A 20-day test finds no clear hard X-ray excess during soft flux decreases in LMC X-3 in 2018–2025

We test for a hard X-ray excess during soft flux decreases in LMC X-3. We compare Swift BAT 15–50 keV rates at the same central MAXI 2–4 keV flux. We use public daily data frozen on 2026-10-22. We use a fixed 20-day grid, coverage cuts, and a cubic adjustment for soft flux. We select 75 bins in 2018–2025: 40 falling and 35 rising. We subtract the rate during soft flux increases from the rate during decreases. The adjusted rate difference is −0.80 × 10⁻⁴ count cm⁻² s⁻¹. A circular block bootstrap uses 200-day blocks. Its nominal two-sided 95% interval is [−2.88, +1.94] × 10⁻⁴ count cm⁻² s⁻¹. All planned sensitivity intervals include zero. Two controls use specified models of correlated noise. Their interval coverage is 92.5% and 93.5% in 200 trials per model. These checks do not establish coverage for the real source. The result applies to the selected bins and a constant branch difference. The test can miss short changes. Errors in soft flux can affect the result. It does not exclude the spectral hysteresis measured in earlier RXTE work. We preserve the frozen data, source hashes, plan, code, figures, tables, and offline repeat-run checks.

LMC X-3 · X-ray binary · MAXI · Swift BAT · hysteresis · public data · block bootstrap · null result