74 × 2-min MWAX observations 1449928536–1449937296, 167–198 MHz; the RFI-heavy night added to the three in the gist. RFI decisions from the
bayesian-rfi mixture model of Ducharme, Li, Wilensky & Liu (2026), https://arxiv.org/abs/2609.26904 . No coarse-channel edge masking (Phase III).
Power spectra: unusable on this night, and no flagger fixes them. Persistent DTV broke the calibration (hyperdrive def4c315, FnxA-v2) of 27 of the 74 observations: their SSINS in coarse channels 137–146 is raised ×1.18 for the whole 2-min obs, hyperdrive fails to converge on up to 370 of 384 chanblocks, all in 174–195 MHz, and the whole-band bayesian-rfi rule flags them no more than the rest (the steady RFI is absorbed by the smooth background). Dropping those observations is not enough: the second figure has one power spectrum per observation. Everything below is independent of calibration.
(a) TV-band SSINS per obs ÷ kept-obs median; (b) calibration chanblocks that failed; (c) whole-band flagged %; (d) failed chanblocks per coarse channel. Orange × = dropped obs. Script: cal_rfi_1216.py, table cal_check_1216.csv.Shown: below the DTV band (167.06–173.4 MHz) every 2025-12-16 observation has the same single-obs power as 2025-10-07, including the 27 with broken calibration. In the DTV band (174.7–190.1 MHz), YY of the good-calibration observations is also normal, but XX is 10–10⁴× higher, median ~200×, and rises with the TV-band SSINS level (Spearman ρ = 0.72; ρ = 0.02 below the band). The sky is the same as on 2025-10-07 (same LST 3.4–5.9 h, same pointings; Moon and Sun below −36°). Interpretation, not proven: DTV that is present in XX for whole observations dominates this night's power spectrum; no flag set removes it (AOFlagger, whole-band and per-channel give the same level), and in the worst observations it also breaks calibration. Not ruled out: an XX-only instrumental problem in that band on this night. Scripts: perobs.sh, perobs_extract.py, perobs_plot.py.
RFI occupancy
Per-coarse-channel flaggers at 2 s × 40 kHz, XX (SSINS, AOFlagger mwa-1cc-sensitive excess over each channel's night median, CASA rflag, tfcrop). YYbayesian-rfi P(RFI) per coarse channel × 2 s.EAVILS |z| from hyperdrive bmetrics.Flag sets applied in the power spectra (8 s × 80 kHz).Occupancy time series of the flag sets.
What each flagger removes
Time on x (true UTC), 167–198 MHz on y. Panels: SSINS |z|; AOFlagger mwa-fullband-relaxed new flags; bayesian-rfi (Ducharme, Li, Wilensky & Liu (2026)) P(RFI) per coarse channel with the whole-band steps it drops (needs XX and YY); the gentle mask's TV-channel votes; and the fraction of the band each removes per 8 s step, all on one time axis. Flags do not depend on calibration. Script: wf_flaggers.py.