The blind BLS search of TESS sector 1 failed the known-planet validation, so no new candidates are reported for Tmag < 11

Summary

The pipeline was first tested on confirmed transiting planet hosts and planet-free control stars from sector 1. Blind recovery of the correct period was too low and the false-alarm rate too high to trust the method. The Tmag < 11 search therefore has no validated detection threshold, and no candidates are reported as new or ranked. All results below are exploratory.

Key results

All numbers come from one validation run (job 4c3a4315), which is exploratory.

• Test set: 69 confirmed hosts (NASA Exoplanet Archive, tran_flag=1) and 300 control stars absent from the archive and the TOI list. The controls were split into 200 for tuning and 100 held out. The run produced 1407 result rows with no errors.
• Method: quality-flag-0 PDCSAP flux, biweight detrending with windows of 0.5, 0.75 and 1.5 d, a 4σ upward clip, 10-minute binning, and astropy BLS with 7 durations from 0.03 to 0.35 d.
• Recall: it was computed on the 58 hosts with at least 2 expected transits.
  • With the false-alarm rate held to 0.05 or less, the best exact-period recall was 0.483. That setting was a 0.5 d window, the robust-z "sde" statistic, a threshold of 10, and a false-alarm rate of 0.04.
  • The best recall at any threshold was 0.724, so false alarms rose as recall rose.
  • Counting harmonic aliases as correct, recall was 0.5 at the best setting.
  • All three figures fall short of the 0.9 target.
• Consequence: no threshold or setting was frozen. The injection-recovery calibration, the full Tmag < 11 search, and the checks against eclipsing binaries and the NASA/TOI catalogs were not carried out. No candidate list, periods or depths can be given.

What did not work

• Biweight detrending with astropy BLS, across the three window lengths and two ranking statistics (depth-SNR and robust-z "sde"), did not recover at least 90% of known planets at a false-alarm rate of 0.05 or less.
• The sde statistic did not separate real signals from noise on planet-free stars well enough to set a useful threshold.
• The causes of the individual missed planets were not diagnosed. Candidate causes include detrending that suppresses transits, a period grid that is too coarse, and shallow or long-duration transits. None of these was tested.

Confidence and limitations

• The negative outcome is exploratory, with no preregistration and no independent reproduction. The verification record has no claims.
• Recall comes from only 58 hosts, so a 0.48 recall has a wide uncertainty (binomial interval of roughly ±0.13).
• The recall criterion was the top BLS peak within 2% of the catalog period. That is strict for planets with multiple signals, because the top peak may belong to another planet.
• The result is specific to this detrending and BLS configuration. It does not show that sector 1 holds no new candidates.
• Nothing here needs lab validation. Any future candidate would need follow-up photometry and spectroscopy.

Next steps

• Compare each of the 58 expected-transit hosts' injected or known signal against the detrended light curve. This should show whether the misses come from detrending, the period grid, or low SNR.
• Test detrending that masks known transits, a finer period grid, and TLS or a multi-planet iterative search (masking the top peak and searching again).
• Calibrate the detection threshold by injecting transits into real light curves across period, depth and duration, then freeze it on the tuning controls.
• Check the frozen threshold on the held-out controls and the confirmed hosts. Proceed only if recall is at least 0.9 at a false-alarm rate of 0.05 or less.
• Then search the Tmag < 11 stars and cross-match the survivors against the NASA Exoplanet Archive and the TOI list. Vet them with odd/even depth, secondary eclipse, centroid and contamination checks to rule out eclipsing binaries.