Of 155 stars in the registry, 81 look like eclipsing binaries and 51 like offset sources; none is a planet candidate

Summary

The goal was to vet only the TICs in candidates_ranked.csv, which contains 149 unique stars (259 rows). Three confirmed-planet controls were added. The current registry (each star's latest vetting with the standard kit) lists 155 stars: eclipsing_binary 81, offset_source 51, false_alarm 16, known_planet 6 and inconclusive 1. Nothing is left to review (toReview is empty). No star has a planet_candidate verdict. Everything here is exploratory.

The track's own earlier vetting of 152 stars gave different counts: eclipsing_binary 95, offset_source 37, needs_review 14, known_planet 3 (the controls) and inconclusive 3. The registry is newer, so it supersedes those counts. I cannot reconcile the two sets from the inputs. The registry has 155 stars, not 152, and it has 6 known_planet verdicts, not 3. I cannot tell which three extra stars are involved. The per-star tables below come from the earlier vetting and are not backed by the current registry. The 14 stars the earlier vetting marked needs_review no longer appear as pending in the registry, but I do not know their current verdicts.

The Gaia false-positive pass and the catalogue cross-match did not deliver results.

What was analyzed

• Input: tables/candidates_ranked.csv, 259 rows, collapsed to 149 unique TICs (110 duplicate detections). All 149 have period, t0 and depth. All 149 have SPOC 2-min data (median 6 sectors per star).
• Controls: 3 confirmed planets chosen in code from pscomppars: TIC 166778944, 470171739 and 353475866. An earlier set of controls chosen from memory was dropped.
• Vetting kit run (earlier vetting): 152 stars (149 + 3), 16 of 16 shards succeeded, none lost.
• Other-sector replication: 152 stars. Light curves were 139 SPOC 2-min, 12 TESS-SPOC FFI and 1 QLP. Sector 1 was excluded.
• Common-mode systematics test: 149 candidates.
• Compute: 148 runs, 99 succeeded, 19.5 CPU hours, 108 of 120 minutes, 4 work groups done and 1 dropped.

Key results

All results are exploratory. Full table of the earlier vetting: tables/vetting_results_final.csv (group 3, merge 5301b5ce).

• Current registry (155 stars): eclipsing_binary 81, offset_source 51, false_alarm 16, known_planet 6, inconclusive 1. No planet_candidate, nothing to review.
• Earlier vetting (152 stars): eclipsing_binary 95, offset_source 37, needs_review 14, known_planet 3, inconclusive 3.
• Controls: all 3 received known_planet in the earlier vetting and matched TOI-4153 b, TOI-1693 b and TOI-4662 b. They are not candidates.
• Needs-review stars (earlier vetting only):

Every one of the 14 failed at least one test, so none was a clean survivor. Periods near 13.85 d with only 2 transits are suspicious.

• Systematics (job bbac492b): 18.1% of candidate periods lie near the 13.7 d TESS orbit or its half or third, against a null mean of 4.7% (p = 0.0005). Pairs of stars with t0 within 0.05 d number 550 against a null mean of 80.8 (p = 0.002). The t0 null is crude. This points to instrumental systematics in part of the list, and it matches the 13.85 d cluster above. 30 candidates were flagged.
• Other-sector repeat: 128 of 149 candidates repeat, as do 3 of 3 controls. However, 4770 of 5960 null folds also exceed SNR 7, so this test barely discriminates. The per-star p ≤ 0.05 count is 47 of 149 (controls 3 of 3).
• Claim status: the earlier claim "95 of 152 are eclipsing binaries" was at level candidate, novelty known. The registry now gives 81 of 155, so that count is outdated. The reviewer returned inconclusive, with no verdict. No preregistered hypotheses were run and there is no leaderboard.

Alternatives explored

These directions went beyond the request.

• Other-sector replication: completed, but weakly informative because of the high null rate.
• Systematics test: partial, and it found orbit-harmonic clustering.
• Gaia DR3 false-positive pass (RUWE, neighbours, giant host, companion radius): unfinished, so no flag rates exist.

What did not work

• The other-sector repeat test does not separate real signals from random folds, since most null folds pass SNR > 7.
• The Gaia false-positive flags (giant host, RUWE, bright neighbours, companion radius, duration vs density) stayed untested, as did the TOI and NASA archive cross-match for the 14 stars.

Confidence and limitations

Confidence is moderate for the broad conclusion that most of the list is false positives, and low for the status of any individual star. The verdicts come from a single automated kit with no independent check. The registry and the earlier vetting disagree in total and in category counts (for example 16 false_alarm in the registry versus none earlier), and I could not explain the difference. Some stars from the earlier needs_review list may have changed class. Any star that could still be a planet needs photometric follow-up and, ultimately, spectroscopic follow-up (radial velocity, high-resolution imaging).

Next steps

• Export the per-star registry verdicts and compare them with the earlier table, especially for the 14 former needs_review stars and the 3 additional known_planet stars.
• Run the Gaia DR3 checks (RUWE, neighbours, stellar radius) on the 14 former needs_review stars.
• Cross-match those 14 against TOI and the NASA archive.
• Inspect pixel-level data and individual transits for the stars with long periods and few transits, to test for the 13.7 d orbit systematics.
• Check the short-period stars with many transits (TIC 197733707, 79247714, 80056984, 350431472, 179120455) for background binaries via difference imaging.
• Fit transit models to any survivor.