No new planet candidate survives in TESS sector 107: USP, M-dwarf and single-dip searches of non-excluded stars end with zero vetted candidates

Summary

We searched TESS sector 107 2-min SPOC light curves for signals that earlier searches missed. The searches covered ultra-short periods (0.2–1 d), M dwarfs (0.3–15 d, with iterative masking) and single dips. Known TOIs, NASA-archive hosts and previously vetted stars were excluded. No signal survived vetting as a planet candidate. All 18 USP signals that passed a physical filter were rejected as false alarms, offset sources, eclipsing binaries or inconclusive. Of the 32 single-dip and M-dwarf stars vetted, 8 were already known TOIs, 15 were offset sources, 5 were eclipsing binaries, 3 were false alarms and 1 was inconclusive. The search finds only about 31% of known planets at a 6.7% false-alarm rate, so this is a statement about what the search can detect, not proof that no planets remain. All results are exploratory.

What was analyzed

• Target table (done): 13,000 sector 107 stars in and 13,000 rows out. TIC v8 parameters came from MAST. The exclusion set holds 855 vetted, 112 NASA, 248 TOI and 246 TCE-proxy TICs. The new search set has 11,876 stars. The TCE flag is only a proxy built from TOI dispositions.
• USP search (done): 12,145 stars were targeted and 12,010 gave rows; 135 (1.1%) failed. Biweight detrending was followed by fine-grid BLS on 0.2–1 d, with TLS refinement on 1,317 stars. The threshold was SDE 26.99, the 99th percentile over 11,742 controls. This gave 118 new USP signals, of which 18 passed the physical filter.
• M-dwarf multi-period search (done): 2,308 M dwarfs were targeted and 2,140 gave rows; 248 (10.7%) were lost, which exceeds the 5% limit, so the search is incomplete. 6,335 signals were found, 25 of them above threshold (SDE 37.35) on 22 new stars. 18 were confirmed by TLS and none had P < 1 d. Only 1 of 15 known planets was recovered. The whole non-excluded sample was not searched in this mode.
• Single-dip search (done): 12,998 stars were searched, with 2 failures. The search left 66 signals on 39 stars. About 24% of injected transits were recovered (682 injections).
• Calibration (done): Thresholds were set on 180 control stars and tested on 180 other controls. Recall was measured on 201 held-out known planets, out of 362 known planets in total.
• Cheap physical filter (done): 123 signals (118 new, 5 known USPs) were tested for density, ellipsoidal, reflection, odd/even and secondary signs.
• SAP+PCA diagnostic (done): 88 known hosts and 286 controls were used.
• Vetting (done): 21 stars (18 USP signals plus 3 known-planet controls), then 32 stars from the single-dip and M-dwarf lists.
• Compute: 744 runs (740 succeeded), 30.4 CPU hours, 10 work groups done, 0 dropped.

Key results

• Recall (exploratory): held-out recall is 0.31 (62/201) at SDE 19.3, with a false-alarm rate of 0.067 on control set B (target 0.05). Recall is about 42% for USPs and 32% for M dwarfs (in-sample group 1). SAP with PCA correction does not raise recall.
• USP vetting:
  • Verdicts for the 21 vetted stars: 5 false alarms, 1 inconclusive, 3 known planets (the controls), 8 offset sources and 4 eclipsing binaries.
  • The three controls were recovered as known planets: NGTS-6 b, LHS 1678 c and TOI-198 b.
  • Result: 0 new USP candidates.
• Single-dip and M-dwarf vetting (32 stars):

• Known TOIs re-detected (checks that the search works):

• The full vetting table is tables/vetting_results_final.csv (group 10; group 6 for the USP stars). Per-star reports are in views/vetting.
• Preregistered tests and claims: none. No result is confirmed.

Alternatives explored

These are directions beyond the original request.

• Single-transit search (group 7): gave 66 unvetted-by-period signals on 39 stars. Vetting found no planet candidate.
• SAP+PCA versus PDCSAP (group 9): no recall gain, so the M-dwarf and bright-star reruns were skipped.
• Declined:
  • Re-running the kit on the single-dip and M-dwarf lists was declined because it was already done.
  • A population-level excess test was declined because the held-out calibration already sets the false-alarm rate.

What did not work

• The signals that passed the filters were almost all explained by known TOIs, contamination from neighbours (offset sources) or eclipsing binaries.
• No USP below 1 d appeared among the new M-dwarf signals (n_new_P_lt1 = 0).
• The M-dwarf search recovered only 1 of 15 known planets, so it has low sensitivity.
• SAP with PCA common-mode correction did not improve detection.

Confidence and limitations

• Confidence is moderate (about 0.7) that no obvious candidate remains in the searched set.
• Recall is only about 31% and the false-alarm rate (6.7%) is above target. The injection tables are not a fully valid recall estimate.
• The M-dwarf search lost 10.7% of its stars and covered M dwarfs only.
• The TCE exclusion is a proxy, so some non-TOI TCEs may remain in the set.
• The centroid-based "offset source" verdicts rest on a 2-min light curve and catalogue matching. Some may hide true planets on blended neighbours, and these need pixel-level or ground-based follow-up.
• Nothing here needs lab validation, because no candidates are claimed.

Next steps

• Search the full non-excluded sample (not just M dwarfs) with iterative masking, and recover the 248 lost M dwarfs.
• Raise sensitivity to small planets, for example by combining sector 107 with its other sectors for the same stars, and re-test recall on injections.
• Follow up the inconclusive star (TIC 317185294) and the offset-source stars whose likely source is a bright neighbour, to check whether the signal can be attributed to a planet on the neighbour.