Multi-night photometric repeatability of STDWeb: an empirical error budget from a 14-night campaign on a single field

Abstract

Web-based photometric pipelines such as STDWeb are routinely used by small observatories and professional-amateur networks across the full range of photometric science: variable stars, supernovae, exoplanet transits, transient follow-up. The formal magnitude uncertainty they report describes a single reduction of a single frame; campaign science instead requires the scatter of the same star, same field, night after night. We measure this empirically on the field of the Einstein Probe trigger EP-WXT 01709274151, observed on 14 nights in 2026 July: 157 Gaia-like G exposures reduced homogeneously with STDWeb, analysed through 101 constant field stars (10.5<G<14.5). The median formal error is 8.1 mmag (6.7 mmag for the bright half), and within a single night the empirical differential scatter matches it ($χ_{\rm within}\simeq1$). Used naively - taking the pipeline's mag_calib, which lives in the instrumental pseudo-band - each star acquires a night-level offset of 7-9 mmag, the campaign scatter grows to 11-12 mmag, and the formal error is roughly half of the true one ($χ_{\rm camp}=1.5$-$1.7$). We show that this entire excess is the epoch-dependent colour term: analysing the catalogue-system magnitude $m_{\rm sys}$ = mag_calib + colour term x (BP-RP) collapses the night term to 3.4 mmag, restores $χ_{\rm camp}=1.0$-$1.1$, and improves the absolute Gaia-anchored per-star repeatability from 43 to 9.5 mmag, with the nightly zero-point peak-to-peak shrinking from 130 to 11 mmag and the airmass and colour correlations of the residuals vanishing. Stellar colours fitted from the light curves themselves reproduce Gaia BP-RP to 0.027 mag (r=0.98) and yield an identical budget: the correction needs no external colour catalogue. We provide the closed variance decomposition, practical recommendations, and a cautionary note on re-deriving zero points outside the pipeline.