Review Hub → Pipeline B

Pipeline B: DESI Spectral Anomaly Miner

Human Review Dashboard — Gate 9


Saved

Top 50 Anomalies

Click any row to expand the detail panel with sky image, AI reasoning, and review form.

# Target ID RA Dec z Type ZWARN Score AI Label

Top 20 QSO Anomalies

QSO-specific anomaly subset. Click any row for details.

# Target ID RA Dec z ZWARN AI Label

Red-Team Checklist

Review each risk before finalizing labels. Check off items as you address them.

Likely Leakage Routes

Survey depth (FLUX_IVAR) was the dominant nuisance — partially mitigated by percentile normalization
Fiber-to-fiber variations not tested (no FIBER column in EDR)
Spectral features not used — only catalog metadata; true spectral anomalies may be missed

Astrophysical Confounders

Stars dominate top anomalies — many are just unusual stellar types, not cosmologically interesting
High-z QSO enrichment may reflect template mismatch in DESI pipeline, not intrinsic anomalousness
Galaxy anomalies may be photometric blends or deblending failures

Instrument Confounders

DESI fiber positioning errors could create systematic spectral artifacts
Sky subtraction residuals may dominate faint-object anomaly scores
Wavelength calibration issues could mimic emission-line anomalies

What Would Falsify the Pipeline

If top anomalies are all known artifacts in DESI validation catalogs
If anomaly scores correlate strongly with observing conditions after correction
If no anomaly survives visual inspection of the actual spectra