TIC 372784556 (G 79-12): an unconfirmed 12-hour transit-like signal in four TESS sectors, with a pre-registered test for Sector 109

We report a periodic transit-like signal on the late-K/early-M dwarf TIC 372784556 (G 79-12, Tmag 11.97, d about 74 pc) in four TESS sectors spanning 1848 days: P = 0.4981329 +/- 0.0000007 d (11.955 h; the pre-registered primary hypothesis, the best of eight aliases; regression estimates 0.498132-0.498134 d), depth about 650-700 ppm, duration about 0.7-1.0 h; if a planet on the target it would be about 1.5 Earth radii. The signal was found in an automated coherent multi-sector search of public light curves (about 1300 stars). It passed two of three search gates; it failed the third in the original run because of a single scattered-light event in the background test, and passes only after that test was changed to a median, a change decided after this result was seen, so the star is not an independently passed candidate. A centroid test in sky coordinates is consistent with the dimming source being on the target (offset -0.1 +/- 1.4, +0.1 +/- 1.2 arcsec; sources beyond about 7 arcsec are excluded; a companion closer than that is not); TRICERATOPS on the TESS data alone gives FPP about 0.09 and NFPP 0.00. This is not a validation: there is no imaging or spectroscopy, and all four sectors entered the discovery. We therefore recorded a test for TESS Sector 109 (5-31 October 2026) with its decision rule, ephemeris hypotheses, full-frame photometry recipe and script, frozen on 23 October 2026 before any data of that sector were retrieved or looked at. The images of its first half had been public since 21 October, so the time stamp can only precede the second half; the data product and the threshold of the rule were each changed once after the first freeze, and a coverage check and a reporting rule were added, all before any data (disclosed). With the full-frame product the test can confirm the ephemeris but can practically not refute it. The signal may be a planet, an eclipsing companion or an artefact. The analysis was AI-assisted (Claude); the author is responsible for the content.