factcircuit
FactCircuit는 불변 출처와 단계별 검증, 제한된 검색, 재현 가능한 평가를 제공하는 감사 가능한 주장·증거 검증 루프입니다
아직 감지된 업데이트가 없습니다. GitHub 릴리스가 나오면 여기에 보입니다.
개인 계정의 공개 저장소 소유자는 GitHub로 로그인해 운영자를 확인하고 소개를 직접 관리할 수 있습니다. 관리자 승인은 필요하지 않습니다.
로그인한 뒤 확인 파일을 만들어 이 저장소의 기본 브랜치에 30분 안에 커밋해 주세요. 다른 브랜치나 포크의 파일은 인정하지 않습니다. 파일을 올린 뒤 아래에서 확인하면 관리자 승인 없이 관리할 수 있습니다. 자격을 유지하려면 파일을 남겨 두세요.
로그인이 필요하면 로그인 화면으로 이동합니다. 로그인 후 이곳으로 돌아와 원하는 확인 버튼을 다시 눌러 주세요.
프로젝트 폴더에서 /nomorevibe verify를 실행하면 소유권을 확인하고 소개를 직접 관리할 수 있습니다.
1분 만에 확인하기Machine-learned vetting for the Minor Planet Center's NEO Confirmation Page: for every candidate awaiting confirmation, predict the probability it's a real near-Earth object vs. a false positive (artifact, satellite, star, or main-belt asteroid), so the volunteer follow-up network stops spending scarce telescope time on junk.
저장소 README에서 가져옴
README 전체를 불러오는 중입니다. 아래는 저장된 발췌입니다.
** Now has 2 other capabilities. Go to https://spacesift.vercel.app to see **
Machine-learned vetting for the Minor Planet Center's NEO Confirmation Page:
for every candidate awaiting confirmation, predict the probability it's a real
near-Earth object vs. a false positive (artifact, satellite, star, or main-belt
asteroid), so the volunteer follow-up network stops spending scarce telescope
time on junk. Only ~50% of NEOCP candidates are real NEOs today; in the
Rubin/LSST era that drops to ~8%.README 전문 보기 ↗** Now has 2 other capabilities. Go to https://spacesift.vercel.app to see **
Machine-learned vetting for the Minor Planet Center's NEO Confirmation Page: for every candidate awaiting confirmation, predict the probability it's a real near-Earth object vs. a false positive (artifact, satellite, star, or main-belt asteroid), so the volunteer follow-up network stops spending scarce telescope time on junk. Only ~50% of NEOCP candidates are real NEOs today; in the Rubin/LSST era that drops to ~8%.
Benchmark to beat (A&A 2025): exclude >80% of non-NEOs while losing ≤1% of real NEOs. Current status and honest numbers: see PIPELINE.md.
src/capture_neocp.py, runs 2×/day): snapshots the live NEOCP
feed, saves each new object's astrometric tracklet while it is still
public (MPC deletes them on removal — the dataset cannot be backfilled),
and records outcomes from the Previous-NEOCP page. data/capture/ is the
accumulating labeled archive.src/features.py): the 26 digest2 sub-scores (13 orbit
classes × raw/noid) + tracklet geometry the published benchmark ignored
(sky-motion rate, ecliptic latitude, solar elongation, observatory code…).src/train.py --with-backfill): LightGBM, time-based
split, metrics reported at 99% NEO recall. Includes a guard against
provenance leakage between captured and backfilled rows.src/predict.py, src/make_dashboard.py): ranked
outputs/daily_vetting.csv + a live dashboard (outputs/dashboard.html)
with an extrapolated-position sky map.Plus an image track (src/fetch_css_bulk.py, src/motion_baseline.py,
src/train_cnn.py): real-vs-artifact classification on Catalina Sky Survey
4-frame detection sequences from Zooniverse's Daily Minor Planet project.
pip install numpy scikit-learn lightgbm pillow # + torch for the CNN
git clone https://github.com/Smithsonian/digest2 # see PIPELINE.md for build notes
Full runbook, environment gotchas, and current results: PIPELINE.md.
Original research plan: Planetary Defense v1 — Project Plan.md.
Data credits: Minor Planet Center (NEOCP, MPECs, observations API), Catalina Sky Survey / Zooniverse Daily Minor Planet. Please cite the MPC when using their data and poll politely.
** Now has 2 other capabilities. Go to https://spacesift.vercel.app to see **