How We Verify
Our Results
OrbVeil's TLE-based conjunction screening is validated against two independent data sources: Space-Track CDMs (official 18th SDS conjunction data messages) and SOCRATES (Satellite Orbital Conjunction Reports Assessing Threatening Encounters in Space).
Automated regression test suite against historical screening runs
OrbVeil detections that align with Space-Track CDM events
Historical CDMs used to train and validate the ML classifier
Precision-Recall AUC of escalation classifier on held-out test data
Validation Methodology
Space-Track CDM Cross-Reference
For each conjunction event we detect, we query the Space-Track CDM archive for the same object pair in the same time window. Match rate: 98%. Events in our catalog but not in Space-Track CDMs are usually below the threshold for official CDM issuance (<10 km).
SOCRATES Comparison
SOCRATES (NASA/Celestrak) independently computes high-interest conjunctions from TLE data. We compare our top-100 daily events against SOCRATES outputs. Agreement is typically 85-95% by rank order, with differences attributable to TLE age and propagation timing.
Historical Backtesting
Our ML escalation classifier was trained on 627,000 historical CDM events (2018–2025) with temporal train/test splits. We never use future data in training — the 0.849 PR-AUC is measured on events the model has never seen.
Known Limitations
TLE-based screening cannot match Special Perturbations (SP) orbit determination accuracy. For active satellites with recent maneuvers, TLE-based Pc estimates may diverge from SP-derived CDMs. We state this clearly in every report.