Curves with one rise and one fall leave large residuals in three public KMT-2025-BLG-1760 streams with baseline error scales

We tested whether public KMTNet light curves of the known alert KMT-2025-BLG-1760 can have one rise and one fall. We used 227 I-band measurements from three sites over 860 <= HJD - 2460000 <= 915. Each site had a separate exact weighted unimodal fit. Errors were scaled by robust scatter outside the event interval. The minimum squared residual cost was Q = 8427.42. The three site costs were 1223.81, 1805.07, and 5398.54. Removing the three brightest measurements at each site left Q = 3043.95 for 218 points. One value per site per integer HJD day left Q = 5949.84 for 117 values. An added 0.05 mag error floor left Q = 953.05. Repeated bright phases occur near times 872-875, 882-883, and 901-903. Australia has no data in the middle phase. The bound applies to straight-motion single point mass lens models with constant light and calibration, positive source flux, and a circular source whose brightness does not increase with radius. It does not cover all lens scenarios. Two public reductions show similar structure but share the source images. The target and intervals were selected after inspection. Q is a descriptive cost, not a calibrated discovery probability. The result identifies an anomaly in the released data. It does not establish a planet or a physical cause. The package preserves source inputs, hashes, code, figures, row tables, and instructions. A clean offline run reproduced all nine final outputs with identical hashes.

microlensing · KMT-2025-BLG-1760 · public photometry · unimodal regression · light curve anomaly · reproducible research