Measured, not claimed
How often is it right?
Each check is measured on cases whose answer was known in advance (last measured 2026-10-30 06:07 UTC). Where a number falls short, it is shown as it is: amber under 90 %, red at zero.
Known asteroids named, the right one150of 150Numbered asteroids drawn at random, their own MPC rows of October 2026 with every name taken off.
Real discoveries wrongly killed2of 200Discovery tracklets of objects the MPC designated from its NEO Confirmation Page (Aug-Sept 2026), each run with itself left out of the catalogue.
Surviving discoveries sent to the NEOCP desk119of 120Of the real discoveries that survived: routed to the NEO Confirmation Page by digest2 (a score of 65 or more, the MPC's own threshold).
Real asteroids not mistaken for a satellite, a star or noise349of 350Real tracklets, known and newly discovered, through the satellite, star and motion checks.
Stars caught pretending to move100of 100Gaia DR3 stars given as four rows over 30 minutes with 0.3" noise.
Chance links caught (killed)120of 120Rows of 2-4 different real objects in one field on one night (the MPC's own rows), linked as one, checked as the site runs (the MPC's WAMO on); a link named as one known object is not counted.
Links of two objects across nights caught60of 60Three-night links with the middle night swapped for a neighbour's own rows (the MPC's), both objects left out of the catalogue.
Real three-night links kept60of 60Three nights of one numbered asteroid within 12 days, left out of the catalogue.
Frames door (beta): a faint candidate found at its place (SNR 6)194of 196Star images injected at a signal-to-noise of 6 into the first frame of 49 real sequences (LCO 0.4 m and our T71).
Frames door (beta): our own T71 frames of numbered asteroids, named26of 2626 T71 sequences of numbered asteroids (25-30 Oct 2026; 3-24 frames over 1-41 minutes), the candidate given by its predicted RA and Dec, the first frame's WCS refit onto Gaia. With every frame's header WCS made 3-8 px off: 26 of 26. The four windows whose header WCS we found off (they shaped the refit) are named too.
Frames door (beta): hot pixels named96of 110Isolated hot pixels of T71's own dark (30 Oct 2026), in 24 raw frames each; the rest were killed as noise, none taken for a mover.
Deep: known objects far off their predicted places, named by the identification in orbit space0of 11Made rows of 11 short-arc asteroids on three nights, moved 0.3-1.0 of their catalogue's 3-sigma. On their catalogue's own places it names 5 of 11. It takes minutes: 7 of 11 stopped at its 300 s cap.
Satellites named, the right one98of 100Real satellites' places from the IAU CPS SatChecker (its own element archive and code), 4 rows over 6 minutes with 1" noise, 8 real stations, the last 10 days; SGP4 first, then Project Pluto's sat_id. No satellite named on 350 real asteroid tracklets.
Real NEOCP false alarms (objects the MPC called not a minor planet) not sent on as new60of 74Distinct objects whose NEOCP astrometry could be read (Project Pluto's pseudo-MPECs; 9 more pages held it redacted), removed June-Oct 2026: 14 of 74 (19 %) still pass as new. JPL Horizons' spacecraft name the Roman Space Telescope or its second stage for 19 of these, and 1 in the suspected-artificial set below. These cases helped shape the checks.
Satellites named at past dates (Jan-Apr 2026), the right one93of 100Real satellites' places from the IAU CPS SatChecker at past dates; Project Pluto's sat_id with its dated snapshots. Under 95 %: monthly snapshots miss low and transfer orbits.
Who can see a survivor next: a station that really followed within 24 h is among the first 5 listed214of 258Fresh 2026 NEO discoveries (June-July), each station ranked by its record of the 30 days before; 5 picked at random from the same list: 61 %.
Real NEOCP false alarms (objects the MPC took down as suspected artificial) not sent on as new13of 15Their own NEOCP astrometry (Project Pluto's pseudo-MPECs), 9 Sept-15 Oct 2026. 2 of 15 still pass as new: their rows fit ordinary orbits about the Sun, and nothing catalogued was there. These cases helped shape the checks.
Frames door (beta): noise pulled out of a stack, not passed60of 60The best peak of a 100-motion shift-and-stack search over empty sky in real LCO 0.4 m frames.
Frames door (beta): faint movers passed (stack z 15)56of 60Moving star images injected into real LCO 0.4 m frames at a stack signal-to-noise of 15.
Frames door (beta): one-frame events killed22of 22Cosmic rays LCO's pipeline flagged, given a motion through them (Skyptic never reads that flag).
Every number above comes straight from its measured run.