Southeastern Tibetan Plateau instrumental records are too short to characterise multidecadal dry-season variability. We reconstruct previous-October to current-May (pOct–cMay) precipitation for the Lugu Lake basin from two <em>Picea likiangensis</em> residual chronologies (204-year mainline 1817–2020, plus a descriptive extension to 1647); a nested principal-component transfer function explains 44.4 % of Ninglang variance (1960–2020), with positive RE and CE in both split-period verifications. Over 1817–2020 (29 |z| ≥ 1.5 extremes; 13 wet, 16 dry), extremes show no monotonic trend (logistic <em>P</em> = 0.63), the recent 30-yr window is not exceptional (<em>P</em> > 0.73), and only the most extreme tail tends upward (<em>P</em> = 0.042 for |z| ≥ 2). The dominant signal is decadal variability: the wettest (2000s, 197.7 mm) and driest (2010s, 128.2 mm) decades occur consecutively, and a 46-yr drought (1764–1809) in the single-tree segment exceeds red-noise expectations but lacks corroboration. Without local gauges, hydrological consistency rests on CNRD v1.0 natural runoff (<em>r</em> = 0.46) and soil-moisture and scPDSI fields (<em>r</em> = 0.33–0.41). Wet extremes coincide with La Niña-like cooling and stronger 700-hPa moisture convergence; dry extremes with the opposite configuration, an El Niño–dry teleconnection opposite in sign to Northeast Asia. Four CMIP6 models project wetter dry seasons with amplified wet tails.