf3),固定提交 `4c9e09931a7814d7500630ffd064b74f1c01bcf3`。目录中存在 `papers/uai-YYYY/papers.jsonl` 与四个 `raw/uaiYYYY.jsonl`,不能仅凭数据集说明宣称拒稿覆盖。
|
||||
|
||||
| 年份 | 整理版及原始论坛数 | 原始回复对象 | 实际回复角色 | 标准结构化决定字段 |
|
||||
| --- | ---: | ---: | --- | --- |
|
||||
| 2022 | 230 | 1059 | Official_Review 829,Decision 230 | Accept (Poster) 194,Accept (Oral) 36 |
|
||||
| 2023 | 243 | 0 | 没有保存回复 | 没有保存决定记录 |
|
||||
| 2024 | 200 | 1994 | Official_Review 905,Rebuttal 889,Meta_Review 200 | 没有单独 Decision 记录或结构化 decision 字段 |
|
||||
| 2025 | 229 | 0 | 没有保存回复 | 没有保存决定记录 |
|
||||
|
||||
整理版902条均被整理者标为 `accepted=true`;原始根记录的 `venue` 为相应会议名称或 oral/poster/spotlight 文本,不能替代逐次决定证据,更不能算作902条已认证的原始录用决定。这不是四届全部投稿或全部录用的完整性证明。2024年还有 `Edit` 邀请,属于同一批回复上的另一个邀请,不能再累加成1994个新回复。
|
||||
|
||||
为排除字段漏读,对含回复的2022、2024快照另做了全部3053个回复的字段名、包装结构、邀请角色和短决定类字段核查;重新读取的正文哈希与首次捕获完全一致。2022的230条明确字段是录用标签。2024仅保存评审、作者回复和元评审,没有可据此新增的结构化拒稿/撤稿决定。没有从元评审文本、评分、推荐或缺少录用推断负面决定,也没有保存评审正文、作者回复正文或论文全文。
|
||||
|
||||
**这只描述上述固定镜像子集,不证明UAI没有公开拒稿。** 本批没有生成负面决定导入队列,也未写数据库。2020、2021、2026及其他可用来源仍需分别核查;某文件未出现在这个镜像目录中不代表全网不存在。
|
||||
|
||||
## 产物与恢复
|
||||
|
||||
服务器目录均相对 `/data/paper-index/collected/`:
|
||||
|
||||
- `uai2022-2025-mirror-metadata-v1`:902条整理版元数据投影、字段盘点与标签计数。
|
||||
- `uai2022-2025-raw-decisions-v1`:902个原始论坛的元数据、明确决定字段及来源正文/逐行哈希。
|
||||
- `uai2022-2024-reply-schema-v2`:3053个回复的元数据、字段名与短标签;评审正文不保存。角色核查包含直接记录和 `note` 包装,未发生静默遗漏。
|
||||
|
||||
源响应正文仅在内存中读取、校验,保存元数据投影与固定来源URL/字节哈希;不是保留了含评审文本的完整原始快照。复核原始源字节时应使用固定提交,并区分投影文件哈希与源正文哈希。恢复包已在本机独立核验全部902个论坛、3053个回复的元数据结构、逐文件哈希及来源关联。
|
||||
|
||||
| 恢复证据 | SHA256 |
|
||||
| --- | --- |
|
||||
| 整理版元数据恢复包 | 55889fb2b9005260cef1001b2cf36e203f3b7e1deeec0291f058b52eeae880e7 |
|
||||
| 原始元数据/回复结构恢复包 | caade7df899dd3e65572d06e2bf2326fad692cc5e1c1e4ab002fdc6d7823743f |
|
||||
| 原始结构独立核验报告 | 5f0ef2948d0db68fe0cd89eb4c8b57a1088b94b53bed98f180d2b96629746e72 |
|
||||
|
||||
服务器恢复包分别为 `/data/paper-index/validation/uai2022-2025-mirror-metadata-recovery-v1.tar.gz` 与 `uai2022-2025-raw-reply-recovery-v1.tar.gz`。本地同名备份位于 `/Users/zzyang/Documents/Codex/2026-11-06/paper-index/`。先检查实际单元、完整manifest和哈希,再恢复;不要因进度文件旧状态或日志安静重跑。
|
||||
|
||||
后续继续寻找能明确覆盖作者公开拒稿选择的来源,保留每次投稿及决定原文标签、原始ID与镜像关系,核对跨来源重叠和后续录用。UAI缺口不关闭,全部43个目标的逐年公开负面决定核查也不缩减;最终全库去重仍为独立必做阶段。
|
||||
@@ -0,0 +1,67 @@
|
||||
# MM 2026 出版条目采集与恢复
|
||||
|
||||
`collect mm2026-publications` 是包内可复用的冻结快照规范化命令,读取完整注册目录和官方节目,不连接数据库、不请求网络、不下载PDF。总体仍为13/17,来源规范化、测试通过或节目匹配都不能替代身份批准、生产入库或最终去重。
|
||||
|
||||
## 使用
|
||||
|
||||
在已配置的服务器环境中,`--registry-snapshots` 是包含 `mm2026-container-registry-v2/` 的父目录;`--program-snapshots` 是包含 `mm2026-program-and-public-probes-v1/` 的父目录。父论文集正文使用独立固定快照,不能从不存在的文章ISBN字段猜测:
|
||||
|
||||
```sh
|
||||
paper-atlas collect mm2026-publications \
|
||||
--registry-snapshots /data/paper-index/collected \
|
||||
--program-snapshots /data/paper-index/collected \
|
||||
--parent-snapshot /data/paper-index/collected/mm2026-publication-metadata-v1/parent.json \
|
||||
--out /data/paper-index/collected/mm2026-publications-native-v1
|
||||
```
|
||||
|
||||
输出目录必须不存在。失败输出保留 `failed.json`,完整 `collection-report.json` 只在全部校验成功后原子生成;不要覆盖输出或因旧进度文件、日志安静重跑。先观察实际单元/报告,再选择新的恢复目录。
|
||||
|
||||
## 注册目录与节目分别保留
|
||||
|
||||
源目录基于 Crossref [官方精确过滤器](https://www.crossref.org/documentation/retrieve-metadata/rest-api/rest-api-filters/) 的 `container-title` 和出版者member320,实际捕获8页、1918个不同DOI。每页请求、时间、原始正文、字节哈希、总量与唯一DOI均核验。父DOI为 `10.1145/3767308`,题名为 *Proceedings of the 34th ACM International Conference on Multimedia*,明确事件为MM2026;文章均注册为 `proceedings-article`。
|
||||
|
||||
ISBN查询和父Unixref XML只返回父项,不能代表逐篇目录。宽泛 `query.container-title` 的检索总量也不能当会议篇数。当前1918是这个明确论文集过滤范围内的注册条目数,不是已经去重的科研作品数、所有MM workshop或全部28个来源track的出版全集。
|
||||
|
||||
每个出版条目都输出,稳定外部标识为真实DOI。身份标识仅为 `doi`,投稿侧标识仅为 `publication_doi`;不推断原始OpenReview/forum/投稿号或arXiv标识。来源名 `mm_registry_program` 明确表示派生的注册/节目材料,不能把其中所有字段都说成Crossref原文。
|
||||
|
||||
原始注册响应及父记录复制到 `evidence/registry/`,完整节目及四个既有403回执复制到 `evidence/program/`,不重新请求受限入口。每项元数据保留注册正文的文件/哈希、数组位置和规范JSON对象哈希;常用书目字段另作投影,完整引用、资助等原始字段仍在原始响应中,未丢弃也不在每条JSONL重复膨胀。
|
||||
|
||||
## 分轨、摘要与版本差异
|
||||
|
||||
按原始题名的Unicode/字母数字规范化与完整作者多重集合比对,1726项有唯一节目对应。源track映射到对应输出分区,并保留原始标签、节目ID、机构和所有展示上下文;不生成全局人物别名。其余122项作者版本差异、70项题名版本差异保留在 `proceedings_unclassified`,不强配分轨或静默排除。
|
||||
|
||||
| 输出分区 | 注册条目 |
|
||||
| --- | ---: |
|
||||
| main | 1533 |
|
||||
| grand_challenge | 77 |
|
||||
| interactive_art | 17 |
|
||||
| datasets | 40 |
|
||||
| reproducibility | 1 |
|
||||
| bni | 22 |
|
||||
| open_source | 11 |
|
||||
| doctoral_symposium | 2 |
|
||||
| demo_and_video | 20 |
|
||||
| industry_talk | 3 |
|
||||
| proceedings_unclassified | 192 |
|
||||
|
||||
注册摘要1918项全缺失。1717项有明确对应的节目摘要,逐字保留并标明 `abstract_source=official_program_exact_metadata_correspondence`;注册原始摘要仍标为缺失,绝不伪称由出版者提供。其余201项保持缺失,包括9项已对应节目本身没有摘要的条目及192个版本待核条目。没有用模糊匹配把待核摘要强加给论文。
|
||||
|
||||
8个作者仅有单一姓名字段,保留登记值并标记,不猜测given/family;37个含排版标记或实体的题名保留字面原文并标记,不能损失数学上下标或假定排版版本已经解决。
|
||||
|
||||
完整2143个节目ID和4326个展示位置仍在匹配证据或417个未匹配节目中,28个原始track均保留。417个节目与192个出版版本待核来自两种观察口径,不能直接相加当成不同论文。Workshop、艺术、行业等贡献不能因为不在这个论文集范围就被当成拒稿、撤稿或不值得保留。
|
||||
|
||||
注册出版支持来源侧 `status=accepted` 的发表观察,`raw_decision` 明确不是原始签署决定;出版日期仅放在注册元数据中,不伪造决定日期。`mm_registry_program:crossref:type` 与 `mm_registry_program:mm_program:track` 分别保存注册书目类型和节目 track;共同的派生来源前缀满足写入器的来源校验,两个标签命名空间仍相互独立,原始平台字段与证据继续保留在各自元数据中,`research_type_classified=false`:书目类型不是研究类型,行业、艺术、demo条目不能默认变成主轨研究论文。
|
||||
|
||||
## 只读数据库预检与导入边界
|
||||
|
||||
每个分区都有带count/SHA256的manifest,coverage始终为partial。规范化完成后,对实际报告中的全部输出文件运行现有 `audit source-identities --files ... --pg-service foundry-research --pg-schema research --out <新目录>`,不能靠已有66个arXiv/DOI候选代表全目录。预检使用只读一致性快照,并保存来源/标识/历次投稿/逐次决定的证据包;缺失或截断必须单列。
|
||||
|
||||
`approved_for_import=0`,每项 `scientific_work_identity_approved=false`。DOI绑定可能有误,精确题名/作者匹配不证明科学作品身份,没有匹配也不证明新作品。进入增量导入前仍须审查已有绑定、科学版本与候选历史,做新鲜保护检查、完整基线/备份,再使用原生写入者。不得覆盖旧拒稿、后续录用、标签、原始元数据或人物姓名版本;既有已消费队列不可重放。
|
||||
|
||||
## 恢复
|
||||
|
||||
完整注册目录及独立节目比对恢复包:`/data/paper-index/validation/mm2026-container-registry-recovery-v2.tar.gz`,SHA256 `ddb05a8134f6437b3fb40250e45ed6350d55431d953812abda3300871b6dd0ac`。其中含注册源目录、父记录 `validation/mm2026-parent.json`、逐项比对和请求方法证据。安全解包后可把该父记录路径传给 `--parent-snapshot`。
|
||||
|
||||
节目原始来源恢复包见 [MM来源说明](mm2026-source-status.md)。本地独立核验备份位于 `/Users/zzyang/Documents/Codex/2026-11-06/paper-index/`。固定回执/正文哈希变化、总量变化、未知注册上下文或节目对应改变都须重新核查,不能放宽为无条件接受。最终全库去重、近似题名作者、标识冲突、摘要、约束、覆盖与查询审计仍是独立必做阶段。
|
||||
|
||||
实际导入回归在隔离的 SQLite 和 PostgreSQL 测试库中运行全部1918条、全部11个输出分区,核验原始字段、1918个书目类型标签、1726个节目标签、publication_doi 身份、partial覆盖,以及已有作品的规范信息和旧拒稿决定不被覆盖。这些测试不会写入生产 research schema;测试通过仍不替代逐条生产身份审阅。
|
||||
@@ -90,3 +90,50 @@ paper-atlas collect mm2026-program \
|
||||
这些文件的格式为 `mm_program_observation_v1_not_ingestion_record`,所有项 `approved_for_import=false`。节目ID仅为 `mm_program:program_id` 来源定位符,track标签使用 `mm_program:track`;不建立科学作品、原始投稿ID或最终决定,不推断 DOI/arXiv/OpenReview 标识、论文类型、拒稿或撤稿。行业报告、艺术、workshop及其他来源贡献均保留,尚需逐轨核对贡献类型与正式出版依据。
|
||||
|
||||
本批尚未形成生产导入队列。后续做贡献类型、作品/出版身份及既往投稿审查,再生成经批准的增量队列。已消费的其他会议导入队列不可重放;最终全库去重和完整审计仍为独立必做阶段。
|
||||
|
||||
## 直接审计节目身份与既往历史
|
||||
|
||||
全量节目观察可以直接进入只读候选审计,无需伪造成录用论文队列:
|
||||
|
||||
```sh
|
||||
PGSERVICEFILE=/etc/paper-atlas/pg_service.conf \
|
||||
paper-atlas audit source-identities \
|
||||
--mm-program-report /data/paper-index/collected/mm2026-native-v1/collection-report.json \
|
||||
--pg-service foundry-research --pg-schema research \
|
||||
--out /data/paper-index/collected/mm2026-preflight-v1
|
||||
```
|
||||
|
||||
`--mm-program-report` 与常规 `--files` 互斥。审计先核对冻结原文、全部28轨的逐文件哈希/数量,以及每个节目ID的完整字段和展示上下文;缺轨、重复ID、篡改后的标识/决定或上下文均在数据库连接前拒绝。报告中记录的绝对文件路径仅作来源证据,实际读取报告同目录中的对应文件。
|
||||
|
||||
候选审计中的 `status=unknown` 仅为内部查询占位符,不代表原始提交的最终状态。节目ID以 `mm_program` 来源键查询,不转成全局 DOI、arXiv 或 OpenReview 标识;保留原始 `listed_in_official_program` 观察与全部展示位置。所有贡献类型、出版身份、最终决定核验标志均为false;缺作者的两项也保留并标记。
|
||||
|
||||
查询使用 PostgreSQL 只读一致性快照,候选作品的原始来源、标识、历次投稿和逐次决定压缩保存并记录哈希;截断会显式标记。标题/作者候选匹配不是作品身份批准,没有匹配也不证明是新论文。既往拒稿和后续录用不能互相覆盖,评分或评审推荐不能充当最终决定。输出目录必须不存在;遇到已有输出或静默运行,先检查具体单元和报告,不重启重放。
|
||||
|
||||
## 已完成的规范化、只读审计与出版核对
|
||||
|
||||
采集器提交 `c8db611` 和直接节目审计提交 `09a2d4e` 已通过PR #65、#66由用户合并。授权远程runner的535/543项测试、wheel构建安装和安装后全量命令检查通过;没有本机测试或构建。服务器规范化和只读候选报告也已独立验收,并在本机备份、复核全部2143项与814个候选证据包。
|
||||
|
||||
| 只读候选分类 | 来源节目项数 |
|
||||
| --- | ---: |
|
||||
| 标题与完整作者多重集合一致,仍需作品核查 | 660 |
|
||||
| 仅部分作者重叠 | 153 |
|
||||
| 无精确标题候选,不证明是新作品 | 1329 |
|
||||
| 有标题候选但无作者一致性 | 1 |
|
||||
|
||||
全部4326个展示上下文一致,814个证据包均无截断。24篇候选作品涉及24次已有负面投稿、37条负面来源观察;37不能直接算作37次拒稿。所有候选均尚未批准身份或入库,旧负面决定必须在之后关联时保留。
|
||||
|
||||
对已有作品的原始arXiv记录、标题/完整作者多重集合和MM2026语境进一步筛查,得到67个 Main Track 出版DOI核对候选。独立查询其明确给出的DOI及父论文集,共68次单次请求,67次HTTP200、1次404,没有认证或重试。父注册记录为[第34届ACM Multimedia论文集](https://doi.org/10.1145/3767308),ISBN `9798400722134`,出版者ACM,出版年2026。
|
||||
|
||||
其中66项的原始arXiv引用DOI与注册DOI一致,按既有审计规则规范化的标题、完整作者多重集合、出版年、出版者、明确会议名和论文集标题也一致。注册文章未提供ISBN,不能伪称逐篇ISBN已核验;归属依据是文章明确给出的会议和论文集字段,以及独立的原始arXiv语境和DOI。仍需逐项作品/版本审查及新鲜数据库保护检查,**不是66项已入库或最终身份批准**。
|
||||
|
||||
`mfp7067` 对应既有候选作品2749947,原始arXiv提供的 `10.1145/3767308.3836169` 返回404,暂缓。404不是撤稿/拒稿证据,也不据此删除或修正既有标识;需独立核对后再处理。
|
||||
|
||||
完整会议出版名单仍需继续核对。实际ISBN查询只返回父论文集一个条目,父DOI的公开Unixref XML也只含该父DOI、0个 `conference_paper` 节点;都不能代表逐篇名单。按会议名检索得到的 `total-results` 是出版者全年范围内的宽泛检索量,不能当作MM篇数;必须逐项核对明确会议/论文集字段、节目与分轨归属,并保留未匹配项。全部28个来源track继续保留,不能把本批Main Track候选当成会议完整覆盖。
|
||||
|
||||
| 已独立验证的恢复证据 | SHA256 |
|
||||
| --- | --- |
|
||||
| 全轨规范化恢复包 | b119081618b8ce98037dc139be159512f9fdb86a0ccfb127c8b8a1c5722f3d87 |
|
||||
| 只读候选/814证据包恢复包 | 42d0d0404bba04e64e638344b050ad4e7287c3e4c75a777b7069ecc7d7a6ed60 |
|
||||
| 67项DOI及父论文集原始注册回执恢复包 | dd7a0ff76f12070f403cc7352ac54269e454758d807896a086c58c1c1ac7b798 |
|
||||
|
||||
服务器恢复包位于 `/data/paper-index/validation/`,分别名为 `mm2026-native-recovery-v1.tar.gz`、`mm2026-preflight-recovery-v1.tar.gz`、`mm2026-publication-metadata-recovery-v1.tar.gz`;本地同名备份在 `/Users/zzyang/Documents/Codex/2026-11-06/paper-index/`。来源目录为 `collected/mm2026-native-v1`、`mm2026-preflight-v1`、`mm2026-publication-metadata-v1`。先核对完整manifest及实际单元结果,不能用旧进度文件的running字样重启。其他会议已消费的导入队列仍禁止重放;本轮均为来源核查、只读候选审计,没有生产写入。
|
||||
|
||||
@@ -0,0 +1,52 @@
|
||||
# ReviewArena 镜像的索引字段投影
|
||||
|
||||
ReviewArena 是公开评审镜像,固定版本为 `c2978add17c2099219eaddbc2599974d69d4d09b`。上游数据卡描述七类会议、2020–2026 年的部分来源;这不是逐会议、逐年度的公开决定全集。文件清单有 15 个 Parquet 分片,覆盖 COLM、CoRL、EMNLP、ICLR、ICML、NeurIPS、TMLR。这里不包括 UAI,不能用它填补 UAI 缺口。
|
||||
|
||||
- [固定版本上游数据卡](https://huggingface.co/datasets/anonymousNeurIPS2026submission4281/reviewarena/blob/c2978add17c2099219eaddbc2599974d69d4d09b/README.md)
|
||||
- [固定版本文件元数据 API](https://huggingface.co/api/datasets/anonymousNeurIPS2026submission4281/reviewarena/tree/c2978add17c2099219eaddbc2599974d69d4d09b/data?recursive=false&expand=false)
|
||||
- [Arrow ParquetFile 投影接口](https://arrow.apache.org/docs/python/generated/pyarrow.parquet.ParquetFile.html)
|
||||
|
||||
## 运行
|
||||
|
||||
仅在已授权远程 runner 安装 `paper-atlas[mirrors]`,使用固定 PyArrow 26.0.1 二进制 wheel。本机不执行安装、构建或项目测试。常规 CI 包含投影、安全边界和镜像决定语义测试;真实网络采集由独立的 `ReviewArena metadata projection` workflow 手动触发,指定已通过 CI 的代码 ref 和 split,例如 `emnlp`。该 workflow 仍使用 `ces-control` / `aliyun-ces-control`,在服务器固定数据目录生成输出和恢复归档,不更改默认 runtime 或数据库。
|
||||
|
||||
包内命令如下,`--out` 必须是既有父目录内的新目录:
|
||||
|
||||
```sh
|
||||
paper-atlas collect reviewarena-metadata \
|
||||
--split emnlp \
|
||||
--out /data/paper-index/collected/reviewarena-emnlp-reviewed-version
|
||||
```
|
||||
|
||||
输出目录已存在时拒绝执行。遇到禁止访问、来源变化或读取失败,保留失败报告,不自动重试或切换到整文件下载。不能把原始访问拒绝解释为没有拒稿。
|
||||
|
||||
已保存的完整元数据投影可以离线重新解释决定字面值。提供原 `collection-report.json` 及其已独立验证的精确文件 SHA-256,程序核对全部元数据文件哈希、逐行来源身份和原始字段,生成新目录;不会再次访问上游 Parquet:
|
||||
|
||||
```sh
|
||||
paper-atlas collect reviewarena-metadata --split emnlp \
|
||||
--from-projection /data/paper-index/collected/<既有投影>/collection-report.json \
|
||||
--projection-sha256 <已验证的原报告SHA256> \
|
||||
--out /data/paper-index/collected/<新解释版本>
|
||||
```
|
||||
|
||||
手动 workflow 对应输入为 `projection_name` 与 `projection_sha256`。离线报告的 `metadata_bytes_read` 为本次请求字节数 `0`,`source_projection_bytes` 保留原始列块投影字节数;原报告哈希建立恢复链,原始记录与来源定位不变。
|
||||
|
||||
## 实际读取和保存范围
|
||||
|
||||
仅投影以下字段:`forum_id`、`conference`、`year`、`track`、`venue_id`、`paper_number`、`title`、`abstract`、`authors`、`keywords`、`venue`、`decision`。
|
||||
|
||||
程序先读取文件魔数、尾部长度和有界 Parquet 页脚,在内存中解析 schema 和列块偏移,然后只请求这些列的字典/数据页。Arrow 可能请求较大的页脚窗口,未获准区域由本地稀疏文件合成零字节,不会为该窗口补下载相邻正文页。逐个选中列块还要检查不与未选列块重叠。单文件上限为 64 MiB 所需字节、512 个范围请求和 100,000 行;响应必须是准确的 HTTP 206 / Content-Range,ETag 变化时停止。该边界适用于每个分片,不能据此启动其他分片的未授权全文件下载。
|
||||
|
||||
不读取 `markdown`、`reviews_json`、`decision_comment`、`author_rebuttal` 等正文或评审内容的数据页;不保存原 Parquet、页脚原文或未选字段统计值。保存的 JSONL 只有上述索引字段、来源定位和审计标记。报告记录每个请求范围及其 SHA-256、选中列名、页脚哈希、投影数量和保存文件哈希,便于独立复核。
|
||||
|
||||
上游 LFS SHA-256 与大小是固定版本 API 提供的对象声明。由于不下载整个 Parquet,本流程**不声称验证了完整 Parquet 内容哈希**。范围对象大小/ETag 和各段哈希用于检查本次读取的一致性。
|
||||
|
||||
## 决定和身份边界
|
||||
|
||||
`mirror_decision_category` 只解释明确的镜像字面值;未知、缺失或未审核的决定保持 `unknown`,不会根据 venue、是否录用、摘要缺失或目录缺席推断拒稿。类别数量是镜像行数,不能解释为独立作品数量或原始官方决定数量。
|
||||
|
||||
真实 EMNLP 2023 固定分片包含 2,009 行:原始决定为 `Accept-Main` 994 行、`Accept-Findings` 1,015 行,原始 `track` 却全部为 `main`。两种已核对字面值属于镜像录用声明;`mirror_decision_track_claim` 分开保留其明确的 main/findings 声明,原 `track` 和所有 `raw_fields` 不被重写。`Accept (Poster)` 等展示形式不能自行推出主会归属。此分片没有明确负面决定,不能用于补齐拒稿缺口,更不能由此断言 EMNLP 原始平台没有公开拒稿。首次实际投影仅请求 1,720,498 字节;原文件 90,794,691 字节,未下载完整文件。
|
||||
|
||||
上游 `forum_id`、`venue_id`、投稿号、作者和决定均作为原始声明保留。`forum_id` 不直接成为全局 OpenReview 标识;稳定行定位由数据集、固定 commit、文件和行号组成。输出 `identifiers` 为空,原始论坛身份和决定签名标记为未验证,保留匿名和缺失作者。源数据中的字符串命名不能被当成真实人员别名。
|
||||
|
||||
输出故意不是可直接 `import records` 的规范化队列,没有授予导入或合并权限。下一步必须检查原始来源、跨来源重合和作品身份,再决定哪些来源观察能进入生产。即使明确镜像拒稿,也要保留该投稿历史并继续查找后续其他会议的录用;不能把“目前没有关联录用”表述为从未录用。本采集不代替最终全库去重、决定冲突、来源覆盖和查询验收。
|
||||
@@ -74,6 +74,74 @@ def load_inputs(files):
|
||||
return records, inputs
|
||||
|
||||
|
||||
def load_mm_program_inputs(report_path):
|
||||
"""Read every MM observation without turning a program into import records.
|
||||
|
||||
Reconcile the complete output against its frozen raw source before any DB
|
||||
connection. The synthetic unknown status is an audit placeholder only.
|
||||
"""
|
||||
from paper_atlas.collectors import mm2026
|
||||
|
||||
report_path = Path(report_path).absolute()
|
||||
raw_report = mm2026.safe_read(report_path, 256*1024)
|
||||
report = json.loads(raw_report)
|
||||
if (report.get('source') != 'mm_program' or report.get('records') != 2143
|
||||
or report.get('program_observations') != 4326 or len(report.get('files', [])) != 28
|
||||
or report.get('source_body_sha256') != mm2026.BODY_SHA
|
||||
or report.get('source_manifest_sha256') != mm2026.MANIFEST_SHA
|
||||
or report.get('database_written') is not False or report.get('approved_for_import') != 0):
|
||||
raise ValueError('Unreviewed MM observation report')
|
||||
root = report_path.parent
|
||||
raw_source = mm2026.safe_read(root/'evidence/program-data.json', 12*1024**2)
|
||||
raw_receipts = mm2026.safe_read(root/'evidence/manifest.json', 128*1024)
|
||||
if digest(raw_source) != mm2026.BODY_SHA or digest(raw_receipts) != mm2026.MANIFEST_SHA:
|
||||
raise ValueError('MM raw evidence changed')
|
||||
expected_rows, _ = mm2026.parse_program(raw_source)
|
||||
expected = {r['paper_id']:r for r in expected_rows}
|
||||
records, inputs, seen, tracks = [], [], set(), set()
|
||||
for item in report['files']:
|
||||
track = item.get('track_literal')
|
||||
if (track not in mm2026.TRACK_COUNTS or track in tracks
|
||||
or item.get('format') != 'mm_program_observation_v1_not_ingestion_record'
|
||||
or item.get('approved_for_import') != 0):
|
||||
raise ValueError('Duplicate or unreviewed MM track format')
|
||||
tracks.add(track)
|
||||
# Stored absolute paths are provenance, not instructions to read elsewhere.
|
||||
path = root/Path(item['file']).name
|
||||
if path.suffix != '.jsonl':
|
||||
raise ValueError('Invalid MM observation filename')
|
||||
raw = mm2026.safe_read(path, 16*1024**2)
|
||||
lines = raw.splitlines(keepends=True)
|
||||
if digest(raw) != item.get('file_sha256') or len(lines) != item.get('count') or len(lines) != mm2026.TRACK_COUNTS[track]:
|
||||
raise ValueError('MM observation file count/hash mismatch')
|
||||
for number,line in enumerate(lines, 1):
|
||||
row = json.loads(line)
|
||||
pid = row.get('paper_id')
|
||||
if not isinstance(pid,str) or pid in seen or row != expected.get(pid) or row['track'] != track:
|
||||
raise ValueError('MM observation differs from frozen source or repeats an ID')
|
||||
seen.add(pid)
|
||||
records.append({'source':'mm_program','external_id':'2026/'+pid,
|
||||
'title':row['title'],'authors':row['authors'],'venue':'MM','year':2026,'track':track,
|
||||
'status':'unknown','identifiers':{},'file':str(path),'line':number,
|
||||
'raw_line_sha256':digest(line),'source_record_hits':[],
|
||||
'identifier_hits':[],'title_candidate_ids':[],
|
||||
'input_format':'mm_program_observation_v1_not_ingestion_record',
|
||||
'program_id':pid,'program_contexts':row['contexts'],
|
||||
'source_observed_status':row['observed_status'],
|
||||
'missing_source_fields':row['missing_fields'],
|
||||
'publication_identity_verified':False,'final_decision_verified':False,
|
||||
'contribution_type_verified':False})
|
||||
inputs.append({'file':str(path),'file_sha256':digest(raw),'count':len(lines),
|
||||
'manifest_sha256':digest(raw_report),'manifest':{
|
||||
'source':'mm_program','venue':'MM','year':2026,'track':track,'status':'unknown',
|
||||
'format':'mm_program_observation_v1_not_ingestion_record',
|
||||
'scope':'Read-only lookup input only; unknown is not a source decision.',
|
||||
'source_report':str(report_path),'source_body_sha256':digest(raw_source)}})
|
||||
if seen != expected.keys() or tracks != mm2026.TRACK_COUNTS.keys():
|
||||
raise ValueError('Incomplete MM observation input; all28tracks required')
|
||||
return records, inputs
|
||||
|
||||
|
||||
def _lookup(connection, values, columns, lateral):
|
||||
"""Bound scalar VALUES allow the existing text codec to encode each key."""
|
||||
width = len(columns)
|
||||
@@ -187,9 +255,11 @@ def _atomic_json(path, value):
|
||||
temporary.replace(path)
|
||||
|
||||
|
||||
def audit(files, out, *, pg_service, pg_schema):
|
||||
def audit(files, out, *, pg_service, pg_schema, mm_program_report=None):
|
||||
# Reject input corruption before opening a database connection or reserving output.
|
||||
records, inputs = load_inputs(files)
|
||||
if bool(files) == bool(mm_program_report):
|
||||
raise ValueError('Choose source files or one MM program report')
|
||||
records, inputs = load_mm_program_inputs(mm_program_report) if mm_program_report else load_inputs(files)
|
||||
out = Path(out)
|
||||
out.mkdir(parents=True, exist_ok=False)
|
||||
(out / 'audit.lock').write_text(timestamp() + '\n')
|
||||
@@ -229,6 +299,7 @@ def audit(files, out, *, pg_service, pg_schema):
|
||||
'Name agreement does not prove person or scientific-work identity.',
|
||||
'Bound identifiers can be incorrect; review titles, authors and original evidence.',
|
||||
'Source/history limits are explicit in compressed entity packets.',
|
||||
'Program observations and unknown audit status do not establish contribution type, publication or a final decision.',
|
||||
'Not a final whole-database deduplication or coverage audit.']}
|
||||
_atomic_json(out / 'postgresql-preflight.json', report)
|
||||
return report
|
||||
@@ -243,10 +314,14 @@ def audit(files, out, *, pg_service, pg_schema):
|
||||
|
||||
def main(argv=None, *, prog=None):
|
||||
parser = argparse.ArgumentParser(prog=prog, description=__doc__)
|
||||
parser.add_argument('--files', nargs='+', required=True, type=Path)
|
||||
inputs = parser.add_mutually_exclusive_group(required=True)
|
||||
inputs.add_argument('--files', nargs='+', type=Path)
|
||||
inputs.add_argument('--mm-program-report', type=Path,
|
||||
help='Read all frozen MM program observations directly; never generate an import queue')
|
||||
parser.add_argument('--pg-service', required=True)
|
||||
parser.add_argument('--pg-schema', required=True)
|
||||
parser.add_argument('--out', required=True, type=Path)
|
||||
args = parser.parse_args(argv)
|
||||
report = audit(args.files, args.out, pg_service=args.pg_service, pg_schema=args.pg_schema)
|
||||
report = audit(args.files, args.out, pg_service=args.pg_service, pg_schema=args.pg_schema,
|
||||
mm_program_report=args.mm_program_report)
|
||||
print(json.dumps({k: report[k] for k in ('status', 'records', 'kind_counts', 'approved_for_import')}, ensure_ascii=False))
|
||||
|
||||
@@ -20,6 +20,7 @@ COMMANDS = {
|
||||
('collect','public-openreview'): ('paper_atlas.collectors.public_openreview', {}, '核验其他会议的公开拒稿及撤稿标签'),
|
||||
('collect','icml-public-archive'): ('paper_atlas.collectors.icml_public_archive', {}, 'ICML 2026 公开拒稿原始投稿快照'),
|
||||
('collect','peerread'): ('paper_atlas.collectors.peerread', {}, 'PeerRead 历史决定补充'),
|
||||
('collect','reviewarena-metadata'): ('paper_atlas.collectors.reviewarena', {}, '仅投影固定版本评审镜像的索引字段,保留决定声明而不下载正文或评审页'),
|
||||
('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'),
|
||||
@@ -33,6 +34,7 @@ COMMANDS = {
|
||||
('collect','rss2026-proceedings'): ('paper_atlas.collectors.rss2026', {}, '核验 RSS 2026 官方论文集与独立录用表'),
|
||||
('collect','recsys2026-program'): ('paper_atlas.collectors.recsys2026', {}, '核验 RecSys 2026 公开录用贡献并保留 DOI 来源版本'),
|
||||
('collect','mm2026-program'): ('paper_atlas.collectors.mm2026', {}, '恢复 MM 2026 全部来源 track 与重复展示上下文,待贡献和身份核查'),
|
||||
('collect','mm2026-publications'): ('paper_atlas.collectors.mm_publications2026', {}, '核验 MM 2026 全部注册出版条目并分别保留节目分轨、摘要与版本差异'),
|
||||
('collect','cvf'): ('paper_atlas.collectors.vision', {'source':'cvf'}, 'CVPR / ICCV 官方开放目录'),
|
||||
('collect','ecva'): ('paper_atlas.collectors.vision', {'source':'ecva'}, 'ECCV 官方开放目录'),
|
||||
('collect','aamas-proceedings'): ('paper_atlas.collectors.aamas', {}, '按官方目录区分 AAMAS 发表形式和摘要'),
|
||||
|
||||
@@ -0,0 +1,227 @@
|
||||
"""Frozen MM 2026 registered publications with separately attributed program evidence.
|
||||
|
||||
All publications survive normalization. A program metadata match is not a
|
||||
scientific-work identity decision; publisher article type is not research type.
|
||||
"""
|
||||
import argparse
|
||||
from collections import Counter, defaultdict
|
||||
from datetime import datetime
|
||||
import hashlib
|
||||
import json
|
||||
from pathlib import Path
|
||||
import re
|
||||
import unicodedata
|
||||
from urllib.parse import parse_qs, urlsplit
|
||||
|
||||
from paper_atlas.collectors import mm2026 as program
|
||||
|
||||
SOURCE = 'mm_registry_program'
|
||||
SNAPSHOT = 'mm2026-container-registry-v2'
|
||||
MANIFEST_SHA = '128b9d02c72261a66edb9a4537d118bdccab3ed80d49a1e971fefdac399f9855'
|
||||
PARENT_SHA = 'bc3ce7da2914f9ddd198c928b3acb2734a8370add3348c23c5f9f12135381628'
|
||||
PARENT_DOI = '10.1145/3767308'
|
||||
CONTAINER = 'Proceedings of the 34th ACM International Conference on Multimedia'
|
||||
EVENT = "MM '26: The 34th ACM International Conference on Multimedia"
|
||||
TRACKS = {'Main Track':'main','Grand Challenge':'grand_challenge',
|
||||
'Interactive Art':'interactive_art','Datasets':'datasets','Reproducibility':'reproducibility',
|
||||
'BNI':'bni','Open Source':'open_source','Doctoral Symposium':'doctoral_symposium',
|
||||
'Demo and Video':'demo_and_video','Industry Talk':'industry_talk'}
|
||||
BIBLIOGRAPHY = {'DOI','URL','title','original-title','subtitle','short-title','author',
|
||||
'publisher','member','type','container-title','short-container-title','event','abstract',
|
||||
'page','ISBN','isbn-type','published','published-print','published-online','issued',
|
||||
'license','resource','relation','alternative-id'}
|
||||
|
||||
|
||||
def digest(raw):
|
||||
return hashlib.sha256(raw).hexdigest()
|
||||
|
||||
|
||||
def key(text):
|
||||
return ''.join(c for c in unicodedata.normalize('NFKD',text).casefold() if c.isalnum())
|
||||
|
||||
|
||||
def author_names(values):
|
||||
return Counter(key(v) for v in values)
|
||||
|
||||
|
||||
def load_registry(root, parent_path):
|
||||
directory = Path(root)/SNAPSHOT
|
||||
raw_manifest = program.safe_read(directory/'manifest.json', 128*1024)
|
||||
raw_parent = program.safe_read(Path(parent_path), 128*1024)
|
||||
if digest(raw_manifest) != MANIFEST_SHA or digest(raw_parent) != PARENT_SHA:
|
||||
raise ValueError('Reviewed registry or parent hash changed')
|
||||
manifest = json.loads(raw_manifest);parent = json.loads(raw_parent)['message']
|
||||
if (manifest['status'] != 'complete_exact_container_registry_capture_not_work_or_track_approval'
|
||||
or manifest['unique_registry_DOIs'] != 1918 or manifest['expected_registry_items'] != 1918
|
||||
or len(manifest['pages']) != 8 or manifest['database_written'] is not False
|
||||
or manifest['approved_for_import'] != 0 or manifest['parent_body_sha256'] != PARENT_SHA
|
||||
or parent['DOI'] != PARENT_DOI or parent['type'] != 'proceedings'
|
||||
or parent['title'] != [CONTAINER] or parent['event']['name'] != EVENT
|
||||
or parent['member'] != '320' or parent['ISBN'] != ['9798400722134']):
|
||||
raise ValueError('Incomplete reviewed registry or wrong proceedings')
|
||||
evidence = {'manifest.json':raw_manifest,'parent.json':raw_parent};rows=[];seen=set()
|
||||
for number,receipt in enumerate(manifest['pages'],1):
|
||||
name=receipt['file']
|
||||
if name != f'page-{number:03}.json':
|
||||
raise ValueError('Unexpected registry page filename')
|
||||
body=program.safe_read(directory/name,8*1024**2)
|
||||
url=urlsplit(receipt['url']);query=parse_qs(url.query)
|
||||
expected_filter='container-title:'+CONTAINER+',member:320'
|
||||
if (receipt['http_status'] != 200 or receipt['url'] != receipt['final_url']
|
||||
or url.scheme != 'https' or url.netloc != 'api.crossref.org' or url.path != '/works'
|
||||
or set(query) != {'filter','rows','cursor'} or query['filter'] != [expected_filter]
|
||||
or query['rows'] != ['250'] or not query['cursor'][0]
|
||||
or datetime.fromisoformat(receipt['at']).utcoffset() is None
|
||||
or receipt['bytes'] != len(body) or receipt['sha256'] != digest(body)):
|
||||
raise ValueError('Registry receipt and source body disagree')
|
||||
obj=json.loads(body);message=obj['message'];expected=250 if number<8 else 168
|
||||
if (obj['status'] != 'ok' or message['total-results'] != 1918
|
||||
or len(message['items']) != expected or receipt['items'] != expected):
|
||||
raise ValueError('Incomplete registry pagination')
|
||||
evidence[name]=body
|
||||
for position,original in enumerate(message['items']):
|
||||
doi=original.get('DOI')
|
||||
if not isinstance(doi,str) or not re.fullmatch(r'10\.1145/3767308\.[0-9]+',doi) or doi in seen:
|
||||
raise ValueError('Missing, foreign or repeated publication DOI')
|
||||
seen.add(doi)
|
||||
if (original.get('URL') != 'https://doi.org/'+doi or original.get('type') != 'proceedings-article'
|
||||
or original.get('container-title') != [CONTAINER] or original.get('event',{}).get('name') != EVENT
|
||||
or original.get('member') != '320' or original.get('publisher') != 'ACM'
|
||||
or original.get('published') != {'date-parts':[[2026,11,9]]} or original.get('abstract')):
|
||||
raise ValueError('Wrong registered publication context or changed abstract availability')
|
||||
titles=original.get('title');authors=original.get('author')
|
||||
if (not isinstance(titles,list) or len(titles)!=1 or not isinstance(titles[0],str) or not titles[0].strip()
|
||||
or not isinstance(authors,list) or not authors):
|
||||
raise ValueError('Missing publication title or authors')
|
||||
names=[];flags=[]
|
||||
for i,author in enumerate(authors,1):
|
||||
family=author.get('family');given=author.get('given','')
|
||||
if not isinstance(family,str) or not family.strip() or not isinstance(given,str):
|
||||
raise ValueError('Invalid registered author fields')
|
||||
name=' '.join(x.strip() for x in (given,family) if x.strip())
|
||||
if '@' in name or name.casefold()=='anonymous':
|
||||
raise ValueError('Unreviewed author identity')
|
||||
names.append(name)
|
||||
if not given.strip():flags.append({'kind':'single_registered_name_field_not_split','author_position':i})
|
||||
if re.search(r'<[^>]+>|&[A-Za-z#0-9]+;',titles[0]):
|
||||
flags.append({'kind':'literal_title_markup_or_entity_retained_requires_review'})
|
||||
canonical=json.dumps(original,ensure_ascii=False,sort_keys=True,separators=(',',':')).encode()
|
||||
rows.append({'doi':doi,'title':titles[0],'authors':names,'flags':flags,
|
||||
'fields':{k:v for k,v in original.items() if k in BIBLIOGRAPHY},
|
||||
'original_field_names':sorted(original),'original_canonical_json_sha256':digest(canonical),
|
||||
'receipt':receipt,'source_position':position})
|
||||
if len(rows)!=1918:
|
||||
raise ValueError('Incomplete registered publication census')
|
||||
return evidence,rows
|
||||
|
||||
|
||||
def correlate(rows, observations):
|
||||
titles=defaultdict(list)
|
||||
for row in observations:titles[key(row['title'])].append(row)
|
||||
matched=set();kinds=Counter()
|
||||
for row in rows:
|
||||
candidates=titles[key(row['title'])]
|
||||
exact=[p for p in candidates if author_names(p['authors'])==author_names(row['authors'])]
|
||||
if len(exact)==1:
|
||||
row['kind']='unique_program_title_fullauthor_agreement_not_identity_approval'
|
||||
row['program']=exact[0];matched.add(exact[0]['paper_id'])
|
||||
if exact[0]['track'] not in TRACKS:
|
||||
raise ValueError('New registered program track requires review')
|
||||
elif len(exact)>1:row['kind']='multiple_program_ids_same_title_authors_hold'
|
||||
elif candidates:row['kind']='program_author_version_requires_review'
|
||||
else:row['kind']='no_exact_program_title_requires_review'
|
||||
row['program_title_candidates']=[p['paper_id'] for p in candidates]
|
||||
row['program_exact_candidates']=[p['paper_id'] for p in exact]
|
||||
kinds[row['kind']]+=1
|
||||
expected={'unique_program_title_fullauthor_agreement_not_identity_approval':1726,
|
||||
'program_author_version_requires_review':122,'no_exact_program_title_requires_review':70}
|
||||
if kinds != expected:
|
||||
raise ValueError('Reviewed program/publication correspondence changed')
|
||||
return dict(kinds),[p for p in observations if p['paper_id'] not in matched]
|
||||
|
||||
|
||||
def collect(out, *, registry_snapshots, program_snapshots, parent_snapshot):
|
||||
out=Path(out)
|
||||
if out.exists() or out.is_symlink():
|
||||
raise ValueError('Output exists; inspect prior operation, never overwrite')
|
||||
if any(p.is_symlink() for p in out.parents):
|
||||
raise ValueError('Unsafe output parent')
|
||||
out.mkdir(parents=True)
|
||||
try:
|
||||
registry_evidence,publications=load_registry(registry_snapshots,parent_snapshot)
|
||||
program_evidence=program.load_snapshot(program_snapshots)
|
||||
observations,_=program.parse_program(program_evidence['program-data.json'])
|
||||
kinds,unmatched=correlate(publications,observations)
|
||||
for group,files in [('registry',registry_evidence),('program',program_evidence)]:
|
||||
directory=out/'evidence'/group;directory.mkdir(parents=True)
|
||||
for name,body in files.items():(directory/name).write_bytes(body)
|
||||
buckets=defaultdict(list);missing=0;copied=0;known_tracks=Counter()
|
||||
for item in sorted(publications,key=lambda r:r['doi']):
|
||||
p=item.get('program');doi=item['doi'];receipt=item['receipt']
|
||||
track=TRACKS[p['track']] if p else 'proceedings_unclassified'
|
||||
abstract=p['abstract'] if p and p['abstract'] else None
|
||||
missing+=abstract is None;copied+=abstract is not None
|
||||
if p:known_tracks[p['track']]+=1
|
||||
tags=[{'namespace':SOURCE+':crossref:type','value':'proceedings-article'}]
|
||||
if p:tags.append({'namespace':SOURCE+':mm_program:track','value':p['track']})
|
||||
buckets[track].append({'source':SOURCE,'external_id':doi,'url':'https://doi.org/'+doi,
|
||||
'title':item['title'],'authors':item['authors'],'abstract':abstract,
|
||||
'identifiers':{'doi':doi},'submission_identifiers':[{'scheme':'publication_doi','value':doi}],
|
||||
'venue':'MM','year':2026,'track':track,'status':'accepted',
|
||||
'raw_decision':'Registered publication in MM2026 proceedings; not an original signed decision.',
|
||||
'evidence_url':receipt['url'],'observed_at':receipt['at'],'tags':tags,
|
||||
'metadata':{'registry_fields':item['fields'],'registry_original_field_names':item['original_field_names'],
|
||||
'registry_source_evidence':{'file':'evidence/registry/'+receipt['file'],'sha256':receipt['sha256'],
|
||||
'url':receipt['url'],'json_pointer':'/message/items/'+str(item['source_position']),
|
||||
'canonical_object_json_sha256':item['original_canonical_json_sha256']},
|
||||
'registry_manifest':{'file':'evidence/registry/manifest.json','sha256':MANIFEST_SHA},
|
||||
'parent_evidence':{'file':'evidence/registry/parent.json','sha256':PARENT_SHA},
|
||||
'program_source_evidence':{'file':'evidence/program/program-data.json','sha256':program.BODY_SHA},
|
||||
'program_fields':p,'program_match_kind':item['kind'],
|
||||
'program_title_candidates':item['program_title_candidates'],
|
||||
'program_exact_candidates':item['program_exact_candidates'],
|
||||
'abstract_source':'official_program_exact_metadata_correspondence' if abstract is not None else 'not_available_from_reviewed_sources',
|
||||
'registry_abstract_available':False,'original_decision_signature_verified':False,
|
||||
'scientific_work_identity_approved':False,'research_type_classified':False,
|
||||
'review_flags':item['flags'],
|
||||
'scope':'Derived publication/program metadata with separate original evidence. All registered articles retained, including art/industry/demo; bibliographic type is not research type. Program correlation does not approve work identity or person aliases. Publication dates are not decision dates. DOI is publication evidence, not original OpenReview ID.'}})
|
||||
outputs=[]
|
||||
for track,rows in sorted(buckets.items()):
|
||||
path=out/f'mm-publications-2026-{track}.jsonl'
|
||||
raw=''.join(json.dumps(row,ensure_ascii=False)+'\n' for row in rows).encode();path.write_bytes(raw)
|
||||
manifest={'source':SOURCE,'parser_version':1,'venue':'MM','year':2026,'track':track,'count':len(rows),
|
||||
'file':str(path),'file_sha256':digest(raw),'list_url':'https://doi.org/'+PARENT_DOI,
|
||||
'status':'partial','missing_abstract_count':sum(row['abstract'] is None for row in rows),
|
||||
'scope':'Registered container partition only; not allMMprogram/workshops, independent works or public-negative coverage.'}
|
||||
path.with_suffix('.manifest.json').write_text(json.dumps(manifest,ensure_ascii=False,indent=2)+'\n');outputs.append(manifest)
|
||||
report={'status':'collected_publications_pending_work_identity_review','records':1918,'unique_registered_dois':1918,
|
||||
'files':outputs,'program_ids':2143,'program_observations':4326,'program_match_counts':kinds,
|
||||
'matched_program_track_counts':dict(known_tracks),'unmatched_program':unmatched,
|
||||
'unmatched_program_count':len(unmatched),'registry_missing_abstract_count':1918,
|
||||
'abstracts_from_exact_program_correspondence':copied,'output_missing_abstract_count':missing,
|
||||
'registered_single_name_fields':sum(f['kind']=='single_registered_name_field_not_split' for r in publications for f in r['flags']),
|
||||
'registered_title_markup_flags':sum(any(f['kind']=='literal_title_markup_or_entity_retained_requires_review' for f in r['flags']) for r in publications),
|
||||
'database_written':False,'approved_for_import':0,'network_requests':0,'pdfs_downloaded':False,
|
||||
'public_negative_availability':'unknown_no_negative_decisions_inferred',
|
||||
'scope':'1918registeredpublication DOIs, not1918proven distinct scientificworks.192title/authorvariants remainunclassified,417unmatchedprogram items retained. All28programtracks remain independent census; article type never implies research. Per-work/source/DOI/history review andfinalwholeDBaudit remain.'}
|
||||
if (sum(f['count'] for f in outputs)!=1918 or copied!=1717 or missing!=201 or len(unmatched)!=417):
|
||||
raise ValueError('Reviewed output partition or abstract census changed')
|
||||
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('--registry-snapshots',required=True)
|
||||
parser.add_argument('--program-snapshots',required=True)
|
||||
parser.add_argument('--parent-snapshot',required=True)
|
||||
parser.add_argument('--out',required=True)
|
||||
args=parser.parse_args(argv)
|
||||
report=collect(args.out,registry_snapshots=args.registry_snapshots,
|
||||
program_snapshots=args.program_snapshots,parent_snapshot=args.parent_snapshot)
|
||||
print(json.dumps({k:report[k] for k in ('status','records','files','program_match_counts',
|
||||
'abstracts_from_exact_program_correspondence','output_missing_abstract_count',
|
||||
'unmatched_program_count','database_written','approved_for_import')},ensure_ascii=False))
|
||||
@@ -0,0 +1,216 @@
|
||||
"""Source-attributed mirror metadata; no paper/review text or identity promotion."""
|
||||
import argparse
|
||||
from collections import Counter
|
||||
from datetime import datetime, timezone
|
||||
import hashlib
|
||||
from importlib.resources import files
|
||||
import json
|
||||
from pathlib import Path
|
||||
import re
|
||||
|
||||
from paper_atlas.io.parquet_metadata import RangeSource, projection
|
||||
|
||||
DATASET = 'anonymousNeurIPS2026submission4281/reviewarena'
|
||||
COMMIT = 'c2978add17c2099219eaddbc2599974d69d4d09b'
|
||||
TREE_SHA256 = 'bbc75f5774b019c4a9ee3efe5c0f37b14761bc3c5a054570db1df8f5603eac20'
|
||||
SOURCE = 'reviewarena_mirror'
|
||||
SPLITS = ('colm', 'corl', 'emnlp', 'iclr', 'icml', 'neurips', 'tmlr')
|
||||
COLUMNS = {'forum_id': 'string', 'conference': 'string', 'year': 'integer',
|
||||
'track': 'string', 'venue_id': 'string', 'paper_number': 'integer',
|
||||
'title': 'string', 'abstract': 'string', 'authors': 'strings',
|
||||
'keywords': 'strings', 'venue': 'string', 'decision': 'string'}
|
||||
|
||||
|
||||
def inventory(split):
|
||||
if split not in SPLITS:
|
||||
raise ValueError('Explicit supported mirror split required')
|
||||
raw = files('paper_atlas.resources').joinpath('reviewarena-files.json').read_bytes()
|
||||
if hashlib.sha256(raw).hexdigest() != TREE_SHA256:
|
||||
raise ValueError('Pinned mirror file inventory changed')
|
||||
items = json.loads(raw)
|
||||
if len(items) != 15:
|
||||
raise ValueError('Pinned mirror shard inventory changed')
|
||||
selected = []
|
||||
for item in items:
|
||||
path = item['path']
|
||||
if not re.fullmatch(r'data/(colm|corl|emnlp|iclr|icml|neurips|tmlr)-\d{5}-of-\d{5}\.parquet', path):
|
||||
raise ValueError('Unexpected mirror shard path')
|
||||
if (item['type'] != 'file' or type(item['size']) is not int or item['size'] < 12
|
||||
or item['lfs']['size'] != item['size'] or not re.fullmatch('[0-9a-f]{64}', item['lfs']['oid'])):
|
||||
raise ValueError('Missing source object size/checksum claim')
|
||||
if path.startswith('data/'+split+'-'):
|
||||
selected.append(item)
|
||||
parts = sorted(selected, key=lambda x: x['path'])
|
||||
if not parts or any(int(x['path'][-13:-8]) != len(parts) for x in parts):
|
||||
raise ValueError('Incomplete selected mirror split inventory')
|
||||
if [int(x['path'].split('-')[-3]) for x in parts] != list(range(len(parts))):
|
||||
raise ValueError('Mirror shard sequence changed')
|
||||
return parts
|
||||
|
||||
|
||||
def decision_claim(raw):
|
||||
value = raw.strip().casefold() if isinstance(raw, str) else ''
|
||||
if value in {'reject', 'rejected'}:
|
||||
return 'rejected'
|
||||
if value in {'desk reject', 'desk rejected', 'desk_rejected'}:
|
||||
return 'desk_rejected'
|
||||
if value in {'withdraw', 'withdrawn'}:
|
||||
return 'withdrawn'
|
||||
if value in {'accept', 'accepted', 'accept-main', 'accept-findings', 'accept (main)', 'accept (findings)',
|
||||
'accept (poster)', 'accept (oral)', 'accept (spotlight)', 'accept (talk)'}:
|
||||
return 'accepted'
|
||||
return 'unknown'
|
||||
|
||||
|
||||
def decision_track_claim(raw):
|
||||
value = raw.strip().casefold() if isinstance(raw, str) else ''
|
||||
if value in {'accept-findings', 'accept (findings)'}:
|
||||
return 'findings'
|
||||
if value in {'accept-main', 'accept (main)'}:
|
||||
return 'main'
|
||||
return None
|
||||
|
||||
|
||||
def record(row, split, item, row_index):
|
||||
if set(row) != set(COLUMNS):
|
||||
raise ValueError('Unexpected projected metadata fields')
|
||||
if (not isinstance(row['conference'], str) or row['conference'].casefold() != split
|
||||
or type(row['year']) is not int or not 2020 <= row['year'] <= 2026):
|
||||
raise ValueError('Mirror conference/year outside pinned source scope')
|
||||
locator = {'dataset': DATASET, 'commit': COMMIT, 'file': item['path'], 'row_index': row_index,
|
||||
'lfs_sha256_claim_not_full_file_verification': item['lfs']['oid']}
|
||||
return {'source': SOURCE, 'external_id': COMMIT+':'+item['path']+':'+str(row_index),
|
||||
'url': 'https://huggingface.co/datasets/'+DATASET+'/blob/'+COMMIT+'/'+item['path'],
|
||||
'title': row['title'], 'abstract': row['abstract'], 'authors': row['authors'],
|
||||
'conference': row['conference'], 'year': row['year'], 'track': row['track'],
|
||||
'mirror_decision_category': decision_claim(row['decision']),
|
||||
'mirror_decision_track_claim': decision_track_claim(row['decision']), 'raw_fields': row,
|
||||
'source_locator': locator, 'identifiers': {},
|
||||
'metadata': {'forum_id_claim': row['forum_id'], 'original_forum_id_verified': False,
|
||||
'original_decision_signature_verified': False, 'source_visibility': 'public_mirror',
|
||||
'scientific_work_identity_approved': False, 'research_type_classified': False,
|
||||
'approved_for_import': False,
|
||||
'scope': 'Mirror row only. No original forum/person/decision verification; absence is not rejection or proof of no later acceptance.'}}
|
||||
|
||||
|
||||
def cached_projections(path, expected_sha256, split, items):
|
||||
path = Path(path)
|
||||
raw = path.read_bytes()
|
||||
if (not isinstance(expected_sha256, str) or not re.fullmatch('[0-9a-f]{64}', expected_sha256)
|
||||
or hashlib.sha256(raw).hexdigest() != expected_sha256):
|
||||
raise ValueError('Cached projection report hash differs')
|
||||
report = json.loads(raw)
|
||||
if (report.get('dataset') != DATASET or report.get('commit') != COMMIT
|
||||
or report.get('split') != split or report.get('inventory_sha256') != TREE_SHA256
|
||||
or report.get('raw_parquet_saved') is not False or report.get('database_written') is not False
|
||||
or report.get('paper_body_saved') is not False or report.get('review_text_saved') is not False
|
||||
or report.get('approved_for_import') != 0):
|
||||
raise ValueError('Cached report scope differs from permitted mirror projection')
|
||||
declared = report.get('files', [])
|
||||
if len(declared) != len(items) or len({f['source_file'] for f in declared}) != len(items):
|
||||
raise ValueError('Incomplete cached mirror shard set')
|
||||
result = {}
|
||||
for item in items:
|
||||
fm = next((f for f in declared if f['source_file'] == item['path']), None)
|
||||
filename = Path(item['path']).stem+'.metadata.jsonl'
|
||||
if not fm or Path(fm['file']).name != filename or fm['source_lfs_sha256_claim'] != item['lfs']['oid']:
|
||||
raise ValueError('Cached mirror source differs')
|
||||
file = path.parent/filename
|
||||
if file.is_symlink():
|
||||
raise ValueError('Symlink metadata cache forbidden')
|
||||
content = file.read_bytes()
|
||||
if hashlib.sha256(content).hexdigest() != fm['file_sha256']:
|
||||
raise ValueError('Cached metadata file hash differs')
|
||||
rows = []
|
||||
for index, line in enumerate(content.splitlines()):
|
||||
previous = json.loads(line); row = previous['raw_fields']; expected = record(row, split, item, index)
|
||||
for key in ('source', 'external_id', 'url', 'title', 'abstract', 'authors', 'conference', 'year', 'track', 'raw_fields', 'source_locator', 'identifiers', 'metadata'):
|
||||
if previous.get(key) != expected[key]:
|
||||
raise ValueError('Cached raw metadata/identity differs: '+key)
|
||||
rows.append(row)
|
||||
proof = fm['projection']
|
||||
if (len(rows) != fm['records'] or len(rows) != proof['rows'] or proof['columns'] != COLUMNS
|
||||
or proof['source_bytes'] != item['size'] or proof['raw_parquet_saved'] is not False
|
||||
or proof['footer_saved'] is not False or proof['excluded_payload_pages_downloaded'] is not False
|
||||
or proof['full_file_checksum_verified'] is not False):
|
||||
raise ValueError('Cached projection proof differs')
|
||||
result[item['path']] = (rows, proof)
|
||||
if sum(len(rows) for rows, _ in result.values()) != report['records']:
|
||||
raise ValueError('Cached projection row count differs')
|
||||
return result
|
||||
|
||||
|
||||
def collect(split, out, *, source_factory=RangeSource, progress=None, from_projection=None, projection_sha256=None):
|
||||
items = inventory(split); out = Path(out)
|
||||
if bool(from_projection) != bool(projection_sha256):
|
||||
raise ValueError('Cached projection and exact report hash required together')
|
||||
if out.exists() or not out.parent.is_dir() or out.is_symlink():
|
||||
raise ValueError('New output directory in existing parent required')
|
||||
out.mkdir()
|
||||
counts = Counter(); track_claims = Counter(); raw_counts = Counter(); years = Counter(); records = 0; reports = []
|
||||
observed = datetime.now(timezone.utc).isoformat()
|
||||
try:
|
||||
cached = cached_projections(from_projection, projection_sha256, split, items) if from_projection else None
|
||||
for item in items:
|
||||
if cached is None:
|
||||
source_url = 'https://huggingface.co/datasets/'+DATASET+'/resolve/'+COMMIT+'/'+item['path']
|
||||
source = source_factory(source_url, item['size'])
|
||||
rows, proof = projection(source, COLUMNS, progress=progress)
|
||||
else:
|
||||
rows, proof = cached[item['path']]
|
||||
if progress:
|
||||
progress({'phase': 'cached_metadata_projection', 'file': item['path'], 'rows': len(rows)})
|
||||
path = out/(Path(item['path']).stem+'.metadata.jsonl')
|
||||
part = path.with_suffix(path.suffix+'.part'); h = hashlib.sha256()
|
||||
with part.open('wb') as stream:
|
||||
for index, row in enumerate(rows):
|
||||
item_record = record(row, split, item, index)
|
||||
raw = (json.dumps(item_record, ensure_ascii=False)+'\n').encode()
|
||||
stream.write(raw); h.update(raw); records += 1
|
||||
counts[item_record['mirror_decision_category']] += 1
|
||||
track_claims[item_record['mirror_decision_track_claim'] or 'unspecified'] += 1
|
||||
raw_counts[json.dumps(row['decision'], ensure_ascii=False)] += 1
|
||||
years[(row['year'], row['track'])] += 1
|
||||
part.rename(path)
|
||||
reports.append({'source_file': item['path'], 'file': str(path), 'records': len(rows),
|
||||
'file_sha256': h.hexdigest(), 'source_lfs_sha256_claim': item['lfs']['oid'],
|
||||
'projection': proof})
|
||||
report = {'status': 'collected_mirror_metadata_pending_original_provenance_and_work_identity',
|
||||
'dataset': DATASET, 'commit': COMMIT, 'inventory_sha256': TREE_SHA256,
|
||||
'split': split, 'observed_at': observed, 'records': records, 'files': reports,
|
||||
'mirror_decision_counts': dict(counts), 'raw_mirror_decision_counts': dict(raw_counts),
|
||||
'mirror_decision_track_claim_counts': dict(track_claims),
|
||||
'year_track_counts': [{'year': y, 'track': t, 'rows': n} for (y, t), n in sorted(years.items(), key=lambda x: (x[0][0], str(x[0][1])))],
|
||||
'metadata_bytes_read': sum(x['projection']['metadata_bytes_read'] for x in reports) if cached is None else 0,
|
||||
'source_projection_bytes': sum(x['projection']['metadata_bytes_read'] for x in reports),
|
||||
'cached_projection_report_sha256': projection_sha256,
|
||||
'excluded_payload_pages_downloaded': False, 'raw_parquet_saved': False,
|
||||
'paper_body_saved': False, 'review_text_saved': False, 'database_written': False,
|
||||
'approved_for_import': 0,
|
||||
'limitations': ['Public mirror copies do not verify original forum IDs, visibility or signed decisions.',
|
||||
'Original source LFS SHA256 is a pinned claim, not a computed full-Parquet checksum.',
|
||||
'Every selected row and row locator is retained, including missing or anonymous author data.',
|
||||
'Decision categories describe only explicit mirror literals; venue labels and missing decisions do not imply rejection.',
|
||||
'Rows are not independent scientific works or a complete venue/year public negative census.',
|
||||
'Cross-source identity, chronological rejection/acceptance history and final dedup remain.']}
|
||||
(out/'collection-report.json').write_text(json.dumps(report, ensure_ascii=False, indent=2)+'\n')
|
||||
return report
|
||||
except BaseException as error:
|
||||
# No source-specific body or signed redirect URLs in failure reports.
|
||||
failure = {'status': 'incomplete_do_not_import', 'split': split, 'commit': COMMIT,
|
||||
'completed_files': reports, 'error_type': type(error).__name__,
|
||||
'database_written': False, 'approved_for_import': 0, 'no_automatic_retry': True}
|
||||
(out/'failure.json').write_text(json.dumps(failure, ensure_ascii=False, indent=2)+'\n')
|
||||
raise
|
||||
|
||||
|
||||
def main(argv=None, *, prog=None):
|
||||
parser = argparse.ArgumentParser(prog=prog)
|
||||
parser.add_argument('--split', required=True, choices=SPLITS)
|
||||
parser.add_argument('--out', required=True)
|
||||
parser.add_argument('--from-projection', help='Offline replay of a complete permitted metadata projection')
|
||||
parser.add_argument('--projection-sha256', help='Exact cached collection-report.json SHA256')
|
||||
args = parser.parse_args(argv)
|
||||
report = collect(args.split, args.out, from_projection=args.from_projection, projection_sha256=args.projection_sha256,
|
||||
progress=lambda row: print(json.dumps(row), flush=True))
|
||||
print(json.dumps({'status': report['status'], 'records': report['records'], 'out': args.out}), flush=True)
|
||||
@@ -0,0 +1,195 @@
|
||||
"""Read only selected Parquet column chunks; never fetch other payload pages."""
|
||||
import hashlib
|
||||
import io
|
||||
import re
|
||||
import struct
|
||||
import urllib.error
|
||||
import urllib.request
|
||||
|
||||
MAX_FOOTER = 4 * 1024**2
|
||||
MAX_SELECTED_BYTES = 64 * 1024**2
|
||||
MAX_REQUESTS = 512
|
||||
|
||||
|
||||
class RangeSource:
|
||||
"""Strict bounded HTTP ranges, no full-file fallback, redirects kept private."""
|
||||
def __init__(self, url, size, *, opener=None):
|
||||
self.url, self.size = url, size
|
||||
self.opener = opener or urllib.request.build_opener()
|
||||
self.etag = None
|
||||
self.receipts = []
|
||||
self.bytes_read = 0
|
||||
|
||||
def read(self, start, length):
|
||||
if (type(start) is not int or type(length) is not int or length <= 0
|
||||
or start < 0 or start + length > self.size
|
||||
or self.bytes_read + length > MAX_SELECTED_BYTES
|
||||
or len(self.receipts) >= MAX_REQUESTS):
|
||||
raise ValueError('Bounded metadata range budget exceeded')
|
||||
headers = {'Range': f'bytes={start}-{start + length - 1}', 'Accept-Encoding': 'identity'}
|
||||
if self.etag and not self.etag.startswith('W/'):
|
||||
headers['If-Match'] = self.etag
|
||||
request = urllib.request.Request(self.url, headers=headers)
|
||||
try:
|
||||
with self.opener.open(request, timeout=45) as response:
|
||||
if response.status != 206:
|
||||
raise ValueError('Server did not honor metadata range; full download forbidden')
|
||||
if response.headers.get('Content-Range') != f'bytes {start}-{start + length - 1}/{self.size}':
|
||||
raise ValueError('Object size or range differs from pinned inventory')
|
||||
encoding = response.headers.get('Content-Encoding', 'identity')
|
||||
if encoding != 'identity':
|
||||
raise ValueError('Encoded HTTP ranges are unsupported')
|
||||
etag = response.headers.get('ETag')
|
||||
if self.receipts and etag != self.etag:
|
||||
raise ValueError('Object ETag changed between metadata ranges')
|
||||
raw = response.read(length + 1)
|
||||
if len(raw) != length:
|
||||
raise ValueError('Truncated or oversized metadata range')
|
||||
if not self.receipts:
|
||||
self.etag = etag
|
||||
except urllib.error.HTTPError as error:
|
||||
# Do not expose signed CDN redirect URLs or retry denied resources.
|
||||
raise RuntimeError('Metadata source HTTP ' + str(error.code) + '; no retry or full-file fallback') from None
|
||||
self.receipts.append({'start': start, 'length': length, 'sha256': hashlib.sha256(raw).hexdigest()})
|
||||
self.bytes_read += length
|
||||
return raw
|
||||
|
||||
|
||||
class SparseFile(io.RawIOBase):
|
||||
"""A logical file backed only by approved bytes; all other positions are zeros.
|
||||
|
||||
Arrow may read a 64 KiB footer window. Synthesizing its unused prefix avoids
|
||||
fetching the preceding OCR/review page. Data pages are explicitly selected
|
||||
from footer offsets before they can be downloaded.
|
||||
"""
|
||||
def __init__(self, size, segments):
|
||||
self.size, self.segments, self.position = size, segments, 0
|
||||
|
||||
def readable(self):
|
||||
return True
|
||||
|
||||
def seekable(self):
|
||||
return True
|
||||
|
||||
def tell(self):
|
||||
return self.position
|
||||
|
||||
def seek(self, offset, whence=0):
|
||||
base = {0: 0, 1: self.position, 2: self.size}.get(whence)
|
||||
if base is None or base + offset < 0:
|
||||
raise ValueError('Invalid sparse metadata seek')
|
||||
self.position = base + offset
|
||||
return self.position
|
||||
|
||||
def read(self, length=-1):
|
||||
if length < 0:
|
||||
length = self.size - self.position
|
||||
length = min(length, max(0, self.size - self.position))
|
||||
if length > MAX_SELECTED_BYTES:
|
||||
raise ValueError('Unbounded sparse read forbidden')
|
||||
start, end = self.position, self.position + length
|
||||
raw = bytearray(length)
|
||||
for offset, data in self.segments:
|
||||
left, right = max(offset, start), min(offset + len(data), end)
|
||||
if left < right:
|
||||
raw[left-start:right-start] = data[left-offset:right-offset]
|
||||
self.position = end
|
||||
return bytes(raw)
|
||||
|
||||
def readinto(self, buffer):
|
||||
raw = self.read(len(buffer)); buffer[:len(raw)] = raw
|
||||
return len(raw)
|
||||
|
||||
|
||||
def arrow():
|
||||
try:
|
||||
import pyarrow as pa
|
||||
import pyarrow.parquet as pq
|
||||
except ImportError:
|
||||
raise RuntimeError('Install paper-atlas[mirrors] only on the approved remote runner') from None
|
||||
return pa, pq
|
||||
|
||||
|
||||
def projection(source, columns, *, progress=None):
|
||||
"""Return complete metadata rows and range receipts, no full Parquet snapshot."""
|
||||
pa, pq = arrow()
|
||||
if type(source.size) is not int or not 12 <= source.size <= 4 * 1024**3:
|
||||
raise ValueError('Bounded pinned Parquet size required')
|
||||
if (not isinstance(columns, dict) or not columns
|
||||
or any(not re.fullmatch('[a-z][a-z0-9_]*', key) for key in columns)):
|
||||
raise ValueError('Explicit top-level metadata columns required')
|
||||
header = source.read(0, 4); trailer = source.read(source.size - 8, 8)
|
||||
if header != b'PAR1' or trailer[4:] != b'PAR1':
|
||||
raise ValueError('Expected unencrypted Parquet framing')
|
||||
length = struct.unpack('<I', trailer[:4])[0]
|
||||
if not 0 < length <= min(MAX_FOOTER, source.size - 12):
|
||||
raise ValueError('Invalid or oversized Parquet footer')
|
||||
footer_start = source.size - 8 - length
|
||||
footer = source.read(footer_start, length)
|
||||
segments = [(0, header), (footer_start, footer), (source.size - 8, trailer)]
|
||||
sparse = SparseFile(source.size, segments)
|
||||
parquet = pq.ParquetFile(sparse, pre_buffer=False, buffer_size=0)
|
||||
try:
|
||||
schema = parquet.schema_arrow
|
||||
if len(set(schema.names)) != len(schema.names) or not set(columns) <= set(schema.names):
|
||||
raise ValueError('Missing or repeated metadata columns')
|
||||
for key, kind in columns.items():
|
||||
value = schema.field(key).type
|
||||
if kind == 'string':
|
||||
okay = pa.types.is_string(value) or pa.types.is_large_string(value)
|
||||
elif kind == 'integer':
|
||||
okay = pa.types.is_integer(value)
|
||||
elif kind == 'strings':
|
||||
okay = (pa.types.is_list(value) or pa.types.is_large_list(value)) and (
|
||||
pa.types.is_string(value.value_type) or pa.types.is_large_string(value.value_type))
|
||||
else:
|
||||
raise ValueError('Unsupported safe metadata type')
|
||||
if not okay:
|
||||
raise ValueError('Metadata column type differs: ' + key)
|
||||
metadata = parquet.metadata
|
||||
if not 0 < metadata.num_rows <= 100000 or metadata.num_row_groups > 1000:
|
||||
raise ValueError('Bounded metadata row count required')
|
||||
wanted, excluded = [], []
|
||||
for group in range(metadata.num_row_groups):
|
||||
rg = metadata.row_group(group)
|
||||
for i in range(rg.num_columns):
|
||||
col = rg.column(i)
|
||||
if col.file_path:
|
||||
raise ValueError('External Parquet column files forbidden')
|
||||
offsets = [x for x in (col.dictionary_page_offset, col.data_page_offset) if x is not None and x > 0]
|
||||
if not offsets:
|
||||
raise ValueError('Missing column chunk offset')
|
||||
start = min(offsets); end = start + col.total_compressed_size
|
||||
if not 4 <= start < end <= footer_start:
|
||||
raise ValueError('Column chunk lies outside the pinned payload')
|
||||
item = (start, end, group, col.path_in_schema)
|
||||
(wanted if col.path_in_schema.split('.')[0] in columns else excluded).append(item)
|
||||
if any(max(a, c) < min(b, d) for a, b, _, _ in wanted for c, d, _, _ in excluded):
|
||||
raise ValueError('Selected range overlaps an excluded payload column')
|
||||
if sum(end-start for start, end, _, _ in wanted) + length + 12 > MAX_SELECTED_BYTES:
|
||||
raise ValueError('Metadata projection exceeds byte budget')
|
||||
if len(wanted) + 3 > MAX_REQUESTS:
|
||||
raise ValueError('Metadata projection exceeds request budget')
|
||||
for start, end, group, path in sorted(wanted):
|
||||
segments.append((start, source.read(start, end-start)))
|
||||
if progress:
|
||||
progress({'phase': 'metadata_column_range', 'row_group': group,
|
||||
'column': path, 'metadata_bytes_read': source.bytes_read})
|
||||
rows = []
|
||||
for group in range(metadata.num_row_groups):
|
||||
table = parquet.read_row_group(group, columns=list(columns), use_threads=False)
|
||||
if set(table.column_names) != set(columns) or len(table.column_names) != len(columns):
|
||||
raise ValueError('Unexpected projected metadata fields')
|
||||
rows.extend(table.select(list(columns)).to_pylist())
|
||||
if len(rows) != metadata.num_rows:
|
||||
raise ValueError('Incomplete metadata projection')
|
||||
report = {'rows': len(rows), 'row_groups': metadata.num_row_groups,
|
||||
'columns': dict(columns), 'footer_sha256': hashlib.sha256(footer).hexdigest(),
|
||||
'column_ranges': [{'start': a, 'length': b-a, 'row_group': g, 'column': p} for a, b, g, p in wanted],
|
||||
'range_receipts': source.receipts, 'metadata_bytes_read': source.bytes_read,
|
||||
'source_bytes': source.size, 'full_file_checksum_verified': False,
|
||||
'excluded_payload_pages_downloaded': False,
|
||||
'footer_saved': False, 'raw_parquet_saved': False}
|
||||
return rows, report
|
||||
finally:
|
||||
parquet.close()
|
||||
@@ -0,0 +1 @@
|
||||
[{"type":"file","oid":"f87639c8a74ab77de91ea6bb4c88718a52e4250f","size":37291354,"lfs":{"oid":"bcb446a0398dfe62593e1db617f82f43bc792fc413d6d4658b0e0d32c358b888","size":37291354,"pointerSize":133},"xetHash":"c9a6d5d70b6f3a9ee09b4e085327e4f61cad91b9b61244e6bf99031fef9dd85b","path":"data/colm-00000-of-00001.parquet"},{"type":"file","oid":"27ef0f337b76d92ac1aaa4e1dc41a1253768545b","size":28627916,"lfs":{"oid":"930dd4ef90fa352fe6bccca7b73eb36eeabfb99f1e77e783572e2db83e15d779","size":28627916,"pointerSize":133},"xetHash":"468b49fe8b9a921e852c1995c4902e175ad7b54fcefb048a485e3fbad8b62bfd","path":"data/corl-00000-of-00001.parquet"},{"type":"file","oid":"c8de3418de080cbd24667e28c55abf684a8a3d9f","size":90794691,"lfs":{"oid":"e9e3aab42d28ca32a02287ed0bd0f1f9e1f3af5cb8ed8d6a9bf28d4f630773da","size":90794691,"pointerSize":133},"xetHash":"c21b83e5820cea8b7a6bcac34bbd334a0a6ac5b7acb1c4c681715a52ec2cceda","path":"data/emnlp-00000-of-00001.parquet"},{"type":"file","oid":"0a226079bb3fe5256a51750aa874c0e7325f60fa","size":167662887,"lfs":{"oid":"3a65801f95fe490b30282df2cb668fa33fe4182e9d80544148ea66da35215ad8","size":167662887,"pointerSize":134},"xetHash":"2225125f6d18015bb3c44037c82a7f53363eeb636d309de3445e1f1b49f51a03","path":"data/iclr-00000-of-00005.parquet"},{"type":"file","oid":"7b53c49c39f2c287a778e64b18969690926710a8","size":196573710,"lfs":{"oid":"736c05be45eed9db00848471d88dc9cd5a1b1a7d5a0db6432639a21dfb8b4562","size":196573710,"pointerSize":134},"xetHash":"48da5a74c33ca8b04411f4d6f41e34a4ba57d8c4aecb1a72be5dcad89a72bd65","path":"data/iclr-00001-of-00005.parquet"},{"type":"file","oid":"cbf09e34f7f00e33f5bb501461a915ab4bda2ec2","size":210807163,"lfs":{"oid":"1c0b77c370b8735e3d4ebd578706ec8dfbb5e53eb6a71e8cbfeaec36c78ee13f","size":210807163,"pointerSize":134},"xetHash":"fa8bf4848b55d274464b6d06413e1183321e6ffd467e1efa369ee32e0749e8af","path":"data/iclr-00002-of-00005.parquet"},{"type":"file","oid":"4082a290b2e2785413340920e3475c4b5a58b05f","size":250525528,"lfs":{"oid":"078d441ef3854be30f8fd73d34e8442a2a119cd26dbaddf9ab8aba2008df507f","size":250525528,"pointerSize":134},"xetHash":"63ba7c9757b53b02b1793f2186ddc55b79149dbda9a37bacedaf2b0466604d61","path":"data/iclr-00003-of-00005.parquet"},{"type":"file","oid":"d4528dcfcd6d7f56619d2bed75d4705c11811bfb","size":263725824,"lfs":{"oid":"de5efe8578b744f59c8ff9d7c1171cdaa760008920dde190c04b73e5c45901a5","size":263725824,"pointerSize":134},"xetHash":"5934eb773ec49ebf9f48c89d9d7e88ae0d2eb5f5e7f83bfd13f51080aa6d4090","path":"data/iclr-00004-of-00005.parquet"},{"type":"file","oid":"68cd41fa1983b2ba666deb74d1e6da2f9fa9003d","size":204752048,"lfs":{"oid":"4f50e2a0fd7e40a0fdb239595b636c753ca89af1ad9692b7d63f366e7a25eda0","size":204752048,"pointerSize":134},"xetHash":"de8ac963c1738ffe68728255bf7f32fbfff3863b3b07ca1f7026f4c76f123ec0","path":"data/icml-00000-of-00001.parquet"},{"type":"file","oid":"580a98d667a2fc7252fd2bcc2eeeaff116f7030d","size":126503813,"lfs":{"oid":"d2c4359745d30900bd19a44762c9fb6b8415554e2aead6f22f902362aca3e61a","size":126503813,"pointerSize":134},"xetHash":"27cd8f2318a0d096cf3a6aa7178fc502ce52c5d026887f64661b46af74b4b9d5","path":"data/neurips-00000-of-00005.parquet"},{"type":"file","oid":"e90e69bdd1985a36f9a72b4b3b3a9b54d90b7b9d","size":166685127,"lfs":{"oid":"32be2572bc42d33ff227ceadbdcc64e27c996b31e82195d25eced11ec4617963","size":166685127,"pointerSize":134},"xetHash":"2e3528c026d585513df96b23efe39d174b368abd3088e494d85fcf806504a241","path":"data/neurips-00001-of-00005.parquet"},{"type":"file","oid":"b7be2e96d3c3726b87c4a997fcb2c144288b70b4","size":229115194,"lfs":{"oid":"7f9dc8025aebe1d78f74bfccc68e1792f6a076142a920305065d805ff04ab248","size":229115194,"pointerSize":134},"xetHash":"77a4b411199fee5bd1b3fb8752ad4bf1ac090d560a4bfde54b75cf19d48c545b","path":"data/neurips-00002-of-00005.parquet"},{"type":"file","oid":"127763711c2fe76631137432291fa839746536a3","size":264212756,"lfs":{"oid":"32d86e74f00d3fce8eef1cf1db1dd78eed341ddda0e09a924ba17d08c3d566b2","size":264212756,"pointerSize":134},"xetHash":"abe6a12cf6f6f2db52ba28b84a0b91fcfab1411dfa0eb0ab7cdae540b86c5c9d","path":"data/neurips-00003-of-00005.parquet"},{"type":"file","oid":"89eb22ff72a8f44d461e601176239d4f997bbec7","size":279556753,"lfs":{"oid":"b4a536cdb6d306049a32121272b8224ef0e37360423ca04c274f0da74d4dc64c","size":279556753,"pointerSize":134},"xetHash":"d2b68be869a02a2cbe81c74e50fcac3c35585422dbaffcb19985e2376a90e0a7","path":"data/neurips-00004-of-00005.parquet"},{"type":"file","oid":"2b6048e08f894bb47b874a485d3bc5655a43a449","size":211604858,"lfs":{"oid":"df21b892ec531bc223b114d03cd51be6f4f2c48563edde0bdd1b153c36a74a98","size":211604858,"pointerSize":134},"xetHash":"1e81fc6612959ef821e97b7b5ed06a9ebbb3748f0ecc2fe2027b41f7626ef863","path":"data/tmlr-00000-of-00001.parquet"}]
|
||||
+2
-1
@@ -12,6 +12,7 @@ dependencies = ["PyYAML==6.0.3"]
|
||||
|
||||
[project.optional-dependencies]
|
||||
postgres = ["psycopg[binary]==3.3.8"]
|
||||
mirrors = ["pyarrow==26.0.1"]
|
||||
|
||||
[project.scripts]
|
||||
paper-atlas = "paper_atlas.cli:main"
|
||||
@@ -22,4 +23,4 @@ exclude = ["tests*"]
|
||||
|
||||
[tool.setuptools.package-data]
|
||||
"paper_atlas.db" = ["schema.sql", "annotations.sql"]
|
||||
"paper_atlas.resources" = ["venues.json"]
|
||||
"paper_atlas.resources" = ["venues.json", "reviewarena-files.json"]
|
||||
|
||||
BIN
Binary file not shown.
@@ -0,0 +1,218 @@
|
||||
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.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.fixture['independent_gold']})
|
||||
for g in self.fixture['independent_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.fixture['independent_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_exact_metadata_correspondence')
|
||||
else:
|
||||
self.assertIsNone(r['abstract']);self.assertEqual(m['abstract_source'],'not_available_from_reviewed_sources')
|
||||
self.assertEqual(count,1717);self.assertEqual(self.report['output_missing_abstract_count'],201)
|
||||
|
||||
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'],417)
|
||||
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),192)
|
||||
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.fixture['independent_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.fixture['independent_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_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,'program_author_version_requires_review':122,'no_exact_program_title_requires_review':70})
|
||||
self.assertEqual(self.report['registry_missing_abstract_count'],1918)
|
||||
self.assertEqual(self.report['abstracts_from_exact_program_correspondence'],1717)
|
||||
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'),1726)
|
||||
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'])
|
||||
@@ -0,0 +1,101 @@
|
||||
import gzip
|
||||
import hashlib
|
||||
import json
|
||||
from pathlib import Path
|
||||
import tempfile
|
||||
import unittest
|
||||
from unittest.mock import patch
|
||||
|
||||
from paper_atlas.collectors import mm2026
|
||||
from paper_atlas.audits.source_identity import audit, load_inputs, load_mm_program_inputs
|
||||
|
||||
|
||||
class MMProgramIdentityInputTests(unittest.TestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.fixture = json.loads(gzip.decompress((Path(__file__).parent/'fixtures/mm2026-source.json.gz').read_bytes()))
|
||||
|
||||
def setUp(self):
|
||||
self.tmp = tempfile.TemporaryDirectory()
|
||||
self.addCleanup(self.tmp.cleanup)
|
||||
self.root = Path(self.tmp.name)
|
||||
src = self.root/'source'
|
||||
for name,body in self.fixture['files'].items():
|
||||
path=src/name;path.parent.mkdir(parents=True,exist_ok=True);path.write_bytes(body.encode())
|
||||
mm2026.collect(self.root/'program', snapshots=src)
|
||||
self.report_path = self.root/'program/collection-report.json'
|
||||
|
||||
def change_report(self, change):
|
||||
report=json.loads(self.report_path.read_text());change(report)
|
||||
self.report_path.write_text(json.dumps(report))
|
||||
|
||||
def assert_rejected_before_database(self):
|
||||
with patch('paper_atlas.audits.source_identity.ReadConnection') as connection:
|
||||
with self.assertRaises(ValueError):
|
||||
audit(None,self.root/'audit',pg_service='unused',pg_schema='research',mm_program_report=self.report_path)
|
||||
connection.assert_not_called()
|
||||
self.assertFalse((self.root/'audit').exists())
|
||||
|
||||
def test_all2143_programs_are_unknown_readonly_inputs_not_import_records(self):
|
||||
rows,inputs=load_mm_program_inputs(self.report_path)
|
||||
self.assertEqual(len(rows),2143);self.assertEqual(len(inputs),28)
|
||||
self.assertEqual(sum(len(r['program_contexts']) for r in rows),4326)
|
||||
actual={r['program_id']:r for r in rows}
|
||||
for gold in self.fixture['independent_inventory']['inventory']:
|
||||
r=actual[gold['paper_id']]
|
||||
self.assertEqual((r['title'],r['authors'],r['track']),
|
||||
(gold['title'],gold['authors'],gold['track_literal']))
|
||||
self.assertEqual(r['external_id'],'2026/'+gold['paper_id'])
|
||||
self.assertEqual(r['status'],'unknown');self.assertEqual(r['identifiers'],{})
|
||||
self.assertEqual(r['source_observed_status'],'listed_in_official_program')
|
||||
self.assertFalse(r['publication_identity_verified']);self.assertFalse(r['final_decision_verified'])
|
||||
self.assertFalse(r['contribution_type_verified'])
|
||||
self.assertEqual([c['source_pointer'] for c in r['program_contexts']],gold['locations'])
|
||||
self.assertEqual({r['program_id'] for r in rows if not r['authors']},{'mfp4971','mfp7801'})
|
||||
self.assertTrue(all('unknown is not a source decision' in i['manifest']['scope'] for i in inputs))
|
||||
self.assertFalse(any(self.root.rglob('*.manifest.json')))
|
||||
|
||||
def test_untrusted_absolute_file_paths_are_not_followed(self):
|
||||
self.change_report(lambda r:[f.update(file='/untrusted/'+Path(f['file']).name) for f in r['files']])
|
||||
rows,inputs=load_mm_program_inputs(self.report_path)
|
||||
self.assertEqual(len(rows),2143)
|
||||
self.assertTrue(all(Path(i['file']).parent==self.report_path.parent for i in inputs))
|
||||
|
||||
def test_program_files_are_not_silently_accepted_by_regular_ingestion_input_loader(self):
|
||||
report=json.loads(self.report_path.read_text())
|
||||
with self.assertRaises(FileNotFoundError):load_inputs([report['files'][0]['file']])
|
||||
|
||||
def test_corrupt_jsonl_fails_before_connection_or_output(self):
|
||||
report=json.loads(self.report_path.read_text());path=Path(report['files'][0]['file'])
|
||||
path.write_bytes(path.read_bytes()+b' ')
|
||||
self.assert_rejected_before_database()
|
||||
|
||||
def test_rehashed_forged_identifiers_decisions_and_contexts_rejected(self):
|
||||
original=self.report_path.read_bytes();report=json.loads(original);path=Path(report['files'][0]['file']);raw=path.read_bytes()
|
||||
for field,value in [('identifiers',{'openreview':'invented'}),('status','accepted'),('contexts',[])]:
|
||||
path.write_bytes(raw);self.report_path.write_bytes(original)
|
||||
lines=raw.splitlines();row=json.loads(lines[0]);row[field]=value;lines[0]=json.dumps(row).encode()
|
||||
changed=b'\n'.join(lines)+b'\n';path.write_bytes(changed)
|
||||
self.change_report(lambda r:r['files'][0].update(file_sha256=hashlib.sha256(changed).hexdigest()))
|
||||
with self.subTest(field=field):self.assert_rejected_before_database()
|
||||
|
||||
def test_missing_or_repeated_track_rejected(self):
|
||||
original=self.report_path.read_bytes()
|
||||
for kind in ('missing','repeated'):
|
||||
self.report_path.write_bytes(original)
|
||||
if kind=='missing':self.change_report(lambda r:r['files'].pop())
|
||||
else:self.change_report(lambda r:r['files'].__setitem__(1,r['files'][0]))
|
||||
with self.subTest(kind=kind):self.assert_rejected_before_database()
|
||||
|
||||
def test_raw_source_corruption_or_symlink_rejected(self):
|
||||
path=self.report_path.parent/'evidence/program-data.json';raw=path.read_bytes()
|
||||
path.write_bytes(raw+b' ');self.assert_rejected_before_database()
|
||||
path.write_bytes(raw);saved=path.with_name('original.json');path.rename(saved);path.symlink_to(saved)
|
||||
self.assert_rejected_before_database()
|
||||
|
||||
def test_mixed_input_modes_rejected_before_connection(self):
|
||||
with patch('paper_atlas.audits.source_identity.ReadConnection') as connection:
|
||||
for files,program in [([],None),([Path('unused')],self.report_path)]:
|
||||
with self.assertRaisesRegex(ValueError,'Choose source files'):
|
||||
audit(files,self.root/'out',pg_service='unused',pg_schema='research',mm_program_report=program)
|
||||
connection.assert_not_called()
|
||||
@@ -0,0 +1,203 @@
|
||||
"""Metadata projection never requests OCR/review column pages or promotes mirror IDs."""
|
||||
from collections import Counter
|
||||
import hashlib
|
||||
import importlib.util
|
||||
import io
|
||||
import json
|
||||
from pathlib import Path
|
||||
import re
|
||||
import struct
|
||||
import tempfile
|
||||
import unittest
|
||||
from unittest.mock import patch
|
||||
import urllib.error
|
||||
|
||||
from paper_atlas.collectors.reviewarena import COLUMNS, COMMIT, collect, decision_claim, decision_track_claim, inventory, record
|
||||
from paper_atlas.io.parquet_metadata import RangeSource, SparseFile, projection
|
||||
|
||||
|
||||
class MemorySource:
|
||||
def __init__(self, raw):
|
||||
self.raw, self.size, self.bytes_read, self.receipts = raw, len(raw), 0, []
|
||||
|
||||
def read(self, start, length):
|
||||
raw = self.raw[start:start+length]
|
||||
self.bytes_read += len(raw)
|
||||
self.receipts.append({'start': start, 'length': length, 'sha256': hashlib.sha256(raw).hexdigest()})
|
||||
return raw
|
||||
|
||||
|
||||
def fixture():
|
||||
import pyarrow as pa
|
||||
import pyarrow.parquet as pq
|
||||
rows = []
|
||||
for i, decision in enumerate(['Accept-Findings', 'Accept-Main', 'Reject', 'Withdrawn', None]):
|
||||
row = {'forum_id': 'unverified-source-'+str(i)+'.xml', 'conference': 'emnlp', 'year': 2023,
|
||||
'track': 'main', 'venue_id': 'source-venue-claim',
|
||||
'paper_number': i, 'title': 'Metadata title '+str(i), 'abstract': 'Metadata abstract\x00 '+str(i),
|
||||
'authors': ['Anonymous'] if i == 2 else ['Ada Author', 'Yi Researcher'],
|
||||
'keywords': ['metadata'], 'venue': 'Source venue label', 'decision': decision,
|
||||
'markdown': 'OCR_FULLTEXT_SENTINEL '+('private-ish fixture body '*10000),
|
||||
'reviews_json': 'REVIEW_FULLTEXT_SENTINEL '+('fixture review '*10000),
|
||||
'decision_comment': 'META_REVIEW_SENTINEL '+('fixture meta-review '*10000)}
|
||||
rows.append(row)
|
||||
sink = io.BytesIO(); pq.write_table(pa.Table.from_pylist(rows), sink, row_group_size=2, compression='snappy')
|
||||
return sink.getvalue(), [{k: v for k, v in r.items() if k in COLUMNS} for r in rows]
|
||||
|
||||
|
||||
class RangeHTTPTests(unittest.TestCase):
|
||||
def response(self, data, status=206, start=2, size=10, etag='"stable"'):
|
||||
response = io.BytesIO(data); response.status = status
|
||||
response.headers = {'Content-Range': f'bytes {start}-{start+len(data)-1}/{size}', 'ETag': etag}
|
||||
return response
|
||||
|
||||
def test_strict_range_and_changed_object_detection(self):
|
||||
opener = unittest.mock.Mock(); opener.open.side_effect = [self.response(b'ab'), self.response(b'cd', start=4, etag='"changed"')]
|
||||
source = RangeSource('https://example.org/public', 10, opener=opener)
|
||||
self.assertEqual(source.read(2, 2), b'ab')
|
||||
self.assertEqual(opener.open.call_args.args[0].get_header('Range'), 'bytes=2-3')
|
||||
with self.assertRaisesRegex(ValueError, 'ETag'):
|
||||
source.read(4, 2)
|
||||
self.assertEqual(source.bytes_read, 2)
|
||||
|
||||
def test_no_full_file_fallback_or_denied_source_retry(self):
|
||||
opener = unittest.mock.Mock(); response = self.response(b'ab', status=200)
|
||||
response.read = unittest.mock.Mock(side_effect=AssertionError('Must not consume full body'))
|
||||
opener.open.return_value = response
|
||||
with self.assertRaisesRegex(ValueError, 'full download forbidden'):
|
||||
RangeSource('https://example.org/public', 10, opener=opener).read(2, 2)
|
||||
response.read.assert_not_called()
|
||||
opener = unittest.mock.Mock(); opener.open.side_effect = urllib.error.HTTPError('https://cdn.example/signed?SECRET=hidden', 403, 'Forbidden', {}, None)
|
||||
with self.assertRaisesRegex(RuntimeError, '^Metadata source HTTP 403; no retry or full-file fallback$'):
|
||||
RangeSource('https://example.org/public', 10, opener=opener).read(2, 2)
|
||||
self.assertEqual(opener.open.call_count, 1)
|
||||
|
||||
def test_sparse_footer_windows_are_synthetic_not_network_reads(self):
|
||||
source = SparseFile(100, [(0, b'PAR1'), (95, b'footer')])
|
||||
source.seek(90); self.assertEqual(source.read(10), b'\x00'*5+b'foote')
|
||||
source.seek(-4, 2); self.assertEqual(source.tell(), 96)
|
||||
with self.assertRaises(ValueError):
|
||||
source.seek(-101, 2)
|
||||
|
||||
|
||||
@unittest.skipUnless(importlib.util.find_spec('pyarrow'), 'Install mirrors extra on the approved remote runner')
|
||||
class ParquetProjectionTests(unittest.TestCase):
|
||||
def test_complete_selected_rows_and_no_excluded_page_requests(self):
|
||||
import pyarrow.parquet as pq
|
||||
raw, expected = fixture(); source = MemorySource(raw); rows, proof = projection(source, COLUMNS)
|
||||
self.assertEqual(rows, expected)
|
||||
metadata = pq.ParquetFile(io.BytesIO(raw)).metadata
|
||||
excluded = []
|
||||
for i in range(metadata.num_row_groups):
|
||||
group = metadata.row_group(i)
|
||||
for j in range(group.num_columns):
|
||||
col = group.column(j)
|
||||
if col.path_in_schema.split('.')[0] not in COLUMNS:
|
||||
start = min(x for x in [col.data_page_offset, col.dictionary_page_offset] if x is not None and x > 0)
|
||||
excluded.append((start, start+col.total_compressed_size))
|
||||
for request in source.receipts:
|
||||
a, b = request['start'], request['start']+request['length']
|
||||
self.assertFalse(any(max(a, c) < min(b, d) for c, d in excluded))
|
||||
self.assertFalse(proof['raw_parquet_saved']); self.assertFalse(proof['full_file_checksum_verified'])
|
||||
self.assertLess(source.bytes_read, len(raw))
|
||||
encoded = json.dumps(rows)+json.dumps(proof)
|
||||
self.assertNotIn('OCR_FULLTEXT_SENTINEL', encoded); self.assertNotIn('REVIEW_FULLTEXT_SENTINEL', encoded)
|
||||
self.assertNotIn('META_REVIEW_SENTINEL', encoded)
|
||||
|
||||
def test_wrong_schema_and_footer_fail_before_any_payload_download(self):
|
||||
raw, _ = fixture(); source = MemorySource(raw)
|
||||
with self.assertRaisesRegex(ValueError, 'Missing'):
|
||||
projection(source, {'missing_column': 'string'})
|
||||
self.assertEqual(len(source.receipts), 3)
|
||||
source = MemorySource(raw)
|
||||
with self.assertRaisesRegex(ValueError, 'type differs'):
|
||||
projection(source, {**COLUMNS, 'year': 'string'})
|
||||
self.assertEqual(len(source.receipts), 3)
|
||||
bad = bytearray(raw); bad[-8:-4] = struct.pack('<I', len(raw))
|
||||
source = MemorySource(bytes(bad))
|
||||
with self.assertRaisesRegex(ValueError, 'footer'):
|
||||
projection(source, COLUMNS)
|
||||
self.assertEqual(len(source.receipts), 2)
|
||||
|
||||
def test_collector_output_preserves_claims_without_original_identity(self):
|
||||
raw, _ = fixture(); item = {'path': 'data/emnlp-00000-of-00001.parquet', 'size': len(raw), 'lfs': {'oid': 'a'*64}}
|
||||
with tempfile.TemporaryDirectory() as temp:
|
||||
out = Path(temp)/'projected'
|
||||
with patch('paper_atlas.collectors.reviewarena.inventory', return_value=[item]):
|
||||
report = collect('emnlp', out, source_factory=lambda url, size: MemorySource(raw))
|
||||
self.assertEqual(report['mirror_decision_counts'], {'accepted': 2, 'rejected': 1, 'withdrawn': 1, 'unknown': 1})
|
||||
self.assertEqual(report['records'], 5); self.assertFalse(report['database_written'])
|
||||
projected = [json.loads(line) for line in next(out.glob('*.jsonl')).read_text().splitlines()]
|
||||
for row in projected:
|
||||
self.assertEqual(row['identifiers'], {})
|
||||
self.assertFalse(row['metadata']['original_forum_id_verified'])
|
||||
self.assertFalse(row['metadata']['original_decision_signature_verified'])
|
||||
self.assertNotIn('status', row)
|
||||
self.assertIn(COMMIT, row['external_id'])
|
||||
self.assertEqual(projected[2]['authors'], ['Anonymous'])
|
||||
self.assertEqual(projected[2]['metadata']['forum_id_claim'], 'unverified-source-2.xml')
|
||||
self.assertEqual(projected[0]['track'], 'main')
|
||||
self.assertEqual(projected[0]['mirror_decision_track_claim'], 'findings')
|
||||
all_saved=''.join(p.read_text() for p in out.iterdir())
|
||||
self.assertNotIn('FULLTEXT_SENTINEL', all_saved); self.assertNotIn('META_REVIEW_SENTINEL', all_saved)
|
||||
with self.assertRaisesRegex(ValueError, 'New output'):
|
||||
collect('emnlp', out)
|
||||
|
||||
def test_offline_replay_reclassifies_literals_preserving_all_raw_fields_without_network(self):
|
||||
raw, _ = fixture(); item = {'path': 'data/emnlp-00000-of-00001.parquet', 'size': len(raw), 'lfs': {'oid': 'a'*64}}
|
||||
with tempfile.TemporaryDirectory() as temp, patch('paper_atlas.collectors.reviewarena.inventory', return_value=[item]):
|
||||
out = Path(temp)/'v1'; collect('emnlp', out, source_factory=lambda url, size: MemorySource(raw))
|
||||
file, = out.glob('*.jsonl'); before = [json.loads(x) for x in file.read_text().splitlines()]
|
||||
# Mimic the first reviewed source version: new literal spellings were unknown.
|
||||
for row in before[:2]:
|
||||
row['mirror_decision_category'] = 'unknown'; row.pop('mirror_decision_track_claim')
|
||||
file.write_text(''.join(json.dumps(x)+'\n' for x in before))
|
||||
report_path = out/'collection-report.json'; report = json.loads(report_path.read_text())
|
||||
report['files'][0]['file_sha256'] = hashlib.sha256(file.read_bytes()).hexdigest()
|
||||
report_path.write_text(json.dumps(report)); report_hash = hashlib.sha256(report_path.read_bytes()).hexdigest()
|
||||
no_network = unittest.mock.Mock(side_effect=AssertionError('Offline replay must not contact source'))
|
||||
result = collect('emnlp', Path(temp)/'v2', from_projection=report_path, projection_sha256=report_hash, source_factory=no_network)
|
||||
no_network.assert_not_called(); self.assertEqual(result['metadata_bytes_read'], 0)
|
||||
self.assertEqual(result['source_projection_bytes'], report['metadata_bytes_read'])
|
||||
self.assertEqual(result['mirror_decision_counts']['accepted'], 2)
|
||||
self.assertEqual(result['mirror_decision_track_claim_counts'], {'findings': 1, 'main': 1, 'unspecified': 3})
|
||||
new = [json.loads(x) for x in next((Path(temp)/'v2').glob('*.jsonl')).read_text().splitlines()]
|
||||
self.assertEqual([r['raw_fields'] for r in new], [r['raw_fields'] for r in before])
|
||||
self.assertEqual([r['source_locator'] for r in new], [r['source_locator'] for r in before])
|
||||
with self.assertRaisesRegex(ValueError, 'hash differs'):
|
||||
collect('emnlp', Path(temp)/'bad', from_projection=report_path, projection_sha256='0'*64, source_factory=no_network)
|
||||
no_network.assert_not_called()
|
||||
|
||||
def test_failure_is_retained_without_source_payload_or_retries(self):
|
||||
item = {'path': 'data/emnlp-00000-of-00001.parquet', 'size': 50, 'lfs': {'oid': 'a'*64}}
|
||||
with tempfile.TemporaryDirectory() as temp:
|
||||
out = Path(temp)/'failed'
|
||||
with patch('paper_atlas.collectors.reviewarena.inventory', return_value=[item]), patch('paper_atlas.collectors.reviewarena.projection', side_effect=RuntimeError('private body example')):
|
||||
with self.assertRaises(RuntimeError):
|
||||
collect('emnlp', out, source_factory=lambda url, size: None)
|
||||
report = json.loads((out/'failure.json').read_text())
|
||||
self.assertEqual(report['status'], 'incomplete_do_not_import')
|
||||
self.assertTrue(report['no_automatic_retry']); self.assertNotIn('private body', (out/'failure.json').read_text())
|
||||
|
||||
|
||||
class MirrorSemanticsTests(unittest.TestCase):
|
||||
def test_pinned_inventory_covers_all_seven_families_and_fifteen_shards(self):
|
||||
splits = ['colm', 'corl', 'emnlp', 'iclr', 'icml', 'neurips', 'tmlr']
|
||||
result = {s: inventory(s) for s in splits}
|
||||
self.assertEqual(sum(len(v) for v in result.values()), 15)
|
||||
self.assertEqual([len(result[s]) for s in splits], [1, 1, 1, 5, 1, 5, 1])
|
||||
self.assertEqual(result['emnlp'][0]['size'], 90794691)
|
||||
self.assertEqual(result['emnlp'][0]['lfs']['oid'], 'e9e3aab42d28ca32a02287ed0bd0f1f9e1f3af5cb8ed8d6a9bf28d4f630773da')
|
||||
with self.assertRaises(ValueError):
|
||||
inventory('uai')
|
||||
|
||||
def test_missing_or_ambiguous_decision_is_never_inferred_from_venue(self):
|
||||
for literal in ['', None, 'Under review', 'Final venue pending', 'Accept maybe', 'Reject?']:
|
||||
self.assertEqual(decision_claim(literal), 'unknown')
|
||||
self.assertEqual(decision_claim('Desk Rejected'), 'desk_rejected')
|
||||
self.assertEqual(decision_claim('Withdrawn'), 'withdrawn')
|
||||
self.assertEqual(decision_claim('Accept (Findings)'), 'accepted')
|
||||
self.assertEqual(decision_claim('Accept-Main'), 'accepted')
|
||||
self.assertEqual(decision_claim('Accept-Findings'), 'accepted')
|
||||
self.assertEqual(decision_track_claim('Accept-Findings'), 'findings')
|
||||
self.assertIsNone(decision_track_claim('Accept (Poster)'))
|
||||
Reference in New Issue
Block a user