The public 2017 WFI2033-4723 data show negative A1–A2 residual correlation in 46 epoch triples

We test simultaneous residual covariance in two public releases of lensed-quasar light curves. A fixed rule forms groups of three epochs without overlap. Each adjacent gap is at most 2.5 days. A contrast removes a locally linear magnitude signal. We report all 27 image pairs from seven lenses, with 1,316 released epochs and 381 selected triples. In the 157-row 2017 WFI2033-4723 release, the A1–A2 residual correlation is −0.593 from 46 triples. A circular block bootstrap gives a 95% interval of [−0.727, −0.465]. Both time halves and all nine grouping checks give negative values. The raw magnitude-difference variance is 1.616 times the sum of the two marginal variances. With a common magnitude zero point, a descriptive flux check gives a summed-flux variance ratio of 0.231 and nearly equal, opposite flux changes. This is consistent with known flux sharing, but does not prove the cause. One of 21 external pairs passes a Holm-adjusted 0.05 test: 2M1134 A–C has positive correlation of 0.487. The contrasts can also contain fast source variation and calibration errors. The released WFI2033 rows do not identify the 136-epoch subset used in the source paper. The result applies to the public file and does not establish a change in time delays or cosmology. We provide code, source hashes, derived data, figures, simulations, and reproduction instructions.

gravitational lensing · quasar light curves · photometric covariance · WFI2033-4723 · COSMOGRAIL · TDCOSMO · public data · reproducible research